feat: add the kettle CLI, replacing the plugin's Python scripts

The plugin resolved its issue store from `__file__`, which put it inside a
versioned plugin cache: issues written from one project were invisible from the
next, and `origin: local` files — the only copy of that work by definition —
were stranded a version bump at a time. The walk that answers "which directory
is the project" was written three times over, and in a linked worktree the three
disagreed. Both are runtime failures rather than logic ones, so the fix is a
compiled binary: one walk, imported rather than re-derived, and a layering rule
the build graph enforces instead of a grep.

Seven packages, knowledge flowing one way. `project` answers which directory is
the project and depends on nothing. `issue` is the domain — format, taxonomy,
validation, checkboxes, dependency graph, the store, eviction — offline, with no
tracker in it. `wire` holds the protocol shapes. `gitea` is the transport,
`mapping` the bridge, `config` the credentials, `cmd` the command tree. Four
tests hold the boundaries, each failing on a real mistake rather than a naming
convention.

The marker moves to `.kettle/` and the login pin moves out of the harness's
settings file into `.kettle/config.yaml`, which pins a login by NAME; the tokens
live in one file per machine, mode 0600, outside every working tree. That
retires the PreToolUse guard hook entirely — the binary holds its own
credentials, so a command running under a login nobody chose is not expressible
rather than caught.

`kettle init` migrates an older `tmp/issues` or `.tea/issues` store in, as a
move: a store left behind at an old path is one somebody edits by accident
months later.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

cli/AGENTS.md

@@ -0,0 +1,233 @@
+# AGENTS.md — the kettle CLI
+
+`kettle` is a globally installed binary. It owns everything that used to be a
+Python script under `plugins/tea/skills/*/scripts/`: what an issue is, where the
+store lives, who this machine is, and (once the transport lands) how issues move
+to and from Gitea.
+
+The plugin keeps what only a plugin can carry — the rules an operator states and
+a binary cannot enforce. Everything mechanical is here.
+
+## Why a binary
+
+Three failures in the Python version were failures of *runtime*, not of logic:
+
+- the store resolved from `__file__`, so it landed inside a versioned plugin
+  cache and issues written from one project were invisible from the next;
+- the walk that answers "which directory is the project" was written three
+  times — store, login pin, guard hook — and in a linked worktree the three
+  disagreed;
+- `sys.path.insert` was the import mechanism, so the layering rule was a
+  convention checked by grep.
+
+A compiled binary answers all three by construction. There is one walk
+(`internal/project`), it is imported rather than re-derived, and the layering
+rule is a build graph a test walks.
+
+## One dependency
+
+`gopkg.in/yaml.v3`, vendored, and that is the whole list. Everything else is the
+standard library: the transport is plain `net/http` against a documented REST
+API, and the CLI has no cobra — commands are values in a registry, which is what
+lets the plugin's SKILL.md files be generated from the same struct that holds
+the code.
+
+`vendor/` is committed, so a build needs no network.
+
+## Layers
+
+Knowledge flows one way. The arrow means "imports".
+
+```
+cmd/kettle              thin main; exit status only
+
+internal/cmd            the command tree: flags, receipts, exit codes
+   │      │      │
+   │      │      └────► internal/config   who this machine is, what this
+   │      │                               project points at; yaml lives here
+   │      │                               and only here
+   │      └───────────► internal/gitea    TRANSPORT: one door for every
+   │                          │           request, pagination, payload dumps,
+   │                          │           the number -> slug ledger
+   │                          ▼
+   ├────────────────────► internal/wire   PROTOCOL: the JSON shapes and the
+   │                          ▲           identifiers. Imports nothing.
+   │                          │
+   └──► internal/mapping ─────┘           BRIDGE: md <-> those shapes, pure,
+              │                           no I/O; label colours live here
+              ▼
+internal/issue    DOMAIN   what an issue is: format, taxonomy, validation,
+      │                    checkboxes, dependency graph, the store, eviction
+      │                    offline — no tracker, no network, no JSON
+      ▼
+internal/project  ROOT     which directory is the project, and every path
+                           resolved from it: store, payload, config
+                           depends on nothing
+```
+
+Read it bottom-up and each layer knows strictly less about trackers than the one
+above it. Four tests hold the line, and each fails on a real mistake rather than
+on a naming convention:
+
+- `internal/issue` may import `internal/project` and the standard library, and
+  nothing else. One test walks `go list -deps` and fails on any path with a dot
+  in its first element — which is also what keeps yaml out of the domain — and
+  another names `net/http`, `net`, `os/exec` and `encoding/json`, standard
+  library the first test would not catch.
+- `internal/wire` imports only the standard library, checked the same two ways.
+- `internal/gitea` must not import `internal/issue` **or** `internal/mapping`:
+  the transport knows numbers, logins, HTTP and JSON, and none of what they mean.
+- `internal/mapping` performs no I/O and imports neither the transport nor the
+  configuration.
+
+`wire` exists because Go needs the JSON shapes to be one type. The transport and
+the bridge were written in parallel and each invented its own `Issue`, `Label`,
+`Milestone` and `Comment`; every command on top would then have copied fields
+from one struct into the other by hand, which is two vocabularies for one thing —
+exactly what this layering exists to prevent. Python did not have the problem
+because it passed dicts.
+
+If a tracker concept — an issue number, a login, an HTTP call, a label colour —
+shows up in `internal/issue`, it is in the wrong place.
+
+## The walk
+
+`internal/project` answers one question and everything else reads the answer.
+
+Anchors, first hit wins: `$CLAUDE_PROJECT_DIR`, then the working directory. Each
+is searched up its parent chain for a `.kettle/` marker, and then — only if that
+found nothing — up the parent chain of the **main working tree of any linked
+worktree** met on the way, reached by reading `gitdir:` out of a `.git` *file*
+and following `commondir`.
+
+Nothing here resolves from the executable's own location. Where an installation
+keeps its files is a fact about the installation; whose issues a tree has is a
+fact about the tree, and a binary installed in one place and pointed at another
+must answer from the one it was pointed at.
+
+**The marker is created by `kettle init`, never inferred.** `.git` was tried and
+is in every clone. No marker anywhere is an answer, not a fallback: the command
+reports which directories it searched and stops.
+
+## Configuration, and where secrets are not
+
+Two files, and the split is the whole design.
+
+`<project>/.kettle/config.yaml` — written by `kettle init`, says which tracker
+repository the issues belong to and which login to reach it under. It pins a
+login by **name**. The name is worth nothing on its own, which is what makes it
+safe to keep in a file inside a working tree.
+
+`~/.config/kettle/logins.yaml` (`$KETTLE_CONFIG_HOME` or `$XDG_CONFIG_HOME`
+override it) — one file per machine, mode 0600, holding the tokens. Managed with
+`kettle auth`, which reads the token from standard input by default because an
+argument is in the shell history the moment it is typed. Nothing prints a token
+back: `kettle config` shows `(set)`.
+
+A token in a file inside a working tree ends up in a commit. Not always, not
+immediately, and not by anyone careless — but a project config is exactly the
+file somebody eventually decides to share, and a secret that has ever been
+pushed has to be rotated.
+
+`KETTLE_LOGIN`, `KETTLE_REPO`, `KETTLE_URL` and `KETTLE_TOKEN` each override the
+file they shadow, for CI and for anyone who would rather have no token on disk.
+
+Unknown keys in either file are an **error**, not a silent drop: an older binary
+reading a newer config would otherwise delete the setting it did not recognize
+the next time it wrote the file.
+
+`init` gitignores `.kettle/` wholesale. An `origin: local` issue is the only copy
+of that work, and what goes into a shared history is the operator's call, not
+this binary's — drop the line if the team decides otherwise.
+
+## The store
+
+`<project root>/.kettle/issues`, one flat markdown file per issue, named by its
+slug, one metadata field per line so plain grep works without a parser.
+
+It holds two kinds of file and only one of them is a store. An `origin: local`
+issue lives here and nowhere else — that file **is** the issue. Anything with a
+tracker origin is a working copy, deleted the moment a push confirms the tracker
+is up to date. Eviction makes the same trade one step earlier and asks the file
+instead of the tracker, which is why it lives in the domain.
+
+`.kettle/payload/` is a sibling, never a child: request bodies are debris of the
+transport, and a scratchpad inside a store makes `ls .kettle/issues` lie about
+what exists.
+
+`kettle init` migrates older layouts in, oldest first — `tmp/issues` and then
+`.tea/issues` — and each is a **move**. A store left behind at an old path is a
+store somebody will edit by accident months later. It refuses to pick a winner
+when both sides hold a file of the same name.
+
+## Tests
+
+```bash
+cd cli && go test ./...
+```
+
+`internal/cmd` builds the binary once in `TestMain` and runs it as a subprocess
+against a throwaway project in a temp directory — the same discipline the Python
+suite arrived at, for the same reason. **The binary is never run in the
+directory it was built in**, because that is exactly the arrangement that hid
+the `__file__` bug: a tool is installed in one place and used on projects in
+another, and a test that collapses the two proves nothing about resolution.
+
+Every fixture strips `CLAUDE_PROJECT_DIR`. It is the first anchor of the walk,
+so the harness's own value would point every fixture at this repository.
+Anything touching credentials sets `KETTLE_CONFIG_HOME` at a temp directory, so
+a test run can neither read nor overwrite the developer's own tokens.
+
+## The round trip
+
+`push` and `pull` are the two halves of one rule, and the rule is that **the
+store holds what has not left this machine.**
+
+A successful push deletes `<id>.md` and every sidecar under that slug, on create
+and on `--update` alike, and prints the number and URL the issue now lives at.
+The deletion happens **only after a confirmed tracker response and only after the
+number -> slug ledger has been written** — network down, non-2xx, or an answer
+that does not carry the right number, and the file stays where it is while the
+run stops. A never-pushed `origin: local` issue is never touched by any of it.
+
+The slug survives that round trip two ways over, and a test proves both: it goes
+up in the body as `<!-- kettle:id … -->`, and it is indexed by number in
+`.kettle/issues/.remote.json`. A rename in the web UI, a lost ledger, a fresh
+clone, another machine — the file comes back under the same name and every
+`depends:` pointing at it still resolves. The marker is written in that spelling
+and read in both it and the older `<!-- tea:id … -->`, because issues pushed
+before the rename are still in the tracker.
+
+Pull by number fetches an issue in **any** state: a number is an address, not a
+query, and `42`, `#42`, `owner/repo#42` and a URL all name one. Only filter mode
+leaves closed issues out. A pull returns the unit of work rather than one row of
+it — blockers come down with it unless `--no-deps` says otherwise — and it
+overwrites the body, because it is a fetch and not a merge. The one exception is
+checkbox state, which is the local half of the work and is merged rather than
+clobbered.
+
+## No guard hook
+
+The Python version needed a `PreToolUse` hook to block any `tea` command that
+would run under a login the model picked instead of the operator. That whole
+apparatus is gone. The binary holds its own credentials and reads the login out
+of the project's own configuration, so there is no argument to police and no way
+for the transport and the guard to disagree — the failure the hook existed to
+catch is not expressible any more.
+
+There is also no `--login` and no `--repo` on any sync command bar `labels`.
+Which login a project runs under, and which repository its issues belong to, are
+facts about the project, stated once by `kettle init`. A cross-repository address
+is still an address: `kettle pull owner/repo#42` re-points the client for that
+one call.
+
+## Status
+
+Done and tested: all seven packages, and the commands `init`, `auth`, `config`,
+`new`, `check`, `ac`, `tree`, `index`, `evict`, `pull`, `push`, `remote`,
+`comment`, `close`, `labels`, `sync-evict`. 89 tests.
+
+Not done: the plugin still ships the Python scripts and the guard hook, and
+still resolves `.tea/`. Rewiring `plugins/tea` onto this binary — and generating
+its SKILL.md files from the command registry, so the docs cannot drift from the
+CLI — is the remaining work.

cli/cmd/kettle/main.go

@@ -0,0 +1,14 @@
+// Command kettle keeps a project's issues as local markdown and moves them to
+// and from a tracker.
+//
+// An issue is a unit of work first and a tracker row second. Everything under
+// `kettle help` that is not in the sync group works with the network off.
+package main
+
+import (
+	"os"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/cmd"
+)
+
+func main() { os.Exit(cmd.Main(os.Args[1:])) }

cli/go.mod

@@ -0,0 +1,5 @@
+module git.noodles.cam/claude-skills/marketplace/cli
+
+go 1.24
+
+require gopkg.in/yaml.v3 v3.0.1

cli/go.sum

@@ -0,0 +1,4 @@
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

cli/internal/cmd/ac.go

@@ -0,0 +1,187 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"regexp"
+	"strconv"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+var numberRe = regexp.MustCompile(`^\d+$`)
+
+func init() {
+	register(&Command{
+		Name:  "ac",
+		Group: GroupIssue,
+		Args:  "<id>",
+		Short: "list and tick an issue's checkboxes",
+		Long: `A checkbox is the one part of a body that is *state* and not prose. Everything
+else is written once; boxes get ticked as the work goes, and the only other ways
+to tick one are a human with an editor or a model rewriting the whole body — the
+second worse than the first, because the rewrite re-flows the text and the
+issue's diff swells around a change of one character. This changes that one
+character and nothing else.
+
+Named after ` + "`## Acceptance criteria`" + `, where most boxes live, but every checkbox in
+the body is listed and tickable: a type/feature keeps its children under
+` + "`## Issues`" + `, and binding this to one heading would silently lose half of them.
+
+A substring picks an item only when it picks exactly one. Two matches is an
+error listing both — a coin flip would tick the wrong box and look like it
+worked.
+
+Delivering the changed body to a tracker is not part of this; that is
+` + "`kettle push --update`" + `.`,
+		Examples: []Example{
+			{"kettle ac wire-sqlc-appclick", "numbered list with state"},
+			{"kettle ac wire-sqlc-appclick --check 3", "tick by number"},
+			{"kettle ac wire-sqlc-appclick --check регресс", "tick by substring"},
+			{"kettle ac wire-sqlc-appclick --uncheck 3", "untick it again"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			check := fs.String("check", "", "tick one item: number or substring")
+			uncheck := fs.String("uncheck", "", "untick one item: number or substring")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				if len(args) != 1 {
+					return Fail("give exactly one issue id")
+				}
+				checking, unchecking := wasSet(fs, "check"), wasSet(fs, "uncheck")
+				if checking && unchecking {
+					return Fail("--check and --uncheck are mutually exclusive")
+				}
+				id := args[0]
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				path := issue.PathOf(root, id)
+				// Raw bytes in and raw bytes out: byte-for-byte means the line
+				// endings too. Reading a CRLF file with translation and writing
+				// it back would rewrite every line while claiming to have
+				// changed one character.
+				raw, err := os.ReadFile(path)
+				if err != nil {
+					return Fail("no issue %q in %s", id, root)
+				}
+				text := string(raw)
+
+				// The whole file, not just the body: line numbers then point at
+				// the file, and the metadata block is rewritten by nobody.
+				// Round-tripping through the parser would re-render metadata and
+				// re-strip the body, which is exactly the churn this avoids.
+				items := issue.Checkboxes(text)
+				needle, checked := *check, true
+				if unchecking {
+					needle, checked = *uncheck, false
+				}
+				selecting := checking || unchecking
+
+				if len(items) == 0 {
+					if selecting {
+						return Fail("%s has no checkboxes", id)
+					}
+					fmt.Printf("%s — no checkboxes\n", id)
+					return nil
+				}
+
+				if !selecting {
+					done, total := issue.CheckboxProgress(text)
+					fmt.Printf("%s — %d/%d   %s\n", id, done, total, path)
+					fmt.Println(strings.Join(listing(items), "\n"))
+					return nil
+				}
+
+				item, err := selectItem(items, needle)
+				if err != nil {
+					return err
+				}
+				updated, err := issue.SetCheckbox(text, item.Line, checked)
+				if err != nil {
+					return err
+				}
+				if updated == text {
+					fmt.Printf("unchanged  %2d %s %s\n", item.Index, box(item.Checked), item.Text)
+					return nil
+				}
+				if err := os.WriteFile(path, []byte(updated), 0o644); err != nil {
+					return err
+				}
+				if _, _, err := issue.BuildIndex(root); err != nil {
+					return err
+				}
+
+				verb := "checked"
+				if !checked {
+					verb = "unchecked"
+				}
+				done, total := issue.CheckboxProgress(updated)
+				fmt.Printf("%s   %2d %s %s\n", verb, item.Index, box(checked), item.Text)
+				fmt.Printf("%s — %d/%d   %s:%d\n", id, done, total, path, item.Line)
+				return nil
+			}
+		},
+	})
+}
+
+func box(checked bool) string {
+	if checked {
+		return "[x]"
+	}
+	return "[ ]"
+}
+
+// listing is the numbered list, grouped by the heading each item sits under.
+func listing(items []issue.Checkbox) []string {
+	var out []string
+	section := "\x00" // no heading can equal this, so the first item opens a group
+	for _, c := range items {
+		if c.Section != section {
+			section = c.Section
+			head := section
+			if head == "" {
+				head = "(above the first heading)"
+			}
+			out = append(out, "", head)
+		}
+		out = append(out, fmt.Sprintf("  %2d %s %s", c.Index, box(c.Checked), c.Text))
+	}
+	return out
+}
+
+// selectItem resolves a --check/--uncheck argument to exactly one item.
+func selectItem(items []issue.Checkbox, needle string) (issue.Checkbox, error) {
+	needle = strings.TrimSpace(needle)
+	if needle == "" {
+		return issue.Checkbox{}, Fail("empty selector — give an item number or a substring")
+	}
+	if numberRe.MatchString(needle) {
+		n, _ := strconv.Atoi(needle)
+		if n < 1 || n > len(items) {
+			return issue.Checkbox{}, Fail("no item %d — the issue has %d", n, len(items))
+		}
+		return items[n-1], nil
+	}
+	var hits []issue.Checkbox
+	for _, c := range items {
+		if strings.Contains(strings.ToLower(c.Text), strings.ToLower(needle)) {
+			hits = append(hits, c)
+		}
+	}
+	switch len(hits) {
+	case 0:
+		return issue.Checkbox{}, Fail("nothing matches %q", needle)
+	case 1:
+		return hits[0], nil
+	}
+	lines := []string{fmt.Sprintf("%q matches %d items — narrow it down, or use a number:", needle, len(hits))}
+	for _, c := range hits {
+		lines = append(lines, fmt.Sprintf("  %2d %s %s", c.Index, box(c.Checked), c.Text))
+	}
+	return issue.Checkbox{}, Fail("%s", strings.Join(lines, "\n"))
+}

cli/internal/cmd/auth.go

@@ -0,0 +1,140 @@
+package cmd
+
+import (
+	"bufio"
+	"flag"
+	"fmt"
+	"io"
+	"os"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+)
+
+func init() {
+	register(&Command{
+		Name:  "auth",
+		Group: GroupProject,
+		Args:  "list | add | remove <name>",
+		Short: "manage the tokens this machine holds",
+		Long: `Credentials live in one file per machine, outside every working tree, mode
+0600. A project pins a login by NAME; the name is worth nothing on its own,
+which is what makes it safe to keep in a file inside the repository.
+
+The token is read from standard input unless --token is given, because an
+argument is in the shell history the moment it is typed:
+
+    kettle auth add --name noodles --url https://git.example.com < token.txt
+    pass show gitea/token | kettle auth add --name noodles --url https://git.example.com
+
+` + "`list`" + ` never prints a token. There is no flag to make it.`,
+		Examples: []Example{
+			{"kettle auth list", "what this machine holds"},
+			{"pass show gitea | kettle auth add --name noodles --url https://git.example.com", "add one, token on stdin"},
+			{"kettle auth remove noodles", "forget it"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			name := fs.String("name", "", "login name (add)")
+			url := fs.String("url", "", "instance URL, e.g. https://git.example.com (add)")
+			user := fs.String("user", "", "account this token belongs to; documentation only (add)")
+			token := fs.String("token", "", "token, if you would rather not use stdin (add)")
+
+			return func(args []string) error {
+				verb := "list"
+				if len(args) > 0 {
+					verb = args[0]
+				}
+				logins, err := config.LoadLogins()
+				if err != nil {
+					return err
+				}
+
+				switch verb {
+				case "list":
+					if len(logins.Logins) == 0 {
+						fmt.Printf("no logins in %s\n", config.LoginsPath())
+						return nil
+					}
+					fmt.Printf("%s\n\n", config.LoginsPath())
+					for _, l := range logins.Logins {
+						who := l.User
+						if who == "" {
+							who = "—"
+						}
+						fmt.Printf("  %-16s %-40s %s\n", l.Name, l.URL, who)
+					}
+					return nil
+
+				case "add":
+					if *name == "" || *url == "" {
+						return Fail("--name and --url are both required")
+					}
+					secret := *token
+					if secret == "" {
+						if secret, err = readToken(os.Stdin); err != nil {
+							return err
+						}
+					}
+					if secret == "" {
+						return Fail("no token — pipe one in, or pass --token")
+					}
+					entry := config.Login{
+						Name:  *name,
+						URL:   strings.TrimRight(*url, "/"),
+						User:  *user,
+						Token: secret,
+					}
+					if existing := logins.Find(*name); existing != nil {
+						*existing = entry
+					} else {
+						logins.Logins = append(logins.Logins, entry)
+					}
+					if err := config.SaveLogins(logins); err != nil {
+						return err
+					}
+					fmt.Printf("%s -> %s   %s\n", entry.Name, entry.URL, config.LoginsPath())
+					return nil
+
+				case "remove":
+					if len(args) != 2 {
+						return Fail("give exactly one login name to remove")
+					}
+					target := args[1]
+					kept := logins.Logins[:0]
+					found := false
+					for _, l := range logins.Logins {
+						if l.Name == target {
+							found = true
+							continue
+						}
+						kept = append(kept, l)
+					}
+					if !found {
+						return Fail("no login %q in %s", target, config.LoginsPath())
+					}
+					logins.Logins = kept
+					if err := config.SaveLogins(logins); err != nil {
+						return err
+					}
+					fmt.Printf("removed %s\n", target)
+					return nil
+				}
+				return Fail("unknown subcommand %q — list, add, or remove", verb)
+			}
+		},
+	})
+}
+
+// readToken takes the first non-empty line of r, trimmed.
+//
+// The first line, not the whole stream: a token piped from a password manager
+// often arrives with a trailing newline and sometimes with notes underneath it.
+func readToken(r io.Reader) (string, error) {
+	sc := bufio.NewScanner(r)
+	for sc.Scan() {
+		if line := strings.TrimSpace(sc.Text()); line != "" {
+			return line, nil
+		}
+	}
+	return "", sc.Err()
+}

cli/internal/cmd/check.go

@@ -0,0 +1,110 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"sort"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+func init() {
+	register(&Command{
+		Name:  "check",
+		Group: GroupIssue,
+		Args:  "[<id>…]",
+		Short: "validate issues against the canonical format",
+		Long: `The same check the sync layer runs before it pushes anything, available on its
+own so a local-only issue can be held to the format without a tracker being
+involved.
+
+Errors mean malformed; warnings mean it deviates from its type's template or its
+graph looks suspect. An unticked checkbox is neither: work not done yet is the
+normal state of a perfectly well-formed issue.
+
+Exit status is 1 when anything has errors, which is what makes this usable in a
+hook or a CI step.`,
+		Examples: []Example{
+			{"kettle check", "every issue in the store"},
+			{"kettle check wire-sqlc-appclick", "one issue"},
+			{"kettle check --quiet", "exit status only"},
+			{"kettle check --strict", "treat warnings as errors"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			quiet := fs.Bool("quiet", false, "exit status only, print nothing")
+			strict := fs.Bool("strict", false, "treat warnings as errors")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				if err := issue.StoreError(root); err != nil {
+					return err
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+
+				ids := args
+				if len(ids) == 0 {
+					for id := range issues {
+						ids = append(ids, id)
+					}
+					sort.Strings(ids)
+				}
+				known := map[string]bool{}
+				for id := range issues {
+					known[id] = true
+				}
+				for _, id := range ids {
+					if !known[id] {
+						return Fail("no issue %q in %s", id, root)
+					}
+				}
+
+				bad := 0
+				for _, id := range ids {
+					errs, warns := issue.Validate(issues[id], known)
+					if *strict {
+						errs, warns = append(errs, warns...), nil
+					}
+					if len(errs) > 0 {
+						bad++
+					}
+					if *quiet {
+						continue
+					}
+					if len(errs) == 0 && len(warns) == 0 {
+						fmt.Printf("ok    %s\n", id)
+						continue
+					}
+					for _, e := range errs {
+						fmt.Printf("ERROR %s: %s\n", id, e)
+					}
+					for _, w := range warns {
+						fmt.Printf("warn  %s: %s\n", id, w)
+					}
+				}
+
+				for _, c := range issue.FindCycles(issue.Graph(issues)) {
+					bad++
+					if !*quiet {
+						fmt.Printf("ERROR cycle: %s\n", strings.Join(c, " -> "))
+					}
+				}
+
+				if !*quiet {
+					fmt.Printf("%d issue(s) checked, %d with errors\n", len(ids), bad)
+				}
+				if bad > 0 {
+					return SilentError{Code: 1}
+				}
+				return nil
+			}
+		},
+	})
+}

cli/internal/cmd/cli_test.go

@@ -0,0 +1,522 @@
+package cmd_test
+
+// The CLI is tested the way the Python suite it replaces was: the binary is
+// built once and run as a subprocess against a throwaway project somewhere
+// else entirely.
+//
+// That separation IS the contract. A tool is installed in one place and used on
+// projects in another, and the bug this discipline exists to catch — a store
+// resolved from the executable's own directory rather than from the tree it was
+// pointed at — is invisible to any test that runs the code in the directory it
+// lives in.
+//
+// Every fixture also strips CLAUDE_PROJECT_DIR unless the test is about it: it
+// is the first anchor of the walk, so the harness's own value would point every
+// fixture at this repository.
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+var kettle string
+
+func TestMain(m *testing.M) {
+	dir, err := os.MkdirTemp("", "kettle-bin")
+	if err != nil {
+		panic(err)
+	}
+	defer os.RemoveAll(dir)
+
+	kettle = filepath.Join(dir, "kettle")
+	build := exec.Command("go", "build", "-o", kettle, "../../cmd/kettle")
+	if out, err := build.CombinedOutput(); err != nil {
+		panic("building kettle: " + err.Error() + "\n" + string(out))
+	}
+	os.Exit(m.Run())
+}
+
+type result struct {
+	stdout, stderr string
+	code           int
+}
+
+func (r result) out() string { return r.stdout + r.stderr }
+
+// run invokes the binary in dir with a clean environment.
+func run(t *testing.T, dir string, args ...string) result {
+	t.Helper()
+	return runWith(t, dir, nil, "", args...)
+}
+
+// runWith is run plus extra environment and standard input.
+func runWith(t *testing.T, dir string, env []string, stdin string, args ...string) result {
+	t.Helper()
+	cmd := exec.Command(kettle, args...)
+	cmd.Dir = dir
+	cmd.Env = append(append(os.Environ(), "CLAUDE_PROJECT_DIR="), env...)
+	if stdin != "" {
+		cmd.Stdin = strings.NewReader(stdin)
+	}
+
+	var stdout, stderr strings.Builder
+	cmd.Stdout, cmd.Stderr = &stdout, &stderr
+	err := cmd.Run()
+
+	code := 0
+	var ee *exec.ExitError
+	if err != nil {
+		if !asExitError(err, &ee) {
+			t.Fatalf("running kettle %v: %v", args, err)
+		}
+		code = ee.ExitCode()
+	}
+	return result{stdout.String(), stderr.String(), code}
+}
+
+func mustRun(t *testing.T, dir string, args ...string) result {
+	t.Helper()
+	r := run(t, dir, args...)
+	if r.code != 0 {
+		t.Fatalf("kettle %v exited %d:\n%s", args, r.code, r.out())
+	}
+	return r
+}
+
+// newProject makes an initialized project in a temp directory and returns it.
+func newProject(t *testing.T) string {
+	t.Helper()
+	dir, err := filepath.EvalSymlinks(t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	mustRun(t, dir, "init")
+	return dir
+}
+
+func TestInitIsIdempotentAndGitignoresTheStore(t *testing.T) {
+	dir := newProject(t)
+
+	for _, d := range []string{".kettle/issues", ".kettle/payload"} {
+		if fi, err := os.Stat(filepath.Join(dir, d)); err != nil || !fi.IsDir() {
+			t.Errorf("%s was not created", d)
+		}
+	}
+	// An `origin: local` issue is the only copy of that work, and what goes in
+	// a shared history is the operator's call, not this command's.
+	ignore, err := os.ReadFile(filepath.Join(dir, ".gitignore"))
+	if err != nil || !strings.Contains(string(ignore), ".kettle/") {
+		t.Errorf(".kettle/ was not gitignored: %q", ignore)
+	}
+
+	again := mustRun(t, dir, "init")
+	if !strings.Contains(again.stdout, "already initialized") {
+		t.Errorf("a second init should be a no-op, got:\n%s", again.stdout)
+	}
+}
+
+func TestNoMarkerIsReportedNotGuessed(t *testing.T) {
+	dir := t.TempDir()
+	r := run(t, dir, "check")
+
+	if r.code == 0 {
+		t.Fatal("a directory that is not a project must not read as an empty store")
+	}
+	// The operator is owed the directories the search began from — that is how
+	// they see whether it began where they meant it to.
+	if !strings.Contains(r.stderr, "no .kettle/ found") || !strings.Contains(r.stderr, dir) {
+		t.Errorf("the failure must name what it searched:\n%s", r.stderr)
+	}
+}
+
+func TestTheGoldenPath(t *testing.T) {
+	dir := newProject(t)
+
+	mustRun(t, dir, "new", "--type", "task", "--title", "Wire sqlc into the appclick layer")
+	const id = "wire-sqlc-into-the-appclick-layer"
+
+	if r := mustRun(t, dir, "check"); !strings.Contains(r.stdout, "ok    "+id) {
+		t.Errorf("a fresh issue from its own template must validate:\n%s", r.out())
+	}
+
+	// Progress is counted off the body every time, never stored.
+	mustRun(t, dir, "ac", id, "--check", "1")
+	index, err := os.ReadFile(filepath.Join(dir, ".kettle", "issues", "INDEX.md"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(index), "| 1/2 |") {
+		t.Errorf("the index did not pick up the ticked box:\n%s", index)
+	}
+
+	if r := mustRun(t, dir, "tree"); !strings.Contains(r.stdout, id) {
+		t.Errorf("tree did not draw the issue:\n%s", r.stdout)
+	}
+}
+
+func TestTickingABoxChangesOneByte(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Tick one box")
+	path := filepath.Join(dir, ".kettle", "issues", "tick-one-box.md")
+
+	before, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	mustRun(t, dir, "ac", "tick-one-box", "--check", "1")
+	after, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if len(before) != len(after) {
+		t.Fatalf("length changed: %d -> %d", len(before), len(after))
+	}
+	diff := 0
+	for i := range before {
+		if before[i] != after[i] {
+			diff++
+		}
+	}
+	if diff != 1 {
+		t.Errorf("%d bytes changed, want 1 — a tick must not re-render the file", diff)
+	}
+
+	// And back again, byte for byte: the metadata block is rewritten by nobody.
+	mustRun(t, dir, "ac", "tick-one-box", "--uncheck", "1")
+	back, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(back) != string(before) {
+		t.Error("unticking did not restore the file byte for byte")
+	}
+}
+
+func TestFlagsWorkAfterPositionalArguments(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Order of arguments")
+
+	// `kettle ac <id> --check 1` is how everybody types it. A flag silently
+	// read as a positional would tick nothing and report success.
+	r := mustRun(t, dir, "ac", "order-of-arguments", "--check", "1")
+	if !strings.Contains(r.stdout, "checked") {
+		t.Errorf("the flag after the id was ignored:\n%s", r.out())
+	}
+}
+
+func TestTheStoreResolvesFromAnywhereInsideTheProject(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Seen from below")
+
+	deep := filepath.Join(dir, "internal", "adapters")
+	if err := os.MkdirAll(deep, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	r := mustRun(t, deep, "check")
+	if !strings.Contains(r.stdout, "seen-from-below") {
+		t.Errorf("a subdirectory saw a different store:\n%s", r.out())
+	}
+}
+
+func TestADifferentProjectAnswersWithItsOwnStore(t *testing.T) {
+	a, b := newProject(t), newProject(t)
+	mustRun(t, a, "new", "--type", "task", "--title", "Belongs to A")
+	mustRun(t, b, "new", "--type", "task", "--title", "Belongs to B")
+
+	r := mustRun(t, b, "check")
+	if strings.Contains(r.stdout, "belongs-to-a") {
+		t.Errorf("project B saw project A's issues:\n%s", r.out())
+	}
+}
+
+func TestALocalIssueIsNeverEvictedEvenWhenNamed(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Only copy there is")
+	path := filepath.Join(dir, ".kettle", "issues", "only-copy-there-is.md")
+	closeIssue(t, path)
+
+	r := mustRun(t, dir, "evict", "only-copy-there-is")
+	if _, err := os.Stat(path); err != nil {
+		t.Fatal("a closed origin: local issue was deleted — that file IS the work")
+	}
+	if !strings.Contains(r.stdout, "kept") {
+		t.Errorf("keeping it must be said out loud:\n%s", r.out())
+	}
+}
+
+func TestAClosedTrackedIssueIsEvictedWithItsSidecars(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Done and elsewhere")
+	store := filepath.Join(dir, ".kettle", "issues")
+	path := filepath.Join(store, "done-and-elsewhere.md")
+	closeIssue(t, path)
+	setField(t, path, "origin", "gitea")
+
+	sidecar := filepath.Join(store, "done-and-elsewhere.comments.md")
+	if err := os.WriteFile(sidecar, []byte("# thread\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	// A dry run touches nothing, and says so.
+	dry := mustRun(t, dir, "evict", "--dry-run")
+	if !strings.Contains(dry.stdout, "would evict") {
+		t.Errorf("dry run said nothing:\n%s", dry.out())
+	}
+	if _, err := os.Stat(path); err != nil {
+		t.Fatal("a dry run deleted the issue")
+	}
+
+	mustRun(t, dir, "evict")
+	if _, err := os.Stat(path); err == nil {
+		t.Error("the issue survived eviction")
+	}
+	// The domain does not need to know what a comment thread is to know a file
+	// named after this issue goes when it goes.
+	if _, err := os.Stat(sidecar); err == nil {
+		t.Error("the sidecar was left behind")
+	}
+}
+
+func TestCheckExitsNonZeroOnAMalformedIssue(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Loses its type")
+	path := filepath.Join(dir, ".kettle", "issues", "loses-its-type.md")
+	setField(t, path, "labels", "[]")
+
+	r := run(t, dir, "check")
+	if r.code != 1 {
+		t.Errorf("exit = %d, want 1 — this is what makes check usable in a hook", r.code)
+	}
+	if !strings.Contains(r.stdout, "need exactly one type/* label") {
+		t.Errorf("the finding was not reported:\n%s", r.out())
+	}
+}
+
+func TestNewRefusesToOverwriteAnExistingIssue(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Same title twice")
+	// Without an explicit id the slug is allocated around the collision…
+	mustRun(t, dir, "new", "--type", "task", "--title", "Same title twice")
+	if _, err := os.Stat(filepath.Join(dir, ".kettle", "issues", "same-title-twice-2.md")); err != nil {
+		t.Error("the second issue did not get its own slug")
+	}
+	// …but an id typed by hand is taken literally, and taken means taken.
+	r := run(t, dir, "new", "--type", "task", "--title", "Third", "--id", "same-title-twice")
+	if r.code == 0 || !strings.Contains(r.stderr, "already exists") {
+		t.Errorf("an explicit id must not overwrite:\n%s", r.out())
+	}
+}
+
+// An older layout is migrated in, and it is a MOVE: a store left behind at the
+// old path is a store somebody will edit by accident months later.
+func TestInitMigratesAnOlderStore(t *testing.T) {
+	dir, err := filepath.EvalSymlinks(t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	old := filepath.Join(dir, ".tea", "issues")
+	if err := os.MkdirAll(old, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	const body = "---\nid: from-the-old-store\nstate: open\nlabels: [type/task]\norigin: local\n---\n# From the old store\n\n## Summary\nx\n\n## Spec\nnone\n\n## Acceptance criteria\n- [ ] x\n"
+	if err := os.WriteFile(filepath.Join(old, "from-the-old-store.md"), []byte(body), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := mustRun(t, dir, "init")
+	if !strings.Contains(r.stdout, "moved 1 file(s)") {
+		t.Errorf("the migration said nothing:\n%s", r.stdout)
+	}
+	if _, err := os.Stat(filepath.Join(dir, ".kettle", "issues", "from-the-old-store.md")); err != nil {
+		t.Fatal("the issue did not arrive in the new store")
+	}
+	if _, err := os.Stat(old); err == nil {
+		t.Error("the old store is still there — two stores is what the marker exists to prevent")
+	}
+	if r := mustRun(t, dir, "check"); !strings.Contains(r.stdout, "from-the-old-store") {
+		t.Errorf("the migrated issue is not readable:\n%s", r.out())
+	}
+}
+
+// A migration never picks a winner. Two files of the same name are two versions
+// of one issue, and choosing quietly is how the wrong one survives.
+func TestInitRefusesToResolveAMigrationClash(t *testing.T) {
+	dir := newProject(t)
+	old := filepath.Join(dir, ".tea", "issues")
+	if err := os.MkdirAll(old, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	mustRun(t, dir, "new", "--type", "task", "--title", "Both sides have this")
+	if err := os.WriteFile(filepath.Join(old, "both-sides-have-this.md"), []byte("older\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := run(t, dir, "init")
+	if r.code == 0 {
+		t.Fatal("a clash must stop the run")
+	}
+	if !strings.Contains(r.stderr, "both hold") || !strings.Contains(r.stderr, "nothing was changed") {
+		t.Errorf("the clash was not explained:\n%s", r.stderr)
+	}
+	if _, err := os.Stat(filepath.Join(old, "both-sides-have-this.md")); err != nil {
+		t.Error("the older file was moved anyway")
+	}
+}
+
+func TestInitWritesTheConfigAndKeepsWhatItWasNotGiven(t *testing.T) {
+	dir, err := filepath.EvalSymlinks(t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	mustRun(t, dir, "init", "--login", "noodles", "--repo", "claude-skills/marketplace")
+
+	cfg := filepath.Join(dir, ".kettle", "config.yaml")
+	raw, err := os.ReadFile(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(raw), "login: noodles") ||
+		!strings.Contains(string(raw), "repo: claude-skills/marketplace") {
+		t.Fatalf("config did not record what it was given:\n%s", raw)
+	}
+
+	// Re-running init to change one setting must not drop the other.
+	mustRun(t, dir, "init", "--repo", "claude-skills/other")
+	raw, err = os.ReadFile(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(raw), "login: noodles") {
+		t.Errorf("the pinned login was dropped by an unrelated init:\n%s", raw)
+	}
+	if !strings.Contains(string(raw), "repo: claude-skills/other") {
+		t.Errorf("the repository was not updated:\n%s", raw)
+	}
+}
+
+func TestInitRefusesAMalformedRepo(t *testing.T) {
+	dir := t.TempDir()
+	r := run(t, dir, "init", "--repo", "marketplace")
+	if r.code == 0 || !strings.Contains(r.stderr, "owner/name") {
+		t.Errorf("a repo without an owner must be rejected before anything is written:\n%s", r.out())
+	}
+	if _, err := os.Stat(filepath.Join(dir, ".kettle")); err == nil {
+		t.Error("the marker was created despite the bad argument")
+	}
+}
+
+// The project pins a login by NAME. The credential lives in one file per
+// machine, outside every working tree — a token in a repository ends up in a
+// commit, and a secret that has been pushed has to be rotated.
+func TestTokensNeverLandInTheProject(t *testing.T) {
+	dir := newProject(t)
+	home := t.TempDir()
+	env := []string{"KETTLE_CONFIG_HOME=" + home}
+
+	runWith(t, dir, env, "s3cr3t-token\n", "auth", "add",
+		"--name", "noodles", "--url", "https://git.example.com/")
+	mustRun(t, dir, "init", "--login", "noodles", "--repo", "owner/name")
+
+	logins, err := os.ReadFile(filepath.Join(home, "logins.yaml"))
+	if err != nil {
+		t.Fatal("the token file was not written where it was told to go")
+	}
+	if !strings.Contains(string(logins), "s3cr3t-token") {
+		t.Errorf("the token was not stored:\n%s", logins)
+	}
+	if fi, err := os.Stat(filepath.Join(home, "logins.yaml")); err != nil || fi.Mode().Perm() != 0o600 {
+		t.Errorf("the token file must be 0600, got %v", fi.Mode().Perm())
+	}
+
+	cfg, err := os.ReadFile(filepath.Join(dir, ".kettle", "config.yaml"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if strings.Contains(string(cfg), "s3cr3t-token") {
+		t.Fatal("the token was written into the project — that file ends up in a commit")
+	}
+
+	// And nothing prints it back, either.
+	shown := runWith(t, dir, env, "", "config")
+	if strings.Contains(shown.out(), "s3cr3t-token") {
+		t.Errorf("`kettle config` printed the token:\n%s", shown.out())
+	}
+	if !strings.Contains(shown.stdout, "https://git.example.com") {
+		t.Errorf("the resolved URL was not shown:\n%s", shown.out())
+	}
+	if !strings.Contains(shown.stdout, "token     (set)") {
+		t.Errorf("whether a token was found must still be visible:\n%s", shown.stdout)
+	}
+}
+
+func TestAuthListNeverPrintsATokenAndRemoveForgetsIt(t *testing.T) {
+	dir := newProject(t)
+	home := t.TempDir()
+	env := []string{"KETTLE_CONFIG_HOME=" + home}
+
+	runWith(t, dir, env, "s3cr3t-token\n", "auth", "add", "--name", "noodles", "--url", "https://git.example.com")
+	listed := runWith(t, dir, env, "", "auth", "list")
+	if strings.Contains(listed.out(), "s3cr3t-token") {
+		t.Errorf("`auth list` printed a token:\n%s", listed.out())
+	}
+	if !strings.Contains(listed.stdout, "noodles") {
+		t.Errorf("`auth list` did not list the login:\n%s", listed.out())
+	}
+
+	runWith(t, dir, env, "", "auth", "remove", "noodles")
+	after := runWith(t, dir, env, "", "auth", "list")
+	if strings.Contains(after.stdout, "noodles") {
+		t.Errorf("the login survived removal:\n%s", after.stdout)
+	}
+}
+
+// A pinned login that is not on this machine is a fixable mistake, and the
+// message has to say which file was read and what it holds.
+func TestAMissingLoginIsExplained(t *testing.T) {
+	dir := newProject(t)
+	home := t.TempDir()
+	env := []string{"KETTLE_CONFIG_HOME=" + home}
+	mustRun(t, dir, "init", "--login", "absent", "--repo", "owner/name")
+
+	r := runWith(t, dir, env, "", "config")
+	if r.code == 0 {
+		t.Fatal("a login that does not exist must not resolve")
+	}
+	if !strings.Contains(r.stderr, `no login "absent"`) || !strings.Contains(r.stderr, "kettle auth add") {
+		t.Errorf("the failure must name the file and the fix:\n%s", r.stderr)
+	}
+}
+
+func closeIssue(t *testing.T, path string) {
+	t.Helper()
+	setField(t, path, "state", "closed")
+}
+
+func setField(t *testing.T, path, key, value string) {
+	t.Helper()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	lines := strings.Split(string(raw), "\n")
+	for i, line := range lines {
+		if strings.HasPrefix(line, key+": ") {
+			lines[i] = key + ": " + value
+		}
+	}
+	if err := os.WriteFile(path, []byte(strings.Join(lines, "\n")), 0o644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func asExitError(err error, target **exec.ExitError) bool {
+	ee, ok := err.(*exec.ExitError)
+	if ok {
+		*target = ee
+	}
+	return ok
+}

cli/internal/cmd/close.go

@@ -0,0 +1,327 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"sort"
+	"strings"
+	"time"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "close",
+		Group: GroupSync,
+		Args:  "<id|number> [<id|number>…]",
+		Short: "close or reopen issues in the tracker, and on disk with them",
+		Long: `STATE ONLY. This sends ` + "`{\"state\": …}`" + ` and nothing else: no title, no body, no
+labels, no milestone. Editing an issue is ` + "`kettle pull`" + ` -> edit ->
+` + "`kettle push --update`" + `; closing it is not an edit.
+
+EXPLICIT IDS ONLY. No --milestone, no --label, no "close everything that looks
+done". Which issues are finished is a judgement about content; this carries that
+judgement out, one named id at a time. Nothing here deletes an issue either —
+the tracker can, and it is not an operation of this workflow.
+
+WHAT MAY BE NAMED: a local slug, or a tracker key (42, #42, owner/repo#42, an
+issue URL). Both, and for the same reason: a push deletes the local file, so
+most issues in the tracker have no slug on disk to name them by. A slug is
+resolved through the file's ` + "`gitea:`" + ` handle when the file is there, and through
+the ledger (` + "`.remote.json`" + `) when push has already dropped it. A bare number is
+this project's repository; a qualified key names its own, so a foreign #42 can
+never be closed against the repository that happens to be configured here.
+
+An ` + "`origin: local`" + ` issue cannot be closed. It is not in the tracker, so there is
+no state there to change, and the run stops naming the id rather than quietly
+editing one field of a local file. Push it first, or delete it.
+
+THE LOCAL FILE IS WRITTEN ONLY AFTER THE TRACKER CONFIRMS: the answer has to be
+the very issue that was patched, in the state that was asked for. Anything else
+and the file is left exactly as it was. An issue whose local copy is gone
+(pushed and dropped) is closed in the tracker and nothing is written; the state
+comes down with the next pull.
+
+A tracker that refuses to close an issue its own dependency graph still blocks
+says so in the answer, and the run stops with its words: close the blockers
+first, or unlink them.`,
+		Examples: []Example{
+			{"kettle close wire-sqlc-appclick", "one issue, by slug"},
+			{"kettle close wire-sqlc-appclick 42 #43", "several, by slug or number"},
+			{"kettle close --reopen 42", "the same thing backwards"},
+			{"kettle close --dry-run 42 43", "what would change; no request at all"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			reopen := fs.Bool("reopen", false, "set the state back to open instead of closed")
+			dryRun := fs.Bool("dry-run", false, "print what would change; makes no request")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				if len(args) == 0 {
+					return Fail("name at least one issue: a slug, or 42, #42, owner/repo#42, a URL")
+				}
+				state, verb, past := "closed", "close", "closed"
+				if *reopen {
+					state, verb, past = "open", "reopen", "reopened"
+				}
+
+				root, client, err := syncStartExisting(*out)
+				if err != nil {
+					return err
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+				ledger := closeLedger(root)
+
+				// Every argument is resolved before anything is sent, so a typo in
+				// the third id does not leave the first two closed.
+				var targets []closeTarget
+				for _, arg := range args {
+					t, err := closeResolve(arg, root, issues, ledger)
+					if err != nil {
+						return err
+					}
+					if !closeHas(targets, t) {
+						targets = append(targets, t)
+					}
+				}
+
+				if *dryRun {
+					for _, t := range targets {
+						where := "no local copy"
+						if i, ok := issues[t.id]; ok {
+							where = fmt.Sprintf("%s (state: %s)", issue.PathOf(root, t.id), i.State)
+						}
+						fmt.Printf("would %-6s  %-24s %-20s %s\n",
+							verb, closeName(t.id), t.key.In(client.Repo()), where)
+					}
+					fmt.Printf("%d issue(s) would be %s; no request was made\n", len(targets), past)
+					return nil
+				}
+
+				touched := 0
+				for _, t := range targets {
+					c := client
+					if !t.key.Repo.Zero() {
+						c = client.For(t.key.Repo)
+					}
+					req := wire.IssueRequest{State: wire.Set(state)}
+					got, err := c.EditIssue(t.key.Number, req, fmt.Sprintf("state-%d", t.key.Number))
+					if err != nil {
+						return err
+					}
+					// The gate. Above it nothing local has been written; below it the
+					// file is about to say something the tracker had better agree
+					// with. An answer counts only when it is the very issue that was
+					// patched, in the state that was asked for.
+					if got.Number != t.key.Number || got.State != state {
+						return Fail("%s: the %s did not go through — the tracker answered for issue #%d in state %q. Nothing local was changed.",
+							t.key.In(c.Repo()), verb, got.Number, got.State)
+					}
+					fmt.Printf("%-8s  %-24s %-20s %s\n",
+						past, closeName(t.id), t.key.In(c.Repo()), got.HTMLURL)
+
+					i, ok := issues[t.id]
+					if !ok {
+						fmt.Printf("    no local copy — `kettle pull %d` to get one\n", t.key.Number)
+						continue
+					}
+					path, err := closeApply(root, i, state, got)
+					if err != nil {
+						return err
+					}
+					fmt.Printf("    state: %s  %s\n", state, path)
+					touched++
+				}
+
+				// Only when a file actually changed: INDEX.md is a view of the
+				// directory, and rewriting it after a run that wrote nothing local
+				// is a write nobody asked for.
+				if touched > 0 {
+					path, n, err := issue.BuildIndex(root)
+					if err != nil {
+						return err
+					}
+					fmt.Printf("index: %s — %d issue(s)\n", path, n)
+				}
+				return nil
+			}
+		},
+	})
+}
+
+// closeTarget is one issue a run will act on: where it is in the tracker, and
+// what this machine calls it, when this machine has a name for it at all.
+type closeTarget struct {
+	// id is the local slug, "" when nothing here names this issue. Closing one
+	// of those is ordinary — push deletes the file it would have been named by.
+	id string
+	// key is the tracker address. Its Repo is zero only when the argument was a
+	// bare number and no local copy or ledger entry qualified it, which means
+	// this project's own repository.
+	key wire.Key
+}
+
+// closeEntry is one row of the number -> slug ledger, parsed.
+type closeEntry struct {
+	key  wire.Key
+	slug string
+}
+
+// closeLedger is `.remote.json` as pairs, sorted so two identical runs report
+// an ambiguity in the same order.
+//
+// Read rather than ignored because it is the only thing on this machine that
+// still names an issue push has dropped: the file is gone, the slug is not.
+func closeLedger(root string) []closeEntry {
+	var out []closeEntry
+	for raw, slug := range gitea.LoadRemoteMap(root) {
+		k, err := wire.ParseKey(raw)
+		if err != nil || k.Repo.Zero() || k.Number < 1 {
+			continue
+		}
+		out = append(out, closeEntry{key: k, slug: slug})
+	}
+	sort.Slice(out, func(i, j int) bool { return out[i].key.String() < out[j].key.String() })
+	return out
+}
+
+// closeResolve turns one argument into a target.
+//
+// The order is the order of what is most authoritative about this machine: a
+// file on disk, then the ledger, then nothing. A key is already the tracker's
+// answer, so the only thing still wanted for it is the slug — so that the local
+// copy, if there is one, can be kept honest — and the file that carries the
+// handle knows that before the ledger does.
+//
+// The repository travels with the number, because a key may name one and a
+// `gitea:` handle always does. Sending a foreign key to whatever repository
+// this project points at would close somebody else's issue of the same number.
+func closeResolve(arg, root string, issues map[string]*issue.Issue, ledger []closeEntry) (closeTarget, error) {
+	// A key first, and a slug never looks like one: slugs hold no `#`, no `/`
+	// and no `:`, so the two vocabularies cannot collide.
+	if k, err := wire.ParseKey(arg); err == nil {
+		var hits []closeEntry
+		for id, i := range issues {
+			if h, ok := mapping.RemoteKeyOf(i); ok && h.Number == k.Number && (k.Repo.Zero() || h.Repo == k.Repo) {
+				hits = append(hits, closeEntry{key: h, slug: id})
+			}
+		}
+		sort.Slice(hits, func(a, b int) bool { return hits[a].slug < hits[b].slug })
+		if len(hits) == 0 {
+			for _, e := range ledger {
+				if e.key.Number == k.Number && (k.Repo.Zero() || e.key.Repo == k.Repo) {
+					hits = append(hits, e)
+				}
+			}
+		}
+		hit, err := closeOne(hits, arg, "a slug")
+		if err != nil {
+			return closeTarget{}, err
+		}
+		t := closeTarget{key: k}
+		if hit != nil {
+			t.id = hit.slug
+			t.key = k.In(hit.key.Repo)
+		}
+		return t, nil
+	}
+
+	if i, ok := issues[arg]; ok {
+		k, ok := mapping.RemoteKeyOf(i)
+		if !ok {
+			return closeTarget{}, Fail("%s is not in the tracker (origin: %s, no usable `%s:` handle) — "+
+				"there is no state there to change; `kettle push %s` first",
+				arg, i.Origin, mapping.GiteaKey, arg)
+		}
+		return closeTarget{id: arg, key: k}, nil
+	}
+
+	var hits []closeEntry
+	for _, e := range ledger {
+		if e.slug == arg {
+			hits = append(hits, e)
+		}
+	}
+	hit, err := closeOne(hits, arg, "a number")
+	if err != nil {
+		return closeTarget{}, err
+	}
+	if hit != nil {
+		return closeTarget{id: arg, key: hit.key}, nil // pushed, and its file went with the push
+	}
+	return closeTarget{}, Fail("no issue %q in %s or in its %s — name a tracker key "+
+		"(42, #42, owner/repo#42, or the issue's URL) to close one this machine has never seen",
+		arg, root, gitea.RemoteMapName)
+}
+
+// closeOne is the single ledger row for an argument, nil when the ledger knows
+// nothing about it, or an error when it knows two.
+//
+// Two answers mean one number (or one slug) under more than one repository, and
+// only a qualified key can settle that. Guessing would close the wrong issue.
+func closeOne(hits []closeEntry, arg, what string) (*closeEntry, error) {
+	seen := map[string]bool{}
+	var uniq []closeEntry
+	for _, h := range hits {
+		if k := h.key.String() + " " + h.slug; !seen[k] {
+			seen[k] = true
+			uniq = append(uniq, h)
+		}
+	}
+	switch len(uniq) {
+	case 0:
+		return nil, nil
+	case 1:
+		return &uniq[0], nil
+	}
+	var where []string
+	for _, h := range uniq {
+		where = append(where, h.key.String())
+	}
+	return nil, Fail("%q matches %s under more than one repository (%s) — say which, as owner/repo#N",
+		arg, what, strings.Join(where, ", "))
+}
+
+// closeApply writes the confirmed state onto the local file and returns its
+// path.
+//
+// `state:` is the domain's own field, so it is set on the issue and written out
+// by the domain's own writer. The sync-owned freshness fields travel with it:
+// the answer that authorized this write is also the newest thing the tracker has
+// said about the issue, so `synced:` and `remote-updated:` are stamped from it
+// rather than left describing an older read.
+func closeApply(root string, i *issue.Issue, state string, got *wire.Issue) (string, error) {
+	i.State = state
+	if i.Extra == nil {
+		i.Extra = map[string]string{}
+	}
+	i.Extra[mapping.SyncedKey] = time.Now().UTC().Format(time.RFC3339)
+	if got.UpdatedAt != "" {
+		i.Extra[mapping.RemoteUpdatedKey] = got.UpdatedAt
+	}
+	return issue.Save(root, i)
+}
+
+func closeHas(targets []closeTarget, t closeTarget) bool {
+	for _, have := range targets {
+		if have == t {
+			return true
+		}
+	}
+	return false
+}
+
+// closeName is what a receipt calls an issue this machine has no name for.
+func closeName(id string) string {
+	if id == "" {
+		return "(no local copy)"
+	}
+	return id
+}

cli/internal/cmd/command.go

@@ -0,0 +1,269 @@
+// Package cmd is the kettle command tree.
+//
+// Commands are values, not init() side effects on a framework: each one carries
+// the metadata a human needs (what it does, what it takes, worked examples) in
+// the same struct that carries the code. That is deliberate — the plugin's
+// SKILL.md files are generated from this list, so a command whose flags changed
+// cannot ship with documentation that says otherwise.
+//
+// The tree is flat. `kettle new`, not `kettle issue new`: an agent pays for
+// every token of every invocation, and the grouping that matters for reading is
+// carried in Group and only shows up in the docs.
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"sort"
+	"strings"
+)
+
+// Groups, in the order they are presented. They name the layer a command
+// belongs to, which is the one thing a reader has to keep straight: the domain
+// works offline and the tracker does not exist to it.
+const (
+	GroupProject = "project"
+	GroupIssue   = "issue"
+	GroupSync    = "sync"
+)
+
+var groupOrder = []string{GroupProject, GroupIssue, GroupSync}
+
+var groupBlurb = map[string]string{
+	GroupProject: "the project itself",
+	GroupIssue:   "issues as units of work — offline, no tracker involved",
+	GroupSync:    "moving issues between the store and the tracker",
+}
+
+// Example is one worked invocation. Both halves are shown in help and in the
+// generated skill docs.
+type Example struct {
+	Cmd  string
+	What string
+}
+
+// Command is one verb.
+type Command struct {
+	// Name is what the user types.
+	Name string
+	// Group is the layer it belongs to; documentation only.
+	Group string
+	// Args is the positional-argument spec, e.g. "<id> [<id>…]".
+	Args string
+	// Short is one line, shown in the command list.
+	Short string
+	// Long is the full explanation, shown by `kettle help <name>`.
+	Long string
+	// Examples are worked invocations.
+	Examples []Example
+	// Setup registers this command's flags on fs and returns the function that
+	// runs it, closing over them. Splitting it this way lets the doc generator
+	// walk the flags without running anything.
+	Setup func(fs *flag.FlagSet) func(args []string) error
+}
+
+var registry []*Command
+
+func register(c *Command) { registry = append(registry, c) }
+
+// Commands lists every command, sorted by group and then by name.
+func Commands() []*Command {
+	out := append([]*Command{}, registry...)
+	sort.SliceStable(out, func(i, j int) bool {
+		gi, gj := groupIndex(out[i].Group), groupIndex(out[j].Group)
+		if gi != gj {
+			return gi < gj
+		}
+		return out[i].Name < out[j].Name
+	})
+	return out
+}
+
+// Lookup finds a command by name.
+func Lookup(name string) *Command {
+	for _, c := range registry {
+		if c.Name == name {
+			return c
+		}
+	}
+	return nil
+}
+
+// Flags returns this command's flags without running it — what the doc
+// generator walks.
+func (c *Command) Flags() []*flag.Flag {
+	fs := flag.NewFlagSet(c.Name, flag.ContinueOnError)
+	fs.SetOutput(discard{})
+	c.Setup(fs)
+	var out []*flag.Flag
+	fs.VisitAll(func(f *flag.Flag) { out = append(out, f) })
+	return out
+}
+
+// Usage is the one-line synopsis.
+func (c *Command) Usage() string {
+	s := "kettle " + c.Name
+	if c.Args != "" {
+		s += " " + c.Args
+	}
+	return s
+}
+
+// SilentError carries an exit status for a command that has already said
+// everything it has to say. `check` uses it: findings went to stdout and a
+// second copy on stderr would be noise.
+type SilentError struct{ Code int }
+
+func (e SilentError) Error() string { return "" }
+
+// Fail is the error every command returns for an ordinary failure. Main
+// prefixes it with the command name.
+func Fail(format string, a ...any) error { return fmt.Errorf(format, a...) }
+
+// Main runs argv (without the program name) and returns the exit status.
+func Main(argv []string) int {
+	if len(argv) == 0 {
+		printUsage(os.Stdout)
+		return 0
+	}
+
+	name := argv[0]
+	switch name {
+	case "help", "-h", "--help":
+		if len(argv) > 1 {
+			c := Lookup(argv[1])
+			if c == nil {
+				fmt.Fprintf(os.Stderr, "kettle: no command %q\n", argv[1])
+				return 2
+			}
+			printCommand(os.Stdout, c)
+			return 0
+		}
+		printUsage(os.Stdout)
+		return 0
+	}
+
+	c := Lookup(name)
+	if c == nil {
+		fmt.Fprintf(os.Stderr, "kettle: no command %q — try `kettle help`\n", name)
+		return 2
+	}
+
+	fs := flag.NewFlagSet(name, flag.ContinueOnError)
+	fs.Usage = func() { printCommand(os.Stderr, c) }
+	run := c.Setup(fs)
+	if err := fs.Parse(permute(fs, argv[1:])); err != nil {
+		if err == flag.ErrHelp {
+			return 0
+		}
+		return 2
+	}
+
+	switch err := run(fs.Args()).(type) {
+	case nil:
+		return 0
+	case SilentError:
+		return err.Code
+	default:
+		fmt.Fprintf(os.Stderr, "kettle %s: %v\n", name, err)
+		return 1
+	}
+}
+
+func printUsage(w *os.File) {
+	fmt.Fprint(w, "kettle — issues as local markdown, and the tracker they sync with\n\n")
+	fmt.Fprint(w, "usage: kettle <command> [flags] [args]\n")
+
+	current := ""
+	for _, c := range Commands() {
+		if c.Group != current {
+			current = c.Group
+			fmt.Fprintf(w, "\n%s — %s\n", current, groupBlurb[current])
+		}
+		fmt.Fprintf(w, "  %-11s %s\n", c.Name, c.Short)
+	}
+	fmt.Fprint(w, "\n`kettle help <command>` for one command in full.\n")
+}
+
+func printCommand(w *os.File, c *Command) {
+	fmt.Fprintf(w, "%s\n\n%s\n", c.Usage(), c.Short)
+	if c.Long != "" {
+		fmt.Fprintf(w, "\n%s\n", strings.TrimSpace(c.Long))
+	}
+	if flags := c.Flags(); len(flags) > 0 {
+		fmt.Fprint(w, "\nflags:\n")
+		for _, f := range flags {
+			name := "--" + f.Name
+			if f.DefValue != "" && f.DefValue != "false" {
+				name += "=" + f.DefValue
+			}
+			fmt.Fprintf(w, "  %-22s %s\n", name, f.Usage)
+		}
+	}
+	if len(c.Examples) > 0 {
+		fmt.Fprint(w, "\nexamples:\n")
+		for _, e := range c.Examples {
+			fmt.Fprintf(w, "  %s\n      %s\n", e.Cmd, e.What)
+		}
+	}
+}
+
+// permute moves flags ahead of positional arguments.
+//
+// The standard flag package stops parsing at the first non-flag argument, so
+// `kettle ac <id> --check 3` would hand --check to the command as a positional
+// and tick nothing. Every other CLI an operator uses interleaves the two, and
+// a tool that silently ignores a flag because of where it was typed is worse
+// than one that rejects it.
+//
+// A flag that takes a value swallows the next argument, which is why this needs
+// the FlagSet: only the set knows whether --check wants one. `--` ends the
+// permutation, and everything after it is positional whatever it looks like.
+func permute(fs *flag.FlagSet, args []string) []string {
+	var flags, positional []string
+	for i := 0; i < len(args); i++ {
+		a := args[i]
+		if a == "--" {
+			positional = append(positional, args[i+1:]...)
+			break
+		}
+		if len(a) < 2 || a[0] != '-' {
+			positional = append(positional, a)
+			continue
+		}
+		flags = append(flags, a)
+		if strings.Contains(a, "=") {
+			continue
+		}
+		f := fs.Lookup(strings.TrimLeft(a, "-"))
+		// An unknown flag consumes nothing; Parse will reject it by name in a
+		// moment, which is a better message than one about its value.
+		if f == nil || isBoolFlag(f.Value) {
+			continue
+		}
+		if i+1 < len(args) {
+			i++
+			flags = append(flags, args[i])
+		}
+	}
+	return append(flags, positional...)
+}
+
+func isBoolFlag(v flag.Value) bool {
+	b, ok := v.(interface{ IsBoolFlag() bool })
+	return ok && b.IsBoolFlag()
+}
+
+func groupIndex(g string) int {
+	for i, name := range groupOrder {
+		if name == g {
+			return i
+		}
+	}
+	return len(groupOrder)
+}
+
+type discard struct{}
+
+func (discard) Write(p []byte) (int, error) { return len(p), nil }

cli/internal/cmd/comment.go

@@ -0,0 +1,146 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "comment",
+		Group: GroupSync,
+		Args:  "<id>",
+		Short: "post or edit a comment on a synced issue",
+		Long: `The target is a LOCAL ID, not a number. Which issue this is, is a fact about the
+work; where it lives in the tracker is bookkeeping, and the ` + "`gitea:`" + ` handle on the
+file is what turns one into the other. An ` + "`origin: local`" + ` issue cannot be
+commented on at all — it is not in the tracker, so there is nothing there to
+comment on; push it first.
+
+The body comes from a file or from --body, and multi-line prose is what --file
+is for. This is why comments go through the API rather than through a tracker
+CLI: an entity command with an empty-looking positional opens $EDITOR, and on a
+TTY that does not exist it hangs forever.
+
+After the write the whole thread is refetched into ` + "`<id>.comments.md`" + `, so the
+local copy is not stale by one comment — the one this run just made.
+
+COMMENTS ARE PULL-ONLY IN THE STORE. Nothing round-trips them back: editing
+` + "`<id>.comments.md`" + ` by hand changes nothing in the tracker. Use --edit with a
+comment id for that.`,
+		Examples: []Example{
+			{"kettle comment wire-sqlc-appclick --file notes.md", "post the contents of a file"},
+			{`kettle comment wire-sqlc-appclick --body "готово, задеплоено"`, "post one line"},
+			{"kettle comment wire-sqlc-appclick --file fix.md --edit 1234", "rewrite comment 1234 instead"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			file := fs.String("file", "", "markdown file holding the comment body")
+			body := fs.String("body", "", "comment body inline (short, single-line)")
+			edit := fs.Int64("edit", 0, "comment id to rewrite, instead of posting a new one")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				if len(args) != 1 {
+					return Fail("give exactly one issue id")
+				}
+				id := args[0]
+				withFile, withBody := wasSet(fs, "file"), wasSet(fs, "body")
+				switch {
+				case withFile && withBody:
+					return Fail("--file and --body are mutually exclusive")
+				case !withFile && !withBody:
+					return Fail("give the comment body: --file <path>, or --body \"…\"")
+				}
+
+				root, client, err := syncStartExisting(*out)
+				if err != nil {
+					return err
+				}
+				i, err := issue.Load(root, id)
+				if err != nil {
+					return Fail("no issue %q in %s", id, root)
+				}
+				key, ok := mapping.RemoteKeyOf(i)
+				if !ok {
+					return Fail("%s is local-only (origin: %s, no usable `%s:` handle) — "+
+						"there is nothing in the tracker to comment on; `kettle push %s` first",
+						id, i.Origin, mapping.GiteaKey, id)
+				}
+				text, err := commentBodyFrom(*file, *body)
+				if err != nil {
+					return err
+				}
+
+				// The handle names the repository, so a comment lands where the
+				// issue actually is — even when the store has ever pointed at two.
+				client = client.For(key.Repo)
+
+				var got *wire.Comment
+				verb := "posted"
+				if *edit != 0 {
+					verb = "edited"
+					got, err = client.EditComment(*edit, text, fmt.Sprintf("comment-%d", *edit))
+				} else {
+					got, err = client.CreateComment(key.Number, text, "comment-"+id)
+				}
+				if err != nil {
+					return err
+				}
+				// A 2xx that carries no id is not a comment. Nothing local has been
+				// written yet, and nothing will be if the answer is that shape.
+				if got.ID == 0 {
+					return Fail("the %s answer carries no comment id — nothing local was changed", verb)
+				}
+				fmt.Printf("%s comment %d on %s (%s)  %s\n", verb, got.ID, id, key, got.HTMLURL)
+
+				comments, err := client.ListComments(key.Number)
+				if err != nil {
+					return Fail("the comment went up, but refetching the thread failed: %v — "+
+						"`kettle pull %d` to refresh the local copy", err, key.Number)
+				}
+				path := commentsSidecarPath(root, id)
+				if len(comments) == 0 {
+					// Only reachable when the thread was emptied elsewhere between
+					// the write and the read. A stale sidecar for a thread that no
+					// longer exists is worse than no sidecar.
+					if err := os.Remove(path); err != nil && !os.IsNotExist(err) {
+						return err
+					}
+					fmt.Printf("thread: none — %s removed\n", path)
+					return nil
+				}
+				if err := os.WriteFile(path, []byte(mapping.RenderComments(comments)), 0o644); err != nil {
+					return err
+				}
+				fmt.Printf("thread: %s (%d comment(s))\n", path, len(comments))
+				return nil
+			}
+		},
+	})
+}
+
+// commentBodyFrom reads the comment body from a file or takes it as given.
+//
+// Trimmed and then required to be non-empty: a file of whitespace is somebody
+// pointing at the wrong path, and posting it would leave an empty comment in a
+// thread that nobody can delete from here.
+func commentBodyFrom(file, inline string) (string, error) {
+	if file != "" {
+		raw, err := os.ReadFile(file)
+		if err != nil {
+			return "", Fail("cannot read the comment body: %v", err)
+		}
+		inline = string(raw)
+	}
+	text := strings.TrimSpace(inline)
+	if text == "" {
+		return "", Fail("the comment body is empty — nothing was posted")
+	}
+	return text, nil
+}

cli/internal/cmd/config.go

@@ -0,0 +1,59 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+)
+
+func init() {
+	register(&Command{
+		Name:  "config",
+		Group: GroupProject,
+		Short: "show what this project resolved to",
+		Long: `Every path and every setting, with the overrides already applied, so a run that
+went somewhere unexpected can be explained without guessing.
+
+The token is never printed — only whether one was found.
+
+This is the command to reach for when the store looks empty, when a push says
+401, or when two directories disagree about which project they are in.`,
+		Examples: []Example{
+			{"kettle config", "resolved paths and settings"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			return func(args []string) error {
+				root := project.Root("")
+				if root == "" {
+					return project.NotFoundError("")
+				}
+
+				fmt.Printf("project   %s\n", root)
+				fmt.Printf("store     %s\n", issue.Root(""))
+				fmt.Printf("payload   %s\n", project.PayloadRoot(""))
+				fmt.Printf("config    %s\n", config.ProjectPath(""))
+				fmt.Printf("logins    %s\n", config.LoginsPath())
+
+				r, err := config.Resolve("")
+				if err != nil {
+					fmt.Println()
+					return err
+				}
+				red := r.Redacted()
+				fmt.Println()
+				fmt.Printf("login     %s\n", orNone(red.Login))
+				fmt.Printf("url       %s\n", orNone(red.URL))
+				fmt.Printf("token     %s\n", orNone(red.Token))
+				if r.Owner != "" {
+					fmt.Printf("repo      %s\n", r.Slug())
+				} else {
+					fmt.Printf("repo      none\n")
+				}
+				return nil
+			}
+		},
+	})
+}

cli/internal/cmd/evict.go

@@ -0,0 +1,115 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+func init() {
+	register(&Command{
+		Name:  "evict",
+		Group: GroupIssue,
+		Args:  "[<id>…]",
+		Short: "remove closed issues from the local store",
+		Long: `The store is a working set, not an archive. What is evicted is two conditions,
+both read off the file:
+
+    state: closed      the work is done
+    origin: <tracker>  the work is somewhere else too
+
+THE SECOND CONDITION IS THE WHOLE SAFETY ARGUMENT. ` + "`origin: local`" + ` means this
+file IS the issue — there is no other copy and deleting it deletes the work. It
+is never evicted, in any state, not even when named explicitly on the command
+line: a closed local issue is reported and kept.
+
+Eviction asks the file rather than the tracker, because state and origin are
+domain fields and the answer is already in the store — which is why this needs
+no network and no login. ` + "`kettle sync-evict`" + ` is the variant that refreshes state
+from the tracker first and then makes the same decision.
+
+Not a one-off migration: a pull by number fetches an issue in any state, so a
+closed issue pulled after an eviction lands on disk again. Evict it again when
+you are done with it.
+
+INDEX.md is rebuilt, because it IS a view of the directory. The number -> slug
+ledger is deliberately not pruned: its entries outlive the files they name, and
+that is what makes a pull land on the same slug afterwards.`,
+		Examples: []Example{
+			{"kettle evict", "every closed issue that is not origin: local"},
+			{"kettle evict old-thing another-thing", "only these"},
+			{"kettle evict --dry-run", "print what would go; touch nothing"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			dryRun := fs.Bool("dry-run", false, "print what would be removed; touch nothing")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				if !issue.StoreExists(root) {
+					return Fail("store %s does not exist — nothing to evict", root)
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+				var missing []string
+				for _, id := range args {
+					if _, ok := issues[id]; !ok {
+						missing = append(missing, id)
+					}
+				}
+				if len(missing) > 0 {
+					return Fail("no such issue(s) in the store: %s", strings.Join(missing, ", "))
+				}
+
+				rep, err := issue.Evict(root, issues, args, *dryRun)
+				if err != nil {
+					return err
+				}
+				printEviction(rep, len(args) > 0)
+				return nil
+			}
+		},
+	})
+}
+
+func printEviction(rep *issue.EvictReport, named bool) {
+	verb := "evicted"
+	if rep.DryRun {
+		verb = "would evict"
+	}
+	for _, e := range rep.Evicted {
+		fmt.Printf("%-11s  %s\n", verb, e.ID)
+		for _, p := range e.Paths {
+			fmt.Printf("    %s\n", p)
+		}
+	}
+	for _, k := range rep.Kept {
+		// An open issue is the normal case and says nothing worth a line —
+		// unless the operator named it, in which case they are owed the reason.
+		if k.Open && !named {
+			continue
+		}
+		if k.Open {
+			fmt.Printf("%-11s  %s  %s\n", "kept", k.ID, k.Why)
+		} else {
+			fmt.Printf("%-11s  %s  closed, %s\n", "kept", k.ID, k.Why)
+		}
+	}
+
+	if rep.DryRun {
+		fmt.Printf("%d issue(s) would be evicted, %d kept — nothing was touched\n",
+			len(rep.Evicted), len(rep.Kept))
+		return
+	}
+	fmt.Printf("%d issue(s) evicted, %d kept\n", len(rep.Evicted), len(rep.Kept))
+	if rep.IndexPath != "" {
+		fmt.Printf("index: %s — %d issue(s)\n", rep.IndexPath, rep.IndexCount)
+	}
+}

cli/internal/cmd/evict_sync.go

@@ -0,0 +1,229 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"sort"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "sync-evict",
+		Group: GroupSync,
+		Args:  "[<id>…]",
+		Short: "refresh state from the tracker, then evict what is closed",
+		Long: `` + "`kettle evict`" + ` is the command that decides and deletes. This adds exactly one
+thing in front of it: a ` + "`state:`" + ` that is not stale. A local ` + "`state:`" + ` is only as
+fresh as the last pull, so an issue closed in the web UI an hour ago still reads
+` + "`open`" + ` here and the offline command will — correctly — leave it alone. That is
+the gap this closes, and before it existed the operator had to pull the five
+closed issues back onto disk before anything could remove them.
+
+ORDER OF OPERATIONS, AND IT IS THE WHOLE SAFETY ARGUMENT:
+
+    1. every candidate's state is fetched — ALL of them, before anything is
+       removed;
+    2. each answer must be the issue that was asked about, in a state the domain
+       recognizes;
+    3. only then is the eviction run, by handing the refreshed issues to the
+       domain — the same decision, the same deletion, the same protection of
+       ` + "`origin: local`" + `, in one place.
+
+A dead connection, a non-2xx, an answer about another issue, a state nobody
+recognizes: the run stops at step 2 and NOTHING is deleted, not even the issues
+whose answers had already arrived. That is stricter than push, which deletes as
+it goes, and it costs nothing here — there is no ordering constraint between
+evictions, so there is no reason to start before every answer is in.
+
+A candidate is an issue carrying a ` + "`gitea:`" + ` handle. ` + "`origin: local`" + ` work has
+none, is never asked about, and is never evicted — it is not in the tracker to
+be closed. A tracked issue whose handle is missing or unreadable cannot be
+verified, so it is reported and kept rather than guessed at.
+
+Cost: one request per candidate. The store is a working set that push keeps
+small, and a wrong answer here deletes a file, so each issue is asked about by
+its own address rather than inferred from a list a limit could have truncated.
+
+The refreshed state is written back even for the issues that stay: the answer is
+already paid for, and a store that keeps a state the tracker has disowned is the
+thing this command exists to fix.`,
+		Examples: []Example{
+			{"kettle sync-evict", "ask about every synced issue; evict the closed ones"},
+			{"kettle sync-evict old-thing another-thing", "only these"},
+			{"kettle sync-evict --dry-run", "ask, report, write and delete nothing"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			dryRun := fs.Bool("dry-run", false, "ask the tracker and report; write and delete nothing")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				root, client, err := syncStartExisting(*out)
+				if err != nil {
+					return err
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+				var missing []string
+				for _, id := range args {
+					if _, ok := issues[id]; !ok {
+						missing = append(missing, id)
+					}
+				}
+				if len(missing) > 0 {
+					return Fail("no such issue(s) in the store: %s", strings.Join(missing, ", "))
+				}
+
+				checkable, unverifiable, local := syncEvictCandidates(issues, args)
+				for _, s := range unverifiable {
+					fmt.Fprintf(os.Stderr, "warning: %s: %s — kept, and not asked about\n", s.id, s.why)
+				}
+				// A local issue is reported only when the operator named it: they
+				// asked about this file by name and are owed the reason it stayed.
+				if len(args) > 0 {
+					for _, s := range local {
+						fmt.Printf("%-11s  %s  %s\n", "kept", s.id, s.why)
+					}
+				}
+				if len(checkable) == 0 {
+					fmt.Println("nothing to check: nothing named carries a `gitea:` handle")
+					return nil
+				}
+
+				// Every answer first, deletions after.
+				fresh := make(map[string]string, len(checkable))
+				for _, c := range checkable {
+					got, err := client.For(c.key.Repo).GetIssue(c.key.Number)
+					if err != nil {
+						return Fail("%s: asking the tracker about %s failed: %v\nNothing was evicted.",
+							c.id, c.key, err)
+					}
+					state, ok := syncEvictConfirms(got, c.key.Number)
+					if !ok {
+						return Fail("%s: the answer for %s does not confirm a state "+
+							"(issue #%d, state %q). Nothing was evicted.",
+							c.id, c.key, got.Number, got.State)
+					}
+					fresh[c.id] = state
+				}
+
+				// The store stops lying even about the issues that stay. This is
+				// the only write made before the decision, and a dry run makes
+				// none.
+				changed := 0
+				for _, c := range checkable {
+					was := issues[c.id].State
+					if was == fresh[c.id] {
+						continue
+					}
+					fmt.Printf("%-11s  %s  %s -> %s\n", "state", c.id, was, fresh[c.id])
+					issues[c.id].State = fresh[c.id]
+					if *dryRun {
+						continue
+					}
+					if _, err := issue.Save(root, issues[c.id]); err != nil {
+						return err
+					}
+					changed++
+				}
+
+				ids := make([]string, 0, len(checkable))
+				for _, c := range checkable {
+					ids = append(ids, c.id)
+				}
+				rep, err := issue.Evict(root, issues, ids, *dryRun)
+				if err != nil {
+					return err
+				}
+				printEviction(rep, len(args) > 0)
+
+				// Evict rebuilds INDEX.md when something went; a state written back
+				// without an eviction changed the store too, and the index is a
+				// view of it. Neither happening means nothing changed on disk, and
+				// then nothing is rewritten.
+				if changed > 0 && rep.IndexPath == "" {
+					path, n, err := issue.BuildIndex(root)
+					if err != nil {
+						return err
+					}
+					fmt.Printf("index: %s — %d issue(s)\n", path, n)
+				}
+				return nil
+			}
+		},
+	})
+}
+
+// syncEvictTarget is one issue the tracker can be asked about, and the address
+// to ask at — its own, so an issue that lives in another repository is asked
+// about there.
+type syncEvictTarget struct {
+	id  string
+	key wire.Key
+}
+
+// syncEvictSkip is an issue that was not asked about, with the reason.
+type syncEvictSkip struct{ id, why string }
+
+// syncEvictCandidates splits the store into what the tracker can be asked
+// about, what names a tracker but cannot be reached, and what is local.
+//
+// An unverifiable issue names a tracker but carries no handle to reach it by,
+// which is a file to report and never one to delete on a guess. A local issue is
+// in neither of those: it has no handle because it has never left this machine,
+// and asking about it is not a question that has an answer.
+//
+// ids restricts the question to those issues; empty asks about the whole store.
+func syncEvictCandidates(issues map[string]*issue.Issue, ids []string) (checkable []syncEvictTarget, unverifiable, local []syncEvictSkip) {
+	chosen := ids
+	if len(chosen) == 0 {
+		for id := range issues {
+			chosen = append(chosen, id)
+		}
+		sort.Strings(chosen)
+	}
+	for _, id := range chosen {
+		i, ok := issues[id]
+		if !ok {
+			continue
+		}
+		if i.IsLocal() {
+			local = append(local, syncEvictSkip{id, issue.LocalReason})
+			continue
+		}
+		key, ok := mapping.RemoteKeyOf(i)
+		if !ok {
+			unverifiable = append(unverifiable, syncEvictSkip{id,
+				fmt.Sprintf("origin: %s but no usable `%s:` handle", i.Origin, mapping.GiteaKey)})
+			continue
+		}
+		checkable = append(checkable, syncEvictTarget{id: id, key: key})
+	}
+	return checkable, unverifiable, local
+}
+
+// syncEvictConfirms is the state the tracker confirmed for this number — the
+// deletion gate.
+//
+// Deliberately boring, and saying no by default, because everything downstream
+// of a yes here may delete a file. An answer counts only when it is about the
+// very issue that was asked about and names a state the domain recognizes. A
+// non-2xx never reaches this: the transport has already returned an error.
+func syncEvictConfirms(got *wire.Issue, number int) (string, bool) {
+	if got == nil || got.Number != number {
+		return "", false
+	}
+	for _, s := range issue.States {
+		if got.State == s {
+			return got.State, true
+		}
+	}
+	return "", false
+}

cli/internal/cmd/flags.go

@@ -0,0 +1,54 @@
+package cmd
+
+import (
+	"flag"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+)
+
+// storeFlag registers the one flag almost every command has.
+//
+// An explicit --out overrides the resolved store and is used exactly as typed:
+// a relative --out stays relative to the working directory, because that is
+// what the operator asked for.
+func storeFlag(fs *flag.FlagSet) *string {
+	return fs.String("out", "", "store root (default: <project>/.kettle/issues)")
+}
+
+// storeRoot resolves the store, or explains which directories were searched.
+//
+// No marker anywhere is an answer, not a fallback: a store placed in a
+// plausible-looking directory is the failure the marker exists to replace.
+func storeRoot(out string) (string, error) {
+	if root := issue.Root(out); root != "" {
+		return root, nil
+	}
+	return "", project.NotFoundError("")
+}
+
+// wasSet reports whether the operator actually typed this flag.
+//
+// Needed wherever the empty string is a legitimate value to reject rather than
+// a synonym for "not given": `--check ""` is an empty selector and an error,
+// while no --check at all means "just list the boxes".
+func wasSet(fs *flag.FlagSet, name string) bool {
+	found := false
+	fs.Visit(func(f *flag.Flag) {
+		if f.Name == name {
+			found = true
+		}
+	})
+	return found
+}
+
+// stringList is a repeatable flag: --label tech/sql --label comp/appclick.
+type stringList []string
+
+func (l *stringList) String() string { return strings.Join(*l, ", ") }
+
+func (l *stringList) Set(v string) error {
+	*l = append(*l, v)
+	return nil
+}

cli/internal/cmd/gen.go

@@ -0,0 +1,322 @@
+package cmd
+
+import (
+	"errors"
+	"flag"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+	"unicode/utf8"
+)
+
+// The region markers. What sits between them belongs to the generator; the
+// rest of the file belongs to whoever wrote it.
+const (
+	genOpen  = "<!-- kettle:gen -->"
+	genClose = "<!-- /kettle:gen -->"
+)
+
+// genBanner opens every generated region. The first thing anybody who finds
+// the block wants to do is edit it in place, so the block says who wrote it and
+// which command writes it again.
+const genBanner = "**Generated from the kettle command registry by `kettle gen skills`.** " +
+	"Everything between the two markers is replaced on the next run — " +
+	"hand-written prose belongs outside them."
+
+// exampleAlign is the widest example command that still gets its `# what`
+// padded into a column. One long pipeline would otherwise push every other
+// comment off the right edge of the page.
+const exampleAlign = 56
+
+func init() {
+	register(&Command{
+		Name:  "gen",
+		Group: GroupProject,
+		Args:  "skills",
+		Short: "write the plugin's SKILL.md files from the command registry",
+		Long: `A SKILL.md tells an agent how to invoke this binary. Hand-written, it drifts: a
+flag is renamed here and the documentation goes on recommending the old one,
+and the agent that reads it fails in a way nobody traces back to a stale
+sentence. Everything those files say about a command — its usage line, its
+flags with their defaults, its worked examples — is already in the registry
+this binary is built from, so it is written from there and cannot disagree.
+
+THE GENERATOR OWNS A REGION, NOT A FILE. Each SKILL.md carries a pair of HTML
+comment markers — ` + "`kettle:gen`" + ` to open and ` + "`/kettle:gen`" + ` to close, both written in
+the ` + "`<!-- … -->`" + ` form and visible at the top and bottom of the block below.
+Everything between them is replaced on every run; every byte outside them comes
+back exactly as it was, which matters most for ` + "`description:`" + `, the prose that
+decides whether an agent loads the skill at all, and the one thing here that no
+generator can write.
+
+A file with no markers is REPORTED AND LEFT ALONE, never overwritten: clobbering
+somebody's prose because they forgot a marker is the failure this design exists
+to prevent. A file that does not exist yet is created with a frontmatter stub
+around a generated block, for a human to fill in.
+
+The output is deterministic to the byte — no timestamps, no map iteration — so
+regenerating something that has not changed produces no diff. --check is that
+property made useful: it writes nothing and exits 1 when any file on disk
+differs from what would be generated, which is what a pre-commit hook or a CI
+step calls. It wins over --dry-run when both are given.`,
+		Examples: []Example{
+			{"kettle gen skills --out ../plugins/tea/skills", "write the region in every group's SKILL.md"},
+			{"kettle gen skills --out ../plugins/tea/skills --dry-run", "print what would change; write nothing"},
+			{"kettle gen skills --out ../plugins/tea/skills --check", "exit 1 if the docs are out of date"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			out := fs.String("out", "", "directory the skills live in; one <group>/SKILL.md under it")
+			dryRun := fs.Bool("dry-run", false, "print what would change; write nothing")
+			check := fs.Bool("check", false, "write nothing, exit 1 if anything is out of date")
+
+			return func(args []string) error {
+				target := "skills"
+				if len(args) > 0 {
+					target = args[0]
+				}
+				if len(args) > 1 || target != "skills" {
+					return Fail("the only target is `skills` — try `kettle gen skills --out <dir>`")
+				}
+				if *out == "" {
+					return Fail("--out is required — the directory the SKILL.md files live under")
+				}
+				return genSkills(*out, *dryRun, *check)
+			}
+		},
+	})
+}
+
+// errNoRegion is what a file that the generator may not touch reports.
+var errNoRegion = errors.New("no " + genOpen + " … " + genClose + " region")
+
+func genSkills(dir string, dryRun, check bool) error {
+	// --check is a read-only question about the working tree, so it overrules
+	// --dry-run rather than combining with it.
+	if check {
+		dryRun = true
+	}
+
+	groups := docGroups()
+	var written, unchanged, outdated, kept int
+	for _, group := range groups {
+		path := filepath.Join(dir, group, "SKILL.md")
+		block, err := renderGroup(commandsIn(group))
+		if err != nil {
+			return err
+		}
+
+		existing, err := os.ReadFile(path)
+		switch {
+		case errors.Is(err, fs.ErrNotExist):
+			outdated++
+			if check {
+				fmt.Printf("%-13s %s\n", "missing", path)
+				continue
+			}
+			if dryRun {
+				fmt.Printf("%-13s %s\n", "would create", path)
+				continue
+			}
+			if err := writeFile(path, stubFile(group, block)); err != nil {
+				return err
+			}
+			written++
+			fmt.Printf("%-13s %s\n", "created", path)
+
+		case err != nil:
+			return err
+
+		default:
+			want, err := spliceRegion(string(existing), block)
+			if err != nil {
+				// Reported, never repaired: a missing marker is somebody's
+				// prose sitting where the block used to be.
+				kept++
+				fmt.Fprintf(os.Stderr, "kettle gen: %s left alone — %v\n", path, err)
+				continue
+			}
+			if want == string(existing) {
+				unchanged++
+				fmt.Printf("%-13s %s\n", "unchanged", path)
+				continue
+			}
+			outdated++
+			if check {
+				fmt.Printf("%-13s %s\n", "stale", path)
+				continue
+			}
+			if dryRun {
+				fmt.Printf("%-13s %s\n", "would update", path)
+				continue
+			}
+			if err := writeFile(path, want); err != nil {
+				return err
+			}
+			written++
+			fmt.Printf("%-13s %s\n", "updated", path)
+		}
+	}
+
+	switch {
+	case check:
+		fmt.Printf("%d file(s) checked, %d out of date, %d without a region\n",
+			len(groups), outdated, kept)
+		if outdated > 0 {
+			fmt.Printf("run `kettle gen skills --out %s`\n", dir)
+			return SilentError{Code: 1}
+		}
+	case dryRun:
+		fmt.Printf("%d file(s) would change, %d unchanged, %d without a region — nothing was written\n",
+			outdated, unchanged, kept)
+	default:
+		fmt.Printf("%d file(s) written, %d unchanged, %d without a region\n", written, unchanged, kept)
+	}
+	return nil
+}
+
+// docGroups lists the groups that have commands, in the order Commands()
+// returns them — the same order twice, so two runs cannot differ.
+func docGroups() []string {
+	var out []string
+	seen := map[string]bool{}
+	for _, c := range Commands() {
+		if c.Group == "" {
+			fmt.Fprintf(os.Stderr, "kettle gen: command %q has no group and is in no skill\n", c.Name)
+			continue
+		}
+		if !seen[c.Group] {
+			seen[c.Group] = true
+			out = append(out, c.Group)
+		}
+	}
+	return out
+}
+
+func commandsIn(group string) []*Command {
+	var out []*Command
+	for _, c := range Commands() {
+		if c.Group == group {
+			out = append(out, c)
+		}
+	}
+	return out
+}
+
+// renderGroup is the generated block for one group, without the markers and
+// without a trailing newline.
+func renderGroup(cmds []*Command) (string, error) {
+	var b strings.Builder
+	b.WriteString(genBanner)
+	b.WriteString("\n")
+
+	for _, c := range cmds {
+		text := renderCommand(c)
+		// A block holding either marker would cut itself in half on the next
+		// run — the splice would end the region in the middle of the prose that
+		// mentions it. Loud here rather than quietly truncated on disk.
+		if strings.Contains(text, genOpen) || strings.Contains(text, genClose) {
+			return "", Fail("command %q spells a region marker out in full; the generated block would then end inside itself — write it another way", c.Name)
+		}
+		b.WriteString(text)
+	}
+	return strings.TrimRight(b.String(), "\n"), nil
+}
+
+func renderCommand(c *Command) string {
+	var b strings.Builder
+	fmt.Fprintf(&b, "\n## `%s`\n\n%s\n", c.Usage(), c.Short)
+	if long := strings.TrimSpace(c.Long); long != "" {
+		b.WriteString("\n" + long + "\n")
+	}
+	if flags := c.Flags(); len(flags) > 0 {
+		b.WriteString("\n| flag | default | what it does |\n| --- | --- | --- |\n")
+		for _, f := range flags {
+			fmt.Fprintf(&b, "| `--%s` | %s | %s |\n", f.Name, defaultCell(f.DefValue), cell(f.Usage))
+		}
+	}
+	if len(c.Examples) > 0 {
+		w := exampleWidth(c.Examples)
+		b.WriteString("\n```bash\n")
+		for _, e := range c.Examples {
+			pad := w - utf8.RuneCountInString(e.Cmd)
+			if pad < 0 {
+				pad = 0
+			}
+			fmt.Fprintf(&b, "%s%s  # %s\n", e.Cmd, strings.Repeat(" ", pad), e.What)
+		}
+		b.WriteString("```\n")
+	}
+	return b.String()
+}
+
+// exampleWidth is the column the `# what` comments line up at. Runes, not
+// bytes: an example with Cyrillic in it would otherwise pull the column left by
+// however many multi-byte characters it holds.
+func exampleWidth(examples []Example) int {
+	w := 0
+	for _, e := range examples {
+		if n := utf8.RuneCountInString(e.Cmd); n > w && n <= exampleAlign {
+			w = n
+		}
+	}
+	return w
+}
+
+func defaultCell(v string) string {
+	if v == "" {
+		return "—"
+	}
+	return "`" + cell(v) + "`"
+}
+
+// cell keeps a value from breaking out of its table row.
+func cell(s string) string {
+	s = strings.ReplaceAll(s, "\n", " ")
+	return strings.ReplaceAll(s, "|", `\|`)
+}
+
+func region(block string) string {
+	return genOpen + "\n" + block + "\n" + genClose
+}
+
+// spliceRegion swaps the block into existing, leaving every other byte alone.
+func spliceRegion(existing, block string) (string, error) {
+	start := strings.Index(existing, genOpen)
+	if start < 0 {
+		return "", errNoRegion
+	}
+	rest := start + len(genOpen)
+	end := strings.Index(existing[rest:], genClose)
+	if end < 0 {
+		return "", fmt.Errorf("%s is missing its %s", genOpen, genClose)
+	}
+	return existing[:start] + region(block) + existing[rest+end+len(genClose):], nil
+}
+
+// stubFile is a new SKILL.md: the least frontmatter that is still a skill,
+// and the region.
+//
+// The description is left as a TODO on purpose. It is the sentence that decides
+// whether an agent loads this skill at all — prose a human tunes against real
+// failures to trigger, and the one thing here a generator has no way to write.
+func stubFile(group, block string) string {
+	title := "# kettle " + group + "\n"
+	if blurb := groupBlurb[group]; blurb != "" {
+		title += "\n" + blurb + "\n"
+	}
+	return "---\n" +
+		"name: " + group + "\n" +
+		"description: TODO — write this by hand. It is the only thing that decides whether an agent loads this skill at all, so it is prose a human tunes; kettle gen never reads or writes it.\n" +
+		"---\n\n" +
+		title + "\n" +
+		region(block) + "\n"
+}
+
+func writeFile(path, content string) error {
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		return err
+	}
+	return os.WriteFile(path, []byte(content), 0o644)
+}

cli/internal/cmd/gen_test.go

@@ -0,0 +1,281 @@
+package cmd_test
+
+// `kettle gen` writes documentation an agent reads to invoke this binary, into
+// files a human also writes prose in. Both halves of that are tested here: what
+// it produces has to be the same twice over, and what it does NOT own has to
+// come back byte for byte.
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+const (
+	genOpen  = "<!-- kettle:gen -->"
+	genClose = "<!-- /kettle:gen -->"
+)
+
+func TestGenWritesOneFilePerGroupAndIsIdempotent(t *testing.T) {
+	dir := t.TempDir()
+	out := filepath.Join(dir, "skills")
+
+	first := mustRun(t, dir, "gen", "skills", "--out", out)
+	for _, group := range []string{"project", "issue", "sync"} {
+		path := filepath.Join(out, group, "SKILL.md")
+		raw, err := os.ReadFile(path)
+		if err != nil {
+			t.Fatalf("%s was not created: %v\n%s", path, err, first.out())
+		}
+		body := string(raw)
+		// The frontmatter is what makes it a skill at all, and the description
+		// is prose a human tunes — the stub says so and generates nothing.
+		if !strings.HasPrefix(body, "---\nname: "+group+"\n") {
+			t.Errorf("%s has no frontmatter naming the group:\n%s", path, firstLines(body, 5))
+		}
+		if !strings.Contains(body, genOpen) || !strings.Contains(body, genClose) {
+			t.Errorf("%s was created without the region markers:\n%s", path, body)
+		}
+		// The block has to say what wrote it: the first thing anybody who finds
+		// it will want to do is edit it in place.
+		if !strings.Contains(body, "kettle gen skills") {
+			t.Errorf("%s does not name the command that regenerates it:\n%s", path, body)
+		}
+	}
+
+	// One command's documentation, end to end: usage line, short, a flag out of
+	// the flag set, and a worked example with its explanation beside it.
+	issues, err := os.ReadFile(filepath.Join(out, "issue", "SKILL.md"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, want := range []string{
+		"## `kettle evict [<id>…]`",
+		"remove closed issues from the local store",
+		"| `--dry-run` | `false` | print what would be removed; touch nothing |",
+		"kettle evict --dry-run",
+		"# print what would go; touch nothing",
+	} {
+		if !strings.Contains(string(issues), want) {
+			t.Errorf("the issue group is missing %q:\n%s", want, issues)
+		}
+	}
+
+	// Deterministic to the byte: a regeneration of something that has not
+	// changed must produce no diff at all, or every run of a CI step is a
+	// spurious one.
+	before := readAll(t, out)
+	second := mustRun(t, dir, "gen", "skills", "--out", out)
+	if strings.Contains(second.stdout, "updated") {
+		t.Errorf("a second run rewrote a file:\n%s", second.out())
+	}
+	for path, content := range before {
+		if now := readFile(t, path); now != content {
+			t.Errorf("%s changed on a second run with nothing else changed", path)
+		}
+	}
+
+	if r := run(t, dir, "gen", "skills", "--out", out, "--check"); r.code != 0 {
+		t.Errorf("--check exited %d on files that were just written:\n%s", r.code, r.out())
+	}
+}
+
+// The generator owns a region, not a file. Everything outside the markers is
+// somebody's prose and comes back exactly as it was.
+func TestGenLeavesHandWrittenProseAlone(t *testing.T) {
+	dir := t.TempDir()
+	out := filepath.Join(dir, "skills")
+	mustRun(t, dir, "gen", "skills", "--out", out)
+
+	path := filepath.Join(out, "issue", "SKILL.md")
+	raw := readFile(t, path)
+	start := strings.Index(raw, genOpen)
+	end := strings.Index(raw, genClose) + len(genClose)
+	if start < 0 || end < len(genClose) {
+		t.Fatalf("no region in the generated file:\n%s", raw)
+	}
+
+	const above = "\n## Identity: the slug\n\nThe file name is the id, and it never changes.\n\n"
+	const below = "\n\n## Layering rule\n\nThis skill must keep working with the sync skill deleted.\n"
+	// A description a human tuned, in the frontmatter the generator must not
+	// touch: it is the only thing that decides whether the skill loads at all.
+	edited := strings.Replace(raw[:start], "description: TODO", "description: Work with this project's issues as units of work", 1)
+	edited += above + raw[start:end] + below
+	if err := os.WriteFile(path, []byte(edited), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	mustRun(t, dir, "gen", "skills", "--out", out)
+	after := readFile(t, path)
+	if after != edited {
+		t.Errorf("a no-op regeneration did not return the file byte for byte:\n--- want ---\n%s\n--- got ---\n%s", edited, after)
+	}
+
+	// And the prose survives a regeneration that actually rewrites the block.
+	shortened := strings.Replace(after, genClose, "the block was gutted by hand\n"+genClose, 1)
+	if err := os.WriteFile(path, []byte(shortened), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	mustRun(t, dir, "gen", "skills", "--out", out)
+	restored := readFile(t, path)
+	if restored != edited {
+		t.Error("regenerating the block did not restore it, or did not preserve the prose around it")
+	}
+	if !strings.Contains(restored, "description: Work with this project's issues") {
+		t.Errorf("the hand-tuned description was overwritten:\n%s", firstLines(restored, 5))
+	}
+	if !strings.Contains(restored, above) || !strings.Contains(restored, below) {
+		t.Errorf("hand-written prose outside the markers was lost:\n%s", restored)
+	}
+}
+
+// Clobbering somebody's prose because they forgot a marker is the failure this
+// whole design exists to prevent.
+func TestGenNeverOverwritesAFileWithoutMarkers(t *testing.T) {
+	dir := t.TempDir()
+	out := filepath.Join(dir, "skills")
+	path := filepath.Join(out, "issue", "SKILL.md")
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	const prose = "---\nname: issue\ndescription: hand written, every word of it\n---\n\n# Everything here is somebody's work\n"
+	if err := os.WriteFile(path, []byte(prose), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := mustRun(t, dir, "gen", "skills", "--out", out)
+	if got := readFile(t, path); got != prose {
+		t.Fatalf("a file with no markers was rewritten:\n%s", got)
+	}
+	// Left alone silently is how it drifts unnoticed, so it is reported — and
+	// on stderr, where a warning belongs.
+	if !strings.Contains(r.stderr, path) {
+		t.Errorf("the skipped file was not named on stderr:\n%s", r.out())
+	}
+	if !strings.Contains(r.stdout, "without a region") {
+		t.Errorf("the receipt did not account for it:\n%s", r.stdout)
+	}
+	// The other groups still got written — one unmanaged file stops nothing.
+	if _, err := os.Stat(filepath.Join(out, "sync", "SKILL.md")); err != nil {
+		t.Error("one file without markers stopped the whole run")
+	}
+}
+
+func TestGenCheckFailsOnAStaleFileAndNamesIt(t *testing.T) {
+	dir := t.TempDir()
+	out := filepath.Join(dir, "skills")
+	mustRun(t, dir, "gen", "skills", "--out", out)
+
+	stale := filepath.Join(out, "sync", "SKILL.md")
+	raw := readFile(t, stale)
+	edited := strings.Replace(raw, genClose, "kettle push --thoroughly-renamed-flag\n"+genClose, 1)
+	if err := os.WriteFile(stale, []byte(edited), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := run(t, dir, "gen", "skills", "--out", out, "--check")
+	if r.code != 1 {
+		t.Fatalf("--check exited %d, want 1 — this is what a hook or a CI step calls:\n%s", r.code, r.out())
+	}
+	if !strings.Contains(r.stdout, stale) {
+		t.Errorf("--check did not say which file is out of date:\n%s", r.out())
+	}
+	// A question about the tree, never an answer written into it.
+	if got := readFile(t, stale); got != edited {
+		t.Error("--check wrote to the file it was asked about")
+	}
+
+	// A file that is not there at all is out of date too, not a nothing.
+	if err := os.Remove(stale); err != nil {
+		t.Fatal(err)
+	}
+	if r := run(t, dir, "gen", "skills", "--out", out, "--check"); r.code != 1 {
+		t.Errorf("--check exited %d for a missing file, want 1:\n%s", r.code, r.out())
+	}
+	if _, err := os.Stat(stale); err == nil {
+		t.Error("--check created the file it was asked about")
+	}
+}
+
+func TestGenDryRunWritesNothingAtAll(t *testing.T) {
+	dir := t.TempDir()
+	out := filepath.Join(dir, "skills")
+
+	fresh := mustRun(t, dir, "gen", "skills", "--out", out, "--dry-run")
+	if !strings.Contains(fresh.stdout, "would create") {
+		t.Errorf("a dry run said nothing about what it would do:\n%s", fresh.out())
+	}
+	if _, err := os.Stat(out); err == nil {
+		t.Fatal("a dry run created the output directory")
+	}
+
+	// And on an existing tree: the file is described, never touched.
+	mustRun(t, dir, "gen", "skills", "--out", out)
+	path := filepath.Join(out, "issue", "SKILL.md")
+	edited := strings.Replace(readFile(t, path), genClose, "gutted\n"+genClose, 1)
+	if err := os.WriteFile(path, []byte(edited), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := mustRun(t, dir, "gen", "skills", "--out", out, "--dry-run")
+	if !strings.Contains(r.stdout, "would update") || !strings.Contains(r.stdout, "nothing was written") {
+		t.Errorf("the dry run did not report the pending change:\n%s", r.out())
+	}
+	if got := readFile(t, path); got != edited {
+		t.Error("a dry run rewrote the file")
+	}
+}
+
+func TestGenRefusesAnUnknownTargetAndAMissingOut(t *testing.T) {
+	dir := t.TempDir()
+
+	if r := run(t, dir, "gen", "skills"); r.code == 0 || !strings.Contains(r.stderr, "--out") {
+		t.Errorf("gen without --out must stop and say so:\n%s", r.out())
+	}
+	if r := run(t, dir, "gen", "agents", "--out", filepath.Join(dir, "x")); r.code == 0 {
+		t.Errorf("an unknown target must be refused:\n%s", r.out())
+	}
+	if _, err := os.Stat(filepath.Join(dir, "x")); err == nil {
+		t.Error("the refused run created its output directory anyway")
+	}
+}
+
+func readFile(t *testing.T, path string) string {
+	t.Helper()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(raw)
+}
+
+// readAll is every file under root, by path, for a byte-for-byte comparison
+// after a second run.
+func readAll(t *testing.T, root string) map[string]string {
+	t.Helper()
+	out := map[string]string{}
+	err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error {
+		if err != nil || info.IsDir() {
+			return err
+		}
+		raw, err := os.ReadFile(path)
+		if err != nil {
+			return err
+		}
+		out[path] = string(raw)
+		return nil
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	return out
+}
+
+func firstLines(s string, n int) string {
+	lines := strings.SplitN(s, "\n", n+1)
+	if len(lines) > n {
+		lines = lines[:n]
+	}
+	return strings.Join(lines, "\n")
+}

cli/internal/cmd/index.go

@@ -0,0 +1,46 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+func init() {
+	register(&Command{
+		Name:  "index",
+		Group: GroupIssue,
+		Short: "rebuild INDEX.md from what is on disk",
+		Long: `A map of the local store, nothing else. The ` + "`origin`" + ` column is the only place
+the index acknowledges that a tracker exists: ` + "`local`" + ` means the issue has never
+left this machine, anything else names the tracker it also lives in. Both are
+ordinary issues here.
+
+` + "`progress`" + ` counts the body's checkboxes, ticked over total, and is read off the
+body at build time rather than stored — a second copy of that state in a
+metadata field would be wrong by the next edit.
+
+An existing store with nothing in it is a legitimate thing to index and gets an
+"_empty_" table. A store that is not there is an error, not a directory to
+create.`,
+		Examples: []Example{
+			{"kettle index", "rebuild the index for this project"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			out := storeFlag(fs)
+			return func(args []string) error {
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				path, n, err := issue.BuildIndex(root)
+				if err != nil {
+					return Fail("%v — nothing was created; create an issue with `kettle new`, or pass --out", err)
+				}
+				fmt.Printf("%s — %d issue(s)\n", path, n)
+				return nil
+			}
+		},
+	})
+}

cli/internal/cmd/init.go

@@ -0,0 +1,157 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+)
+
+// writeConfig creates or updates .kettle/config.yaml, touching only the
+// settings it was given.
+//
+// Init is idempotent, and that has to include the config: re-running it to add
+// a repository must not silently drop the login somebody pinned last week.
+func writeConfig(root, login, repo string, dryRun bool) (string, error) {
+	path := filepath.Join(root, project.Marker, "config.yaml")
+	rel := filepath.Join(project.Marker, "config.yaml")
+
+	cfg, existed, err := config.ReadProjectFile(path)
+	if err != nil {
+		return "", err
+	}
+	changed := !existed
+	if login != "" && cfg.Login != login {
+		cfg.Login, changed = login, true
+	}
+	if repo != "" && cfg.Repo != repo {
+		cfg.Repo, changed = repo, true
+	}
+	if !changed {
+		return "", nil
+	}
+
+	verb := "updated"
+	if !existed {
+		verb = "created"
+	}
+	detail := "no login or repository pinned yet — `kettle init --login … --repo …`"
+	if cfg.Login != "" || cfg.Repo != "" {
+		detail = fmt.Sprintf("login: %s, repo: %s", orNone(cfg.Login), orNone(cfg.Repo))
+	}
+	if dryRun {
+		return fmt.Sprintf("%s %s (%s)", verb, rel, detail), nil
+	}
+	if err := config.SaveProject(path, cfg); err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("%s %s (%s)", verb, rel, detail), nil
+}
+
+func orNone(s string) string {
+	if s == "" {
+		return "none"
+	}
+	return s
+}
+
+func init() {
+	register(&Command{
+		Name:  "init",
+		Group: GroupProject,
+		Short: "make this directory a project that tracks issues",
+		Long: `Creates ` + "`.kettle/`" + ` — the marker every other command resolves the store from,
+and ` + "`.kettle/config.yaml`" + `, which says which tracker repository these issues
+belong to and which login to reach it under.
+
+The marker is deliberately something an operator makes, not something inferred
+from the tree: ` + "`.git`" + ` is in every clone, so anything that inferred a root from
+one would write issues into whatever it happened to be installed in.
+
+--login pins a name, never a credential. The tokens live in one file per
+machine, outside every working tree, managed with ` + "`kettle auth`" + `.
+
+All of it is idempotent: it creates .kettle/issues and .kettle/payload, migrates
+an older store in if it finds one (tmp/ or .tea/), writes the config without
+disturbing settings it was not given, and adds .kettle/ to .gitignore. Each
+migration is a move, not a copy — two stores is the state the marker exists to
+prevent — and it refuses to pick a winner when both sides hold a file of the
+same name.
+
+Do NOT run this inside a linked worktree. A worktree is the same project on
+another branch and reaches the store by a hop out to the main checkout; a marker
+here would give one project two stores, and the directory holding the second one
+disappears with the branch.`,
+		Examples: []Example{
+			{"kettle init", "initialize the current directory"},
+			{"kettle init --login noodles --repo claude-skills/marketplace", "and point it at a tracker"},
+			{"kettle init --at ~/code/x", "initialize somewhere else"},
+			{"kettle init --dry-run", "say what it would do, touch nothing"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			at := fs.String("at", "", "directory to initialize (default: the working directory)")
+			login := fs.String("login", "", "name of a login in the machine-wide file (see `kettle auth`)")
+			repo := fs.String("repo", "", "tracker repository, as owner/name")
+			dryRun := fs.Bool("dry-run", false, "report what would happen; change nothing")
+
+			return func(args []string) error {
+				root := *at
+				if root == "" {
+					wd, err := os.Getwd()
+					if err != nil {
+						return err
+					}
+					root = wd
+				}
+				root, err := filepath.Abs(root)
+				if err != nil {
+					return err
+				}
+				if fi, err := os.Stat(root); err != nil || !fi.IsDir() {
+					return Fail("%s is not a directory", root)
+				}
+
+				// A second marker inside an existing project gives it a second
+				// store, and the nearer one wins — which is a surprise worth
+				// naming before it happens, not after.
+				if existing := project.Root(root); existing != "" && existing != root {
+					fmt.Fprintf(os.Stderr,
+						"warning: %s already sits inside the project at %s — a second marker here gives it a second store, and the nearer one wins.\n",
+						root, existing)
+				}
+
+				if *repo != "" {
+					if owner, name, ok := strings.Cut(*repo, "/"); !ok || owner == "" || name == "" {
+						return Fail("--repo %q is not owner/name", *repo)
+					}
+				}
+
+				done, err := project.Init(root, *dryRun)
+				if err != nil {
+					return err
+				}
+
+				line, err := writeConfig(root, *login, *repo, *dryRun)
+				if err != nil {
+					return err
+				}
+				if line != "" {
+					done = append(done, line)
+				}
+
+				prefix := ""
+				if *dryRun {
+					prefix = "would: "
+				}
+				for _, l := range done {
+					fmt.Println(prefix + l)
+				}
+				return nil
+			}
+		},
+	})
+}

cli/internal/cmd/labels.go

@@ -0,0 +1,285 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"regexp"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "labels",
+		Group: GroupSync,
+		Short: "put the canonical type/* and severity/* labels into a repository",
+		Long: `Every ` + "`type/*`" + ` and every ` + "`severity/*`" + ` the domain taxonomy defines, created up
+front instead of trickling in as a side effect of whichever push first happens
+to use one. Until a name exists in the repository nobody can filter by it in the
+web UI, so somebody makes their own — foreign colour, no ` + "`exclusive`" + ` — and the
+set arrives in pieces over months.
+
+NO LABEL NAME IS SPELLED OUT HERE. The names come from the domain taxonomy and
+are painted by the mapping layer, because a hex code is how a tracker paints a
+chip and not what an issue is. Add a type over in the domain and the next run
+creates it.
+
+THE REPOSITORY'S OWN LABELS ARE READ BEFORE ANYTHING IS WRITTEN, and read from
+the repository, never from a cache — a cache answers "what did we create last
+time" and the question here is "what does this repository have right now". A
+name that matches exactly is left alone; a colour or ` + "`exclusive`" + ` that disagrees
+with the spec is reported, and corrected only under --fix. A name that merely
+RESEMBLES a canonical one (the same tail, up to case, separator and whatever
+namespace is in front: ` + "`x`" + `, ` + "`X`" + `, ` + "`kind/x`" + `, ` + "`type: x`" + ` against ` + "`type/x`" + `) is
+reported with its id and never touched — renaming somebody else's label is a
+decision, not a step.
+
+Out of scope by design: ` + "`tech/*`" + ` and ` + "`comp/*`" + `, which are open-ended and are
+created by push as they come up, and deleting or renaming anything at all. Only
+repository labels are read; an organization's own labels sit behind a different
+endpoint and are neither read nor written.
+
+The issue store is out of scope too, and not incidentally: a label belongs to
+the repository and not to any issue, so this neither reads the store nor creates
+it. Request bodies go to the transport's own scratchpad, which is a sibling of
+the store and never a child.`,
+		Examples: []Example{
+			{"kettle labels --dry-run", "print the plan; not one writing request"},
+			{"kettle labels", "create whatever is missing"},
+			{"kettle labels --fix", "also patch colour / exclusive drift"},
+			{"kettle labels --repo owner/name", "bootstrap another repository"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			dryRun := fs.Bool("dry-run", false, "print the plan; not one writing request")
+			fix := fs.Bool("fix", false, "also patch colour/exclusive on labels that already exist")
+			repo := fs.String("repo", "", "repository to bootstrap, as owner/name (default: this project's)")
+
+			return func(args []string) error {
+				if len(args) > 0 {
+					return Fail("labels takes no arguments — the set comes from the taxonomy, not the command line")
+				}
+				// The store root is resolved and then deliberately dropped: this
+				// command must fail the same way as every other sync command when
+				// there is no project, and must touch no issue once there is one.
+				_, client, err := syncStart("")
+				if err != nil {
+					return err
+				}
+				if *repo != "" {
+					r, err := wire.ParseRepo(*repo)
+					if err != nil {
+						return Fail("--repo %v", err)
+					}
+					client = client.For(r)
+				}
+
+				specs := mapping.CanonicalLabelSpecs()
+				existing, err := client.ListLabels()
+				if err != nil {
+					return err
+				}
+				rows, similar := labelPlan(specs, existing)
+
+				created, fixed, drifted := 0, 0, 0
+				for _, row := range rows {
+					mark := ""
+					if row.spec.Exclusive {
+						mark = "  exclusive"
+					}
+
+					if row.got == nil {
+						created++
+						if *dryRun {
+							fmt.Printf("create   %-20s %s%s\n", row.spec.Name, row.spec.Color, mark)
+							continue
+						}
+						made, err := client.CreateLabel(row.spec)
+						if err != nil {
+							return err
+						}
+						fmt.Printf("created  %-20s id %-5d %s%s\n", row.spec.Name, made.ID, row.spec.Color, mark)
+						continue
+					}
+					if len(row.drift) == 0 {
+						fmt.Printf("present  %-20s id %d\n", row.spec.Name, row.got.ID)
+						continue
+					}
+
+					drifted++
+					shown := labelShowDrift(row.drift)
+					if !*fix {
+						fmt.Printf("present  %-20s id %-5d drift: %s\n", row.spec.Name, row.got.ID, shown)
+						continue
+					}
+					if *dryRun {
+						fmt.Printf("fix      %-20s id %-5d %s\n", row.spec.Name, row.got.ID, shown)
+						continue
+					}
+					// The unchanged name rides along because a server that reads an
+					// absent field as empty would blank it, and the description is
+					// the repository's own: a description somebody rewrote is
+					// theirs, and this run is about colour and exclusivity.
+					patch := row.spec
+					patch.Description = row.got.Description
+					if _, err := client.EditLabel(row.got.ID, patch); err != nil {
+						return err
+					}
+					fixed++
+					fmt.Printf("fixed    %-20s id %-5d %s\n", row.spec.Name, row.got.ID, shown)
+				}
+
+				for _, s := range similar {
+					fmt.Fprintf(os.Stderr, "warning: %q (id %d) resembles %s — left alone; rename it by hand or ignore it\n",
+						s.name, s.id, strings.Join(s.hits, ", "))
+				}
+
+				verb := "created"
+				if *dryRun {
+					verb = "to create"
+				}
+				line := fmt.Sprintf("%d canonical label(s): %d %s, %d present",
+					len(rows), created, verb, len(rows)-created)
+				if drifted > 0 {
+					line += fmt.Sprintf(" (%d drifted, %d fixed)", drifted, fixed)
+				}
+				if len(similar) > 0 {
+					line += fmt.Sprintf(", %d similar", len(similar))
+				}
+				fmt.Println(line)
+				if drifted > 0 && !*fix {
+					fmt.Println("drift is shown, not applied — re-run with --fix to patch colour/exclusive")
+				}
+				if *dryRun {
+					fmt.Println("dry-run — nothing was written")
+				}
+				return nil
+			}
+		},
+	})
+}
+
+// labelRow is one canonical label, decided before anything is written: what the
+// taxonomy says it should be, what the repository already has under that exact
+// name (nil when it has nothing), and where the two disagree.
+type labelRow struct {
+	spec  wire.LabelRequest
+	got   *wire.Label
+	drift []labelDiff
+}
+
+// labelDiff is one field that disagrees, with both readings, so a receipt can
+// show the change without the caller re-deriving it.
+type labelDiff struct{ field, is, want string }
+
+// labelLookalike is a label of the repository's own that resembles a canonical
+// name. Reported with its id and never touched.
+type labelLookalike struct {
+	name string
+	id   int64
+	hits []string
+}
+
+// labelPlan pairs the canonical set with what the repository holds.
+//
+// In taxonomy order, because a bootstrap prints its plan in that order and a map
+// would shuffle it on every run — two identical runs would look like different
+// ones.
+func labelPlan(specs []wire.LabelRequest, existing []wire.Label) ([]labelRow, []labelLookalike) {
+	byName := make(map[string]*wire.Label, len(existing))
+	for i := range existing {
+		byName[existing[i].Name] = &existing[i]
+	}
+
+	canonical := make(map[string]map[string]bool, len(specs))
+	rows := make([]labelRow, 0, len(specs))
+	for _, spec := range specs {
+		canonical[spec.Name] = labelAkin(spec.Name)
+		row := labelRow{spec: spec, got: byName[spec.Name]}
+		if row.got != nil {
+			row.drift = labelDrift(spec, row.got)
+		}
+		rows = append(rows, row)
+	}
+
+	var similar []labelLookalike
+	for i := range existing {
+		l := &existing[i]
+		if _, exact := canonical[l.Name]; exact {
+			continue
+		}
+		mine := labelAkin(l.Name)
+		var hits []string
+		for _, spec := range specs {
+			if labelIntersects(canonical[spec.Name], mine) {
+				hits = append(hits, spec.Name)
+			}
+		}
+		if len(hits) > 0 {
+			similar = append(similar, labelLookalike{name: l.Name, id: l.ID, hits: hits})
+		}
+	}
+	return rows, similar
+}
+
+// labelDrift is where an existing label disagrees with the spec.
+//
+// Colour and `exclusive` only. A description somebody rewrote is theirs, and the
+// name matched exactly or this row would not exist.
+func labelDrift(spec wire.LabelRequest, got *wire.Label) []labelDiff {
+	var out []labelDiff
+	if labelHex(got.Color) != labelHex(spec.Color) {
+		out = append(out, labelDiff{"color", labelHex(got.Color), labelHex(spec.Color)})
+	}
+	if got.Exclusive != spec.Exclusive {
+		out = append(out, labelDiff{"exclusive",
+			fmt.Sprintf("%t", got.Exclusive), fmt.Sprintf("%t", spec.Exclusive)})
+	}
+	return out
+}
+
+// labelHex normalizes a colour for comparison. Gitea reports them bare
+// (`ee0701`) and the mapping layer writes them with a `#`; same colour, so a
+// comparison has to strip before it compares.
+func labelHex(v string) string { return strings.ToLower(strings.TrimPrefix(strings.TrimSpace(v), "#")) }
+
+var labelWords = regexp.MustCompile(`[^a-z0-9]+`)
+
+// labelAkin is the comparison keys for a label name: its tail, and the whole
+// name squashed.
+//
+// Case, separators and the namespace in front are noise — what a person meant is
+// the tail. `x`, `X` and `kind/x` all reduce to the same tail as `type/x`, and
+// `severity: x y` to the same squashed form as `severity/xy`. Two names resemble
+// each other when these sets intersect.
+func labelAkin(name string) map[string]bool {
+	var parts []string
+	for _, p := range labelWords.Split(strings.ToLower(name), -1) {
+		if p != "" {
+			parts = append(parts, p)
+		}
+	}
+	if len(parts) == 0 {
+		return nil
+	}
+	return map[string]bool{parts[len(parts)-1]: true, strings.Join(parts, ""): true}
+}
+
+func labelIntersects(a, b map[string]bool) bool {
+	for k := range a {
+		if b[k] {
+			return true
+		}
+	}
+	return false
+}
+
+func labelShowDrift(drift []labelDiff) string {
+	var out []string
+	for _, d := range drift {
+		out = append(out, fmt.Sprintf("%s %s -> %s", d.field, d.is, d.want))
+	}
+	return strings.Join(out, ", ")
+}

cli/internal/cmd/new.go

@@ -0,0 +1,144 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+func init() {
+	register(&Command{
+		Name:  "new",
+		Group: GroupIssue,
+		Short: "create a local issue from its type template",
+		Long: `The issue is real the moment this writes the file. Nothing is pending, nothing
+is a draft awaiting a tracker: ` + "`origin: local`" + ` is a complete state and pushing it
+later is optional.
+
+While it says local, this file is the ONLY copy of the work — the store, not a
+cache of anything. That is what a push changes: it hands the issue to the
+tracker and removes the file.
+
+Writes .tea/issues/<slug>.md prefilled with the type's template, prints the
+path, and rebuilds INDEX.md. Fill the sections in an editor, then run
+` + "`kettle check <id>`" + `.
+
+Body prose is Russian, section headers and the title are English.`,
+		Examples: []Example{
+			{`kettle new --type task --title "Wire sqlc into the appclick repo layer" --label tech/sql --label comp/appclick`,
+				"a task with two free-form labels"},
+			{`kettle new --type bug --title "Fix tea-guard crash on empty settings" --depends wire-sqlc-appclick --milestone v0.2`,
+				"a bug that is blocked by another issue"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			typ := fs.String("type", "", "issue type, one of: "+strings.Join(issue.TypeNames(), ", ")+" (becomes the exclusive type/* label)")
+			title := fs.String("title", "", "English, imperative, no type prefix")
+			id := fs.String("id", "", "slug (default: derived from the title)")
+			severity := fs.String("severity", "", "severity/* label, one of: "+strings.Join(issue.Severities, ", "))
+			milestone := fs.String("milestone", "", "milestone title")
+			var labels, assignees, depends stringList
+			fs.Var(&labels, "label", "extra label, e.g. tech/sql; repeat")
+			fs.Var(&assignees, "assignee", "assignee login; repeat")
+			fs.Var(&depends, "depends", "id this issue depends on; repeat")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				if *typ == "" || *title == "" {
+					return Fail("--type and --title are both required")
+				}
+				if !issue.KnownType(*typ) {
+					return Fail("unknown --type %q — known: %s", *typ, strings.Join(issue.TypeNames(), ", "))
+				}
+				if *severity != "" && !issue.KnownSeverity(*severity) {
+					return Fail("unknown --severity %q — known: %s", *severity, strings.Join(issue.Severities, ", "))
+				}
+				// Before anything reads the store path. There is no store to be
+				// second-guessed about when there is no project.
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+
+				labelSet := []string{"type/" + *typ}
+				if *severity != "" {
+					labelSet = append(labelSet, "severity/"+*severity)
+				}
+				for _, l := range labels {
+					if !contains(labelSet, l) {
+						labelSet = append(labelSet, l)
+					}
+				}
+
+				slug := *id
+				if slug == "" {
+					if slug, err = issue.UniqueID(root, issue.Slugify(*title, 0), nil); err != nil {
+						return err
+					}
+				} else if !issue.IsSlug(slug) {
+					return Fail("--id %q is not a slug (lowercase, digits, single dashes)", slug)
+				}
+				if _, err := os.Stat(issue.PathOf(root, slug)); err == nil {
+					return Fail("%s already exists", issue.PathOf(root, slug))
+				}
+
+				known := map[string]bool{}
+				for _, k := range issue.AllIDs(root) {
+					known[k] = true
+				}
+				for _, d := range depends {
+					if !known[d] {
+						fmt.Fprintf(os.Stderr, "warning: depends on %q, which is not in the store yet\n", d)
+					}
+				}
+
+				i := &issue.Issue{
+					ID:        slug,
+					Title:     *title,
+					Body:      issue.Template(*typ, depends),
+					State:     "open",
+					Labels:    labelSet,
+					Assignees: assignees,
+					Milestone: *milestone,
+					Depends:   depends,
+					Origin:    issue.Local,
+				}
+
+				// The first issue in a fresh checkout has to create the store,
+				// but it says so — and it says where, because the path is
+				// absolute.
+				created, err := issue.CreateStore(root)
+				if err != nil {
+					return err
+				}
+				if created {
+					abs, _ := filepath.Abs(root)
+					fmt.Fprintf(os.Stderr, "created store %s\n", abs)
+				}
+
+				path, err := issue.Save(root, i)
+				if err != nil {
+					return err
+				}
+				if _, _, err := issue.BuildIndex(root); err != nil {
+					return err
+				}
+				fmt.Printf("%s  [type/%s] %s\n", path, *typ, *title)
+				fmt.Printf("fill the sections, then: kettle check %s\n", slug)
+				return nil
+			}
+		},
+	})
+}
+
+func contains(xs []string, x string) bool {
+	for _, v := range xs {
+		if v == x {
+			return true
+		}
+	}
+	return false
+}

cli/internal/cmd/pull.go

@@ -0,0 +1,594 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"time"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "pull",
+		Group: GroupSync,
+		Args:  "[<key>…]",
+		Short: "fetch issues from the tracker into the local store",
+		Long: `THIS IS HOW A PUSHED ISSUE COMES BACK. ` + "`kettle push`" + ` deletes the local file the
+moment the tracker confirms the write, so a pull is not a refresh of a copy you
+kept — it is how the copy comes to exist at all.
+
+It lands under the SAME slug it had before, after a rename in the web UI and on
+a machine that has never seen the issue. Three sources answer "what is this
+issue called here", in this order:
+
+    .remote.json        the number -> slug ledger; the only one that knows what
+                        is on disk right now, so it wins
+    <!-- kettle:id … -->  the marker in the tracker-side body; it survives a lost
+                        ledger, a fresh clone, another machine, and a retitling
+    the title           slugified — where an issue filed in the web UI gets its
+                        first local name
+
+A marker is taken at its word only when the slug is free; a name already in use
+is a collision, not an identity, and is uniquified rather than allowed to
+overwrite somebody else's issue. The marker itself is stripped out of what lands
+on disk.
+
+TWO WAYS TO NAME WHAT TO PULL, and they are not the same operation:
+
+    kettle pull 42 #43 owner/repo#44     by key — an ADDRESS
+    kettle pull --milestone v0.2         by filter — a QUERY
+
+A key fetches an issue in ANY state, because a number is an address and not a
+question about state. Only filter mode leaves closed issues out — a closed issue
+is not a unit of work — and only ` + "`--state closed`" + ` puts one in the store. An issue
+already on disk is refreshed either way, so a local copy learns it was closed
+instead of staying open forever, and the count that stayed out goes to stderr.
+
+` + "`--limit`" + ` IS ON THE WRITE, NOT ON THE SELECTION. It counts the issues this run puts
+in the store and never the closed ones it enumerated and threw away, so pages
+keep coming until the budget is full — and stop the moment it is. A filter that
+matches almost only closed issues ends in a warning and a short answer rather
+than a walk of the whole tracker.
+
+A PULL RETURNS THE UNIT OF WORK, NOT ONE ROW OF IT. ` + "`depends:`" + ` is filled from the
+tracker's own dependency graph and every blocker comes down with it, recursively,
+to --depth. What that costs, stated rather than hidden: one request per issue
+that lands in the store, plus one per blocker the selection did not already
+carry. ` + "`--no-deps`" + ` is the way back to one request, and narrows the answer to the
+one issue you asked for. Dependencies are outside --limit: a blocker is followed
+because a stored issue named it, not because the filter selected it, so a
+filtered pull can leave more files behind than its limit — including one from
+another milestone. The one blocker that does not land is a closed one.
+
+PULLING OVERWRITES THE BODY: a fetch, not a merge. Local edits you have not
+pushed are lost, with exactly one exception — checkbox state. A tick is monotone,
+so a ` + "`[x]`" + ` on either side wins for any item whose text matches; unticking is not,
+so untick locally and push. ` + "`--cached`" + ` skips an issue before any of that.
+
+Comments ride along: the thread lands beside the issue in <id>.comments.md. It
+costs no request when the payload says there are none, and a file left over from
+an earlier pull is deleted — so no file means "no comments", never "not asked
+for". The thread is pull-only; post with ` + "`kettle comment`" + `.`,
+		Examples: []Example{
+			{"kettle pull 42", "the issue and everything blocking it, in any state"},
+			{"kettle pull 42 --no-deps", "just that one issue — one request"},
+			{"kettle pull owner/repo#42", "an issue in another repository"},
+			{"kettle pull --milestone v0.2 --limit 20", "20 open issues from a milestone, blockers included"},
+			{"kettle pull --label type/bug --state all", "every bug; the closed ones are enumerated, not stored"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			milestone := fs.String("milestone", "", "pull a whole milestone (id or title)")
+			var labels stringList
+			fs.Var(&labels, "label", "filter by label; repeat for AND")
+			// Both spellings, because both are what somebody has in hand: `-q` is
+			// what a person types and `--query` is what a script reads back.
+			var query string
+			fs.StringVar(&query, "q", "", "search text in title and body")
+			fs.StringVar(&query, "query", "", "the long spelling of -q")
+			state := fs.String("state", "open", "filter mode only: open, closed or all")
+			limit := fs.Int("limit", 100, "filter mode: how many issues to STORE, not to enumerate")
+			noDeps := fs.Bool("no-deps", false, "do not fill depends: and do not follow blockers")
+			depth := fs.Int("depth", 3, "how deep to follow blockers")
+			cached := fs.Bool("cached", false, "skip issues already on disk instead of refetching")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				filtered := *milestone != "" || len(labels) > 0 || query != ""
+				switch {
+				case len(args) > 0 && filtered:
+					return Fail("pass issue keys OR filters, not both")
+				case len(args) == 0 && !filtered:
+					return Fail("nothing to pull: pass an issue key, or --milestone / --label / -q")
+				case !contains([]string{"open", "closed", "all"}, *state):
+					return Fail("--state %q must be open, closed or all", *state)
+				case *limit < 1:
+					return Fail("--limit must be 1 or more, got %d", *limit)
+				case *depth < 0:
+					return Fail("--depth must be 0 or more, got %d", *depth)
+				}
+
+				// Keys are parsed before anything is opened: a typo in a key is
+				// not a network problem and must not be reported as one.
+				keys, named, err := pullKeys(args)
+				if err != nil {
+					return err
+				}
+
+				root, client, err := syncStart(*out)
+				if err != nil {
+					return err
+				}
+				// A key may name its own repository; the project's is the
+				// fallback, never an override.
+				if !named.Zero() {
+					client = client.For(named)
+				}
+				repo := client.Repo()
+
+				// A first pull into a fresh checkout has to create the store, and
+				// it says so — with an absolute path, so it cannot be a stray
+				// working directory.
+				created, err := issue.CreateStore(root)
+				if err != nil {
+					return err
+				}
+				if created {
+					abs, _ := filepath.Abs(root)
+					fmt.Fprintf(os.Stderr, "created store %s\n", abs)
+				}
+
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+				ledger := loadLedgerOrFold(root, issues)
+				namer := &pullNamer{root: root, repo: repo, ledger: ledger, taken: map[string]bool{}}
+				for id := range issues {
+					namer.taken[id] = true
+				}
+				// What the ledger already knows, so a `#N` in a body resolves to
+				// a slug this run never fetched.
+				numberOf := map[int]string{}
+				for raw, slug := range ledger {
+					if k, err := wire.ParseKey(raw); err == nil && k.Repo == repo {
+						numberOf[k.Number] = slug
+					}
+				}
+
+				// A closed issue is not a unit of work: filter mode enumerates it
+				// and keeps it out of the store unless the operator named the
+				// state. A key is an address, not a bulk read, so key mode is
+				// exempt.
+				dropClosed := filtered && *state != "closed"
+
+				queue, err := pullSeed(client, keys, filtered, gitea.IssueFilter{
+					State: *state, Labels: labels, Query: query, Milestone: *milestone,
+					Limit: *limit,
+					Keep: func(p *wire.Issue) bool {
+						return pullLandsInStore(p, dropClosed, namer)
+					},
+				})
+				if err != nil {
+					return err
+				}
+
+				var written, skipped []string
+				var dropped []int
+				type unresolved struct {
+					id      string
+					numbers []int
+				}
+				var pending []unresolved
+
+				seen := map[int]bool{}
+				for _, t := range queue {
+					seen[t.payload.Number] = true
+				}
+
+				for len(queue) > 0 {
+					task := queue[0]
+					queue = queue[1:]
+					p := task.payload
+					id, err := namer.idFor(p)
+					if err != nil {
+						return err
+					}
+					stored := pullStored(root, id)
+
+					// Closed and not already ours: nothing is written and nothing
+					// is asked of the server for it — not its comments, not its
+					// links, and its blockers are not followed. The slug stays
+					// unclaimed too, so no other issue ends up pointing
+					// `depends:` at a file that is not there.
+					if dropClosed && p.State == "closed" && !stored {
+						dropped = append(dropped, p.Number)
+						continue
+					}
+
+					namer.taken[id] = true
+					numberOf[p.Number] = id
+
+					// The native links, fetched ONCE for the two things they are
+					// for: filling this issue's `depends:` and telling the walk
+					// where to go next. One request per issue that lands in the
+					// store, and only one.
+					var blockers []int
+					if !*noDeps {
+						if blockers, err = pullBlockers(client, p.Number, repo); err != nil {
+							return err
+						}
+					}
+
+					if *cached && stored {
+						skipped = append(skipped, id) // body and thread unread; only the links cost
+					} else {
+						// The copy already on disk, as it was when this run
+						// started. It contributes its ticked checkboxes and
+						// nothing else.
+						local := ""
+						if prev := issues[id]; prev != nil {
+							local = prev.Body
+						}
+						next, missing := mapping.FromPayload(p, id, repo, mapping.PayloadOptions{
+							IDForNumber:  numberOf,
+							ExtraNumbers: blockers,
+							// The clock is the caller's: mapping is a pure layer
+							// and a package with a clock in it is not one.
+							Synced:    time.Now().UTC().Format(time.RFC3339),
+							LocalBody: local,
+						})
+						if _, err := issue.Save(root, next); err != nil {
+							return err
+						}
+						if _, err := pullSyncComments(client, root, id, p.Number, p.Comments); err != nil {
+							return err
+						}
+						ledger.Set(wire.Key{Repo: repo, Number: p.Number}, id)
+						written = append(written, id)
+						pending = append(pending, unresolved{id, missing})
+					}
+
+					if *noDeps || task.depth >= *depth {
+						continue
+					}
+					for _, n := range append(mapping.NumbersInBody(p.Body), blockers...) {
+						if seen[n] {
+							continue
+						}
+						seen[n] = true
+						child, err := client.GetIssue(n)
+						if err != nil {
+							return err
+						}
+						queue = append(queue, pullTask{payload: child, depth: task.depth + 1})
+					}
+				}
+
+				// Nothing is dropped in silence.
+				if len(dropped) > 0 {
+					fmt.Fprintf(os.Stderr, "%d closed issue(s) enumerated, not stored"+
+						" (--state closed to pull them)\n", len(dropped))
+				}
+
+				// Second pass: a `#N` that named an issue this run had not written
+				// yet. The first write could not resolve it to a slug; by now the
+				// file it names is on disk.
+				for _, u := range pending {
+					var newly []string
+					for _, n := range u.numbers {
+						if slug := numberOf[n]; slug != "" && slug != u.id {
+							newly = append(newly, slug)
+						}
+					}
+					if len(newly) == 0 {
+						continue
+					}
+					i, err := issue.Load(root, u.id)
+					if err != nil {
+						return err
+					}
+					for _, slug := range newly {
+						if !contains(i.Depends, slug) {
+							i.Depends = append(i.Depends, slug)
+						}
+					}
+					if _, err := issue.Save(root, i); err != nil {
+						return err
+					}
+				}
+
+				if err := ledger.Save(root); err != nil {
+					return err
+				}
+				indexPath, _, err := issue.BuildIndex(root)
+				if err != nil {
+					return err
+				}
+				return pullReceipt(root, written, skipped, indexPath)
+			}
+		},
+	})
+}
+
+// pullTask is one issue to walk, and how far from a seed it was found.
+type pullTask struct {
+	payload *wire.Issue
+	depth   int
+}
+
+// pullKeys parses the positional arguments and the one repository they may name.
+//
+// All of them or none: a run addresses one repository, because the client, the
+// ledger keys and the `gitea:` field all have to agree about which one.
+func pullKeys(args []string) ([]wire.Key, wire.Repo, error) {
+	var keys []wire.Key
+	var named wire.Repo
+	for _, a := range args {
+		k, err := wire.ParseKey(a)
+		if err != nil {
+			return nil, wire.Repo{}, err
+		}
+		if !k.Repo.Zero() {
+			if !named.Zero() && named != k.Repo {
+				return nil, wire.Repo{}, Fail("all keys must name one repository, got %s and %s", named, k.Repo)
+			}
+			named = k.Repo
+		}
+		keys = append(keys, k)
+	}
+	return keys, named, nil
+}
+
+// pullSeed is what the walk starts from: the issues a key addresses, or the ones
+// a filter selected.
+func pullSeed(c *gitea.Client, keys []wire.Key, filtered bool, f gitea.IssueFilter) ([]pullTask, error) {
+	if !filtered {
+		out := make([]pullTask, 0, len(keys))
+		for _, k := range keys {
+			p, err := c.GetIssue(k.Number)
+			if err != nil {
+				return nil, err
+			}
+			out = append(out, pullTask{payload: p})
+		}
+		return out, nil
+	}
+
+	listing, err := c.ListIssues(f)
+	if err != nil {
+		return nil, err
+	}
+	if len(listing.Issues) == 0 {
+		return nil, Fail("no issues match that filter")
+	}
+	if listing.Warning != "" {
+		fmt.Fprintf(os.Stderr, "warning: %s\n", listing.Warning)
+	}
+
+	var what []string
+	if listing.Milestone != "" {
+		what = append(what, "milestone "+listing.Milestone)
+	}
+	for _, l := range f.Labels {
+		what = append(what, "label "+l)
+	}
+	if f.Query != "" {
+		what = append(what, fmt.Sprintf("q=%q", f.Query))
+	}
+	fmt.Fprintf(os.Stderr, "%d issue(s) match %s (%s)\n",
+		len(listing.Issues), strings.Join(what, " + "), f.State)
+
+	out := make([]pullTask, 0, len(listing.Issues))
+	for i := range listing.Issues {
+		out = append(out, pullTask{payload: &listing.Issues[i]})
+	}
+	return out, nil
+}
+
+// pullLandsInStore is the --limit predicate: would this payload leave a file in
+// the store?
+//
+// It has to be the same test the walk applies, or the budget is spent on issues
+// that never land — which is the bug it exists to prevent. A closed issue counts
+// only when the store already has it (it is refreshed, and that is a write);
+// anything else counts, including one --cached will skip, because a skipped
+// issue is still an issue the store holds when the run ends.
+func pullLandsInStore(p *wire.Issue, dropClosed bool, namer *pullNamer) bool {
+	if !dropClosed || p.State != "closed" {
+		return true
+	}
+	id, err := namer.idFor(p)
+	if err != nil {
+		// An id that cannot be allocated is the walk's failure to report, not a
+		// reason to spend the page budget differently.
+		return true
+	}
+	return pullStored(namer.root, id)
+}
+
+// pullBlockers is the numbers of the issues that block this one, in this
+// repository.
+//
+// A blocker in ANOTHER repository is dropped here, and deliberately: everything
+// downstream — `depends:`, the number -> slug ledger, the walk's own GETs — reads
+// a bare number against the repository being pulled, so a foreign number would
+// either resolve to the wrong issue or invent an edge. The body still names it,
+// so nothing is lost.
+func pullBlockers(c *gitea.Client, number int, repo wire.Repo) ([]int, error) {
+	deps, err := c.Dependencies(number)
+	if err != nil {
+		return nil, err
+	}
+	var out []int
+	for i := range deps {
+		if k := deps[i].KeyIn(repo); k.Repo == repo {
+			out = append(out, k.Number)
+		}
+	}
+	return out, nil
+}
+
+// pullSyncComments brings <id>.comments.md in line with the tracker and returns
+// its path, or "" when the issue has no thread.
+//
+// count is the payload's own comment count, so an issue with none costs no
+// request. A file from an earlier pull is removed when the thread is empty: the
+// absence of the file is the answer, not a gap in what was asked for.
+func pullSyncComments(c *gitea.Client, root, id string, number, count int) (string, error) {
+	path := commentsSidecarPath(root, id)
+	var thread []wire.Comment
+	if count > 0 {
+		var err error
+		if thread, err = c.ListComments(number); err != nil {
+			return "", err
+		}
+	}
+	if len(thread) > 0 {
+		if err := os.WriteFile(path, []byte(mapping.RenderComments(thread)), 0o644); err != nil {
+			return "", err
+		}
+		return path, nil
+	}
+	if err := os.Remove(path); err != nil && !os.IsNotExist(err) {
+		return "", err
+	}
+	return "", nil
+}
+
+// pullReceipt is the only thing that lands in a reader's context: one line per
+// issue, the raw payload nowhere.
+func pullReceipt(root string, written, skipped []string, indexPath string) error {
+	cached := map[string]bool{}
+	all := map[string]bool{}
+	for _, id := range written {
+		all[id] = true
+	}
+	for _, id := range skipped {
+		cached[id], all[id] = true, true
+	}
+	ids := make([]string, 0, len(all))
+	for id := range all {
+		ids = append(ids, id)
+	}
+	sort.Strings(ids)
+
+	graph := false
+	for _, id := range ids {
+		i, err := issue.Load(root, id)
+		if err != nil {
+			return err
+		}
+		graph = graph || len(i.Depends) > 0
+		note := ""
+		if cached[id] {
+			note = "  (cached)"
+		}
+		if path := commentsSidecarPath(root, id); pullIsFile(path) {
+			n := i.Extra[mapping.CommentsKey]
+			if n == "" {
+				n = "?"
+			}
+			note += fmt.Sprintf("  +%s comments: %s", n, path)
+		}
+		labels := strings.Join(i.Labels, ", ")
+		if labels == "" {
+			labels = "no labels"
+		}
+		fmt.Printf("%s [%s] %s — %s    %s%s\n",
+			id, labels, i.Title, i.State, issue.PathOf(root, id), note)
+	}
+	fmt.Printf("index: %s\n", indexPath)
+	// Worth printing when there is something to draw, not on every run that
+	// could have drawn something.
+	if graph {
+		fmt.Println("graph: run `kettle tree` (offline) to draw it")
+	}
+	return nil
+}
+
+// --------------------------------------------------------------------------
+// naming, and the files the sync layer parks beside an issue
+// --------------------------------------------------------------------------
+
+// pullNamer answers "what is this remote issue called here", and remembers what
+// it has already handed out so one run cannot name two issues the same thing.
+type pullNamer struct {
+	root   string
+	repo   wire.Repo
+	ledger gitea.RemoteMap
+	// taken is every slug the store holds plus every one this run has claimed.
+	taken map[string]bool
+}
+
+// idFor is the slug this remote issue belongs under. Three sources, in order —
+// see the command's own documentation for why that order and not another.
+func (n *pullNamer) idFor(p *wire.Issue) (string, error) {
+	if got := n.ledger.Slug(wire.Key{Repo: n.repo, Number: p.Number}); got != "" {
+		return got, nil
+	}
+	marked := mapping.IDInBody(p.Body)
+	// A marker is an identity only while the name is free. A file of that name
+	// already in the store, or a ledger entry holding it under another number,
+	// makes it a collision — and overwriting somebody else's issue is worse than
+	// allocating a suffix.
+	if marked != "" && !n.taken[marked] && !n.ledgerHolds(marked) {
+		return marked, nil
+	}
+	base := marked
+	if base == "" {
+		base = issue.Slugify(p.Title, 0)
+	}
+	taken := make([]string, 0, len(n.taken))
+	for id := range n.taken {
+		taken = append(taken, id)
+	}
+	return issue.UniqueID(n.root, base, taken)
+}
+
+func (n *pullNamer) ledgerHolds(slug string) bool {
+	for _, s := range n.ledger {
+		if s == slug {
+			return true
+		}
+	}
+	return false
+}
+
+// loadLedgerOrFold is the number -> slug ledger, with the `gitea:` fields still
+// on disk folded in when there is no ledger to read.
+//
+// A MERGE and never a replacement, which is why the fold only happens when the
+// file is missing or empty: push deletes the file it has just sent, so the store
+// is a SUBSET of what the ledger knows and a rebuild from the files alone would
+// throw away every entry it cannot see. What a fold cannot recover — a
+// pushed-and-dropped issue whose entry was also lost — is not lost either: the
+// next pull of that number reads the slug off the marker in the body and writes
+// the entry back.
+func loadLedgerOrFold(root string, issues map[string]*issue.Issue) gitea.RemoteMap {
+	m := gitea.LoadRemoteMap(root)
+	if len(m) > 0 {
+		return m
+	}
+	for id, i := range issues {
+		if k, ok := mapping.RemoteKeyOf(i); ok {
+			m.Set(k, id)
+		}
+	}
+	return m
+}
+
+// pullStored reports whether the store already holds this issue.
+func pullStored(root, id string) bool { return pullIsFile(issue.PathOf(root, id)) }
+
+func pullIsFile(path string) bool {
+	fi, err := os.Stat(path)
+	return err == nil && !fi.IsDir()
+}

cli/internal/cmd/push.go

@@ -0,0 +1,652 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"os/exec"
+	"sort"
+	"strings"
+	"time"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/mapping"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func init() {
+	register(&Command{
+		Name:  "push",
+		Group: GroupSync,
+		Args:  "[<id>…]",
+		Short: "send local issues to the tracker; the local copy goes with them",
+		Long: `A SUCCESSFUL PUSH DELETES THE LOCAL FILE — <id>.md and every sidecar under that
+slug — and prints the number and the URL the issue now lives at. Once the tracker
+has the issue, the tracker IS the issue: what is left in the store is what has
+not left this machine. Get it back with ` + "`kettle pull <n>`" + `, which returns it under
+the same slug, because the slug travelled up in the body as <!-- kettle:id … -->
+and was recorded in the number -> slug ledger.
+
+ONE RULE, NO EXCEPTION: --update deletes as well. A PATCH is a push, and an issue
+that has just been sent is no more local than one that was just created. Two
+rules would put back exactly the question this removes — "is my copy the fresh
+one?".
+
+THE DELETION IS THE LAST THING THAT HAPPENS TO AN ISSUE, and only after all
+three of:
+
+    1. the call came back without an error and with a 2xx,
+    2. the answer carries a plausible number — on --update the very number that
+       was PATCHed, and
+    3. the ledger has been written with number -> slug.
+
+Network down, non-2xx, an answer that does not confirm the write: the file stays
+and the run stops. Nothing removes a file it has not just watched the tracker
+accept, and nothing removes a file for an issue it did not send — ` + "`origin: local`" + `
+work that has never been pushed is never touched by any of this. Get the ordering
+wrong and a slug is lost at exactly the moment the local copy stops being the
+record, which is why the ledger is written before anything is deleted and not
+after.
+
+Every issue is validated against the canonical format first, offline and before
+a socket is opened. --force posts anyway; say why when you use it.
+
+DEPENDENCIES GO FIRST, in topological order, so a blocker has its number before
+the issue that names it. Every ` + "`depends:`" + ` entry that has a number becomes a NATIVE
+tracker link — the same /dependencies a pull reads back, so the tracker shows the
+blocking panel and refuses to close a blocked issue first. A link that is already
+there is skipped, not re-POSTed, which is what makes a repeat push a no-op. A
+dependency that is still local-only has no number and becomes no link: it is
+reported, never silently dropped.
+
+REMOVING a link is out of scope — push only ever adds. A dependency deleted from
+` + "`depends:`" + ` leaves its tracker link standing; unlink it in the web UI.
+
+The ` + "`## Depends on`" + ` prose is never touched: slugs stay slugs and are not rewritten
+to #N, so a pull -> push round trip is byte for byte.
+
+Labels the repository is missing are created with the canonical colour and, for
+type/* and severity/*, exclusive: true. ` + "`branch:`" + ` carries the tracker's ` + "`ref`" + `: an
+empty one is filled with the current git branch and an already-set one is sent as
+written. Detached HEAD or no repository at all is not an error — no ref is sent
+and a warning says so.`,
+		Examples: []Example{
+			{"kettle push", "every issue the tracker does not have yet, blockers first"},
+			{"kettle push wire-sqlc-appclick", "one issue"},
+			{"kettle push --update wire-sqlc-appclick", "PATCH one that is already there — the file still goes"},
+			{"kettle push --dry-run", "validate and print the plan; no network, nothing deleted"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			update := fs.Bool("update", false, "PATCH issues that already carry a gitea: field")
+			dryRun := fs.Bool("dry-run", false, "validate and print the plan; no network, nothing deleted")
+			force := fs.Bool("force", false, "push despite format violations")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				// A dry run touches no network, so it must not need a credential
+				// to say what it would do. The real run goes through
+				// syncStartExisting below, which resolves the store before it
+				// builds a client — a missing store reported as a network
+				// problem sends the operator to the wrong place.
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				if err := issue.StoreError(root); err != nil {
+					return Fail("%s — create an issue with `kettle new` first", err)
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+
+				chosen, err := pushSelect(issues, args, *update)
+				if err != nil {
+					return err
+				}
+
+				// The domain's own check, offline, before anything is sent.
+				known := map[string]bool{}
+				for id := range issues {
+					known[id] = true
+				}
+				blocked := false
+				for _, id := range chosen {
+					errs, warns := issue.Validate(issues[id], known)
+					for _, w := range warns {
+						fmt.Fprintf(os.Stderr, "warning: %s: %s\n", id, w)
+					}
+					for _, e := range errs {
+						fmt.Fprintf(os.Stderr, "%s: %s\n", id, e)
+					}
+					blocked = blocked || len(errs) > 0
+				}
+				if blocked && !*force {
+					return Fail("format violations (see above); --force overrides")
+				}
+
+				// Dependencies first, so a blocker has its number by the time the
+				// issue that names it is sent. A cycle is reported and ordered
+				// around rather than refused: it is a data problem, not a reason
+				// to send nothing.
+				edges := map[string][]string{}
+				for _, id := range chosen {
+					var deps []string
+					for _, d := range issues[id].Depends {
+						if _, ok := issues[d]; ok {
+							deps = append(deps, d)
+						}
+					}
+					edges[id] = deps
+				}
+				pushing := map[string]bool{}
+				for _, id := range chosen {
+					pushing[id] = true
+				}
+				var order []string
+				for _, id := range issue.TopoOrder(chosen, edges) {
+					if pushing[id] {
+						order = append(order, id)
+					}
+				}
+				for _, c := range issue.FindCycles(edges) {
+					fmt.Fprintf(os.Stderr, "warning: dependency cycle: %s\n", strings.Join(c, " -> "))
+				}
+
+				// Only an EMPTY branch: one written by hand is the author's
+				// decision and a push does not argue with it. The value is set on
+				// the in-memory issue only — the file it came from is about to be
+				// deleted, and the branch comes back with the next pull.
+				var blank []string
+				for _, id := range order {
+					if strings.TrimSpace(issues[id].Extra[mapping.BranchKey]) == "" {
+						blank = append(blank, id)
+					}
+				}
+				if len(blank) > 0 {
+					if branch := pushGitBranch(); branch != "" {
+						for _, id := range blank {
+							if issues[id].Extra == nil {
+								issues[id].Extra = map[string]string{}
+							}
+							issues[id].Extra[mapping.BranchKey] = branch
+						}
+					} else {
+						fmt.Fprintf(os.Stderr, "warning: no current git branch (detached HEAD, or "+
+							"outside a git repository) — no `ref` on: %s\n", strings.Join(blank, ", "))
+					}
+				}
+
+				if *dryRun {
+					// Not one request is made here: everything below is read off
+					// the store and the ledger, which costs nothing.
+					pushPlan(root, issues, order, pushing, *update)
+					return nil
+				}
+
+				_, client, err := syncStartExisting(*out)
+				if err != nil {
+					return err
+				}
+				repo := client.Repo()
+
+				var wanted []string
+				for _, id := range order {
+					for _, l := range issues[id].Labels {
+						if !contains(wanted, l) {
+							wanted = append(wanted, l)
+						}
+					}
+				}
+				sort.Strings(wanted)
+				labelIDs, err := pushLabelIDs(client, wanted)
+				if err != nil {
+					return err
+				}
+
+				milestones := map[string]*int64{}
+				ledger := loadLedgerOrFold(root, issues)
+				keyOf := pushLedgerKeys(ledger, repo)
+
+				for _, id := range order {
+					i := issues[id]
+
+					// Local-only means "this machine has never sent it": no
+					// `gitea:` on the file AND no entry in the ledger. A blocker
+					// whose file an earlier push already dropped is in the ledger
+					// and is not one of these.
+					var unsynced []string
+					for _, d := range i.Depends {
+						dep, onDisk := issues[d]
+						if !onDisk || pushing[d] {
+							continue
+						}
+						if _, synced := mapping.RemoteKeyOf(dep); synced {
+							continue
+						}
+						if _, inLedger := keyOf[d]; !inLedger {
+							unsynced = append(unsynced, d)
+						}
+					}
+					if len(unsynced) > 0 {
+						fmt.Fprintf(os.Stderr, "warning: %s: depends on local-only issue(s) %s"+
+							" — no cross-link in the tracker\n", id, strings.Join(unsynced, ", "))
+					}
+
+					msID, ok := milestones[i.Milestone]
+					if i.Milestone != "" && !ok {
+						m, err := client.FindMilestone(i.Milestone)
+						if err != nil {
+							return err
+						}
+						if m != nil {
+							msID = wire.Set(m.ID)
+						}
+						milestones[i.Milestone] = msID
+					}
+					if i.Milestone != "" && msID == nil {
+						fmt.Fprintf(os.Stderr, "warning: %s: milestone %q does not exist in %s"+
+							" — not set\n", id, i.Milestone, repo)
+					}
+
+					opt := mapping.RequestOptions{LabelIDs: labelIDs, MilestoneID: msID}
+					sent, synced := mapping.NumberOf(i)
+					var got *wire.Issue
+					verb := "created"
+					if synced {
+						// An edit says what state it means; a create takes the
+						// tracker's default.
+						opt.IncludeState = true
+						verb = "updated"
+						got, err = client.EditIssue(sent, *mapping.ToRequest(i, opt), "issue-"+id)
+					} else {
+						sent = 0
+						got, err = client.CreateIssue(*mapping.ToRequest(i, opt), "issue-"+id)
+					}
+					// THE GATE. Below this line a local file is going to be
+					// deleted, so anything short of a confirmed write stops the
+					// run right here.
+					if err != nil {
+						return Fail("%s: not %s: %v — %s is untouched", id, verb, err, issue.PathOf(root, id))
+					}
+					number, confirmed := pushConfirmedNumber(got, sent)
+					if !confirmed {
+						return Fail("%s: the tracker's answer does not confirm the write "+
+							"(it carries number %d) — %s is untouched, nothing was deleted",
+							id, got.Number, issue.PathOf(root, id))
+					}
+
+					// The number is confirmed, so the ledger learns it NOW —
+					// before the label fix-up and the links, both of which can
+					// still fail, and well before the file is removed. This entry
+					// is what a later `kettle pull <n>` lands on; an interrupted
+					// run must cost a re-pull, never a slug.
+					key := wire.Key{Repo: repo, Number: number}
+					ledger.Set(key, id)
+					keyOf[id] = key
+					if err := ledger.Save(root); err != nil {
+						return Fail("%s: the tracker has it as #%d but the ledger could not be "+
+							"written (%v) — %s is untouched", id, number, err, issue.PathOf(root, id))
+					}
+
+					// Gitea occasionally drops labels handed to it on create, so
+					// the echo is checked and the set re-applied rather than
+					// trusted. A failure here is a warning and not an abort: the
+					// issue IS in the tracker, and a run that stopped now would
+					// leave a file whose `gitea:` field was never written — which
+					// the next push would file all over again as a new issue.
+					applied := map[string]bool{}
+					for _, l := range got.Labels {
+						applied[l.Name] = true
+					}
+					var ids []int64
+					var missing []string
+					for _, name := range i.Labels {
+						lid, in := labelIDs[name]
+						if !in {
+							continue
+						}
+						ids = append(ids, lid)
+						if !applied[name] {
+							missing = append(missing, name)
+						}
+					}
+					if len(missing) > 0 {
+						if _, err := client.SetLabels(number, ids, "labels-"+id); err != nil {
+							fmt.Fprintf(os.Stderr, "warning: %s: could not re-apply labels (%s): %v\n",
+								id, strings.Join(missing, ", "), err)
+						} else {
+							fmt.Fprintf(os.Stderr, "warning: %s: labels re-applied via PUT (%s)\n",
+								id, strings.Join(missing, ", "))
+						}
+					}
+
+					// The in-memory issue is stamped even though its file is
+					// going: the rest of this loop reads `gitea:` off it to link
+					// dependencies, and a later issue in topological order asks
+					// the same of this one.
+					mapping.ApplyRemote(i, got, repo, time.Now().UTC().Format(time.RFC3339))
+
+					// The number and the URL lead, because in a moment the local
+					// path is gone and this is the only address the issue has.
+					fmt.Printf("%s  %s  #%d  %s\n", verb, id, number, got.HTMLURL)
+
+					if err := pushLinks(client, id, number, i, issues, pushing, keyOf, repo); err != nil {
+						return err
+					}
+
+					// And now the local copy goes: the last thing that happens to
+					// this issue, after the write, the ledger and the links. A
+					// warning above lands here anyway — the issue is in the
+					// tracker, and keeping a stale file beside it would put back
+					// exactly the two-copies question this removes.
+					gone, err := issue.Remove(root, id)
+					for _, p := range gone {
+						fmt.Printf("    dropped  %s\n", p)
+					}
+					if err != nil {
+						return Fail("%s: the tracker has it as #%d, but the local copy could not "+
+							"be removed: %v", id, number, err)
+					}
+					fmt.Printf("    kettle pull %d to work on it again\n", number)
+				}
+
+				if err := ledger.Save(root); err != nil {
+					return err
+				}
+				path, n, err := issue.BuildIndex(root)
+				if err != nil {
+					return err
+				}
+				fmt.Printf("index: %s — %d issue(s)\n", path, n)
+				return nil
+			}
+		},
+	})
+}
+
+// pushSelect is which issues to send, and the refusal of the ambiguous
+// combinations.
+//
+// Named ids are taken as typed. With none, the default is everything this
+// machine has never sent — pushing the whole store on a bare `kettle push` would
+// re-PATCH every working copy in it.
+func pushSelect(issues map[string]*issue.Issue, ids []string, update bool) ([]string, error) {
+	var chosen []string
+	if len(ids) > 0 {
+		var missing []string
+		for _, id := range ids {
+			if _, ok := issues[id]; !ok {
+				missing = append(missing, id)
+			}
+		}
+		if len(missing) > 0 {
+			return nil, Fail("no such issue(s) in the store: %s", strings.Join(missing, ", "))
+		}
+		chosen = append(chosen, ids...)
+	} else {
+		for id, i := range issues {
+			if _, synced := mapping.RemoteKeyOf(i); update || !synced {
+				chosen = append(chosen, id)
+			}
+		}
+		sort.Strings(chosen)
+		if len(chosen) == 0 {
+			return nil, Fail("nothing to push: every issue in the store is already in the " +
+				"tracker — pass --update to PATCH them, or `kettle new` to make one")
+		}
+	}
+	if !update {
+		var already []string
+		for _, id := range chosen {
+			if _, synced := mapping.RemoteKeyOf(issues[id]); synced {
+				already = append(already, id)
+			}
+		}
+		if len(already) > 0 {
+			return nil, Fail("already in the tracker: %s — pass --update to PATCH them",
+				strings.Join(already, ", "))
+		}
+	}
+	return chosen, nil
+}
+
+// pushLabelIDs is name -> id for the labels these issues carry, creating what
+// the repository is missing.
+//
+// Decided against the repository as it is right now, in one request, and never
+// against a cache: a cache answers "what did we create last time", and the
+// question here is "what does this repository have". A label the tracker does
+// not have and this cannot create is the one failure worth stopping for — an
+// issue filed without its `type/*` label is an issue nothing can find again.
+func pushLabelIDs(c *gitea.Client, names []string) (map[string]int64, error) {
+	out := map[string]int64{}
+	if len(names) == 0 {
+		return out, nil
+	}
+	have, err := c.ListLabels()
+	if err != nil {
+		return nil, err
+	}
+	known := make(map[string]int64, len(have))
+	for _, l := range have {
+		known[l.Name] = l.ID
+	}
+	// The spec — colour, description, exclusivity — is the bridge's, read off the
+	// domain's taxonomy. This layer only decides which names are wanted.
+	for _, spec := range mapping.LabelSpecs(names) {
+		if id, ok := known[spec.Name]; ok {
+			out[spec.Name] = id
+			continue
+		}
+		created, err := c.CreateLabel(spec)
+		if err != nil {
+			return nil, err
+		}
+		out[spec.Name] = created.ID
+		note := ""
+		if spec.Exclusive {
+			note = " (exclusive)"
+		}
+		fmt.Fprintf(os.Stderr, "created label %s%s\n", spec.Name, note)
+	}
+	return out, nil
+}
+
+// pushDep is what one `depends:` entry is, as far as linking is concerned.
+type pushDep struct {
+	// Slug is the dependency as `depends:` spells it.
+	Slug string
+	// Key is where it lives in the tracker; HasKey is false while it is
+	// local-only.
+	Key    wire.Key
+	HasKey bool
+	// InRun says this push is about to give it a number.
+	InRun bool
+}
+
+// pushDepState is every dependency this run can say anything about.
+//
+// A dependency's key is read from its `gitea:` field while the file is still on
+// disk, and from the ledger when it is not — which, since push deletes what it
+// sends, is the normal state of an already-published blocker. Without that
+// fallback the graph would quietly lose an edge every time a blocker was pushed
+// before its dependent: the file is gone, the field goes with it, and the link is
+// never made.
+//
+// A slug that is in neither the store nor the ledger names nothing this machine
+// has ever seen, and is dropped — validation has already warned about it.
+func pushDepState(i *issue.Issue, issues map[string]*issue.Issue, pushing map[string]bool,
+	keyOf map[string]wire.Key) []pushDep {
+	var out []pushDep
+	for _, d := range i.Depends {
+		dep, onDisk := issues[d]
+		var key wire.Key
+		found := false
+		if onDisk {
+			key, found = mapping.RemoteKeyOf(dep)
+		}
+		if !found {
+			key, found = keyOf[d]
+		}
+		if !onDisk && !found {
+			continue
+		}
+		out = append(out, pushDep{Slug: d, Key: key, HasKey: found, InRun: pushing[d]})
+	}
+	return out
+}
+
+// pushLinks turns `depends:` into the tracker's own dependency links.
+//
+// Topological order means every blocker that is going to have a number has one
+// already. The GET is the idempotence check — one request per issue that has
+// dependencies at all, and what makes a repeat push a no-op. A failure is a
+// warning, never an abort: one missing cross-link must not undo a push that has
+// already created issues.
+func pushLinks(c *gitea.Client, id string, number int, i *issue.Issue,
+	issues map[string]*issue.Issue, pushing map[string]bool,
+	keyOf map[string]wire.Key, repo wire.Repo) error {
+	var wanted []pushDep
+	for _, d := range pushDepState(i, issues, pushing, keyOf) {
+		if d.HasKey && d.Key.Number > 0 {
+			wanted = append(wanted, d)
+		}
+	}
+	if len(wanted) == 0 {
+		return nil
+	}
+
+	have, err := c.DependencyKeys(number)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "warning: %s: could not read the links #%d already has (%v)"+
+			" — no link was made\n", id, number, err)
+		return nil
+	}
+	for _, d := range wanted {
+		key := d.Key.In(repo)
+		if containsKey(have, key) {
+			continue
+		}
+		if err := c.AddDependency(number, key); err != nil {
+			fmt.Fprintf(os.Stderr, "warning: %s: could not link #%d -> %s (%s): %v — link it by "+
+				"hand, or `kettle pull %d` and push it again\n", id, number, key, d.Slug, err, number)
+			continue
+		}
+		fmt.Printf("    depends on  %s  (%s)\n", key, d.Slug)
+	}
+	return nil
+}
+
+// pushPlan is the --dry-run receipt: what would be sent, and which links would
+// exist. `#?` is a number this run has not handed out yet.
+func pushPlan(root string, issues map[string]*issue.Issue, order []string,
+	pushing map[string]bool, update bool) {
+	// The ledger costs no request, so a dry run resolves an already-pushed
+	// blocker exactly the way the real run does.
+	keyOf := pushLedgerKeys(loadLedgerOrFold(root, issues), wire.Repo{})
+	links := 0
+	for _, id := range order {
+		i := issues[id]
+		typ := i.Type()
+		if typ == "" {
+			typ = "?"
+		}
+		labels := strings.Join(i.Labels, ", ")
+		if labels == "" {
+			labels = "no labels"
+		}
+		fmt.Printf("ok  %s  [type/%s] %s  (%s)\n", id, typ, i.Title, labels)
+		for _, d := range pushDepState(i, issues, pushing, keyOf) {
+			switch {
+			case d.HasKey:
+				fmt.Printf("    link -> %s (%s)\n", d.Key, d.Slug)
+				links++
+			case d.InRun:
+				fmt.Printf("    link -> #? (%s, created by this run)\n", d.Slug)
+				links++
+			default:
+				fmt.Printf("    no link: %s is local-only\n", d.Slug)
+			}
+		}
+	}
+	verb := "created"
+	if update {
+		verb = "updated"
+	}
+	fmt.Printf("%d issue(s) would be %s, %d dependency link(s) would be created\n",
+		len(order), verb, links)
+}
+
+// pushLedgerKeys is slug -> key, the reverse of the ledger.
+//
+// Where a dependency's number comes from once push has deleted its file. The
+// ledger is keyed by number because that is what a pull has in hand; a push has a
+// slug, so it needs the other direction. An entry in the repository being pushed
+// to wins when a slug somehow appears under two keys.
+func pushLedgerKeys(m gitea.RemoteMap, repo wire.Repo) map[string]wire.Key {
+	raw := make([]string, 0, len(m))
+	for k := range m {
+		raw = append(raw, k)
+	}
+	sort.Strings(raw)
+
+	out := map[string]wire.Key{}
+	for _, r := range raw {
+		key, err := wire.ParseKey(r)
+		if err != nil {
+			continue
+		}
+		slug := m[r]
+		if _, seen := out[slug]; !seen || key.Repo == repo {
+			out[slug] = key
+		}
+	}
+	return out
+}
+
+// pushConfirmedNumber is the number the tracker confirmed for a write, or ok
+// false — the deletion gate.
+//
+// Every local file this command removes is removed because this returned ok, so
+// it is written to be boring and to say no by default: a positive number, and on
+// a PATCH the very number that was addressed. What it does not have to catch,
+// because none of it gets this far: a non-2xx answer, a body that is not the JSON
+// expected, or a `number` that is not a number — the transport fails all three
+// before returning, and the file survives by never reaching the delete.
+func pushConfirmedNumber(got *wire.Issue, sent int) (int, bool) {
+	if got == nil || got.Number <= 0 {
+		return 0, false
+	}
+	if sent != 0 && got.Number != sent {
+		return 0, false
+	}
+	return got.Number, true
+}
+
+// pushGitBranch is the branch HEAD is on, or "".
+//
+// The one git call this binary makes — read, never write. A detached HEAD prints
+// `HEAD` and outside a repository git exits non-zero; both mean "no branch to
+// name", which is not an error.
+func pushGitBranch() string {
+	out, err := exec.Command("git", "rev-parse", "--abbrev-ref", "HEAD").Output()
+	if err != nil {
+		return ""
+	}
+	name := strings.TrimSpace(string(out))
+	if name == "HEAD" {
+		return ""
+	}
+	return name
+}
+
+func containsKey(keys []wire.Key, want wire.Key) bool {
+	for _, k := range keys {
+		if k == want {
+			return true
+		}
+	}
+	return false
+}

cli/internal/cmd/remote.go

@@ -0,0 +1,114 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// labelColumn is how much of the label list a row shows before it is cut.
+const labelColumn = 38
+
+func init() {
+	register(&Command{
+		Name:  "remote",
+		Group: GroupSync,
+		Short: "list what exists in the tracker, one line each",
+		Long: `Discovery only: this prints and WRITES NOTHING. The local store is a store, not a
+search-results folder, and a listing that landed in it would leave files nobody
+asked for beside the issues somebody did. Pick the numbers here, then pull them.
+
+    #42    open    type/task, tech/sql          Wire sqlc into the repo layer
+                 └─ local: wire-sqlc-appclick
+
+The second line appears when the number is already in the local ledger, so it is
+obvious what a pull would refresh and what it would add.
+
+--limit here caps the LISTING: N lines out, closed ones among them. That is not
+what the same flag means to ` + "`kettle pull`" + `, and the difference is not an oversight —
+pull bounds what it WRITES, this command writes nothing, and enumeration is the
+whole job.
+
+Projects are not filterable: the projects API is not exposed by Gitea. Use
+milestones or labels, or the web UI.`,
+		Examples: []Example{
+			{"kettle remote", "the open issues, 30 of them"},
+			{"kettle remote --state all --label type/bug --limit 50", "every bug, open and closed"},
+			{"kettle remote --milestone v0.2", "what is in a milestone"},
+			{"kettle remote -q sqlc", "keyword search over title and body"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			state := fs.String("state", "open", "open, closed or all")
+			var labels stringList
+			fs.Var(&labels, "label", "filter by label; repeat for AND")
+			// Both spellings, the way the Python this replaces took them.
+			var query string
+			fs.StringVar(&query, "q", "", "search text in title and body")
+			fs.StringVar(&query, "query", "", "the long spelling of -q")
+			milestone := fs.String("milestone", "", "milestone id or title")
+			limit := fs.Int("limit", 30, "how many lines to print")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				if len(args) > 0 {
+					return Fail("remote takes no arguments — filter with --label, --milestone or -q")
+				}
+				if !contains([]string{"open", "closed", "all"}, *state) {
+					return Fail("--state %q must be open, closed or all", *state)
+				}
+				if *limit < 1 {
+					return Fail("--limit must be 1 or more, got %d", *limit)
+				}
+
+				root, client, err := syncStart(*out)
+				if err != nil {
+					return err
+				}
+				listing, err := client.ListIssues(gitea.IssueFilter{
+					State: *state, Labels: labels, Query: query,
+					Milestone: *milestone, Limit: *limit,
+				})
+				if err != nil {
+					return err
+				}
+
+				// The ledger, not the files: a pushed issue has no file left and
+				// is still something a pull would land on a known slug.
+				ledger := gitea.LoadRemoteMap(root)
+				repo := client.Repo()
+
+				for i := range listing.Issues {
+					p := &listing.Issues[i]
+					labels := "-"
+					if names := p.LabelNames(); len(names) > 0 {
+						labels = strings.Join(names, ", ")
+					}
+					// One line per issue is the whole point; a repository that
+					// namespaces heavily would wrap the column otherwise.
+					if len(labels) > labelColumn {
+						labels = labels[:labelColumn]
+					}
+					fmt.Printf("#%-5d %-7s %-38s %s\n", p.Number, p.State, labels, p.Title)
+					if local := ledger.Slug(wire.Key{Repo: repo, Number: p.Number}); local != "" {
+						fmt.Printf("%13s└─ local: %s\n", "", local)
+					}
+				}
+
+				scope := ""
+				if listing.Milestone != "" {
+					scope = " in milestone " + listing.Milestone
+				}
+				hint := "<n>"
+				if *milestone != "" {
+					hint = "--milestone " + *milestone
+				}
+				fmt.Printf("%d issue(s)%s — pull them with: kettle pull %s\n",
+					len(listing.Issues), scope, hint)
+				return nil
+			}
+		},
+	})
+}

cli/internal/cmd/sync.go

@@ -0,0 +1,70 @@
+package cmd
+
+import (
+	"path/filepath"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+// commentsSidecarPath is where an issue's comment thread lives — beside it,
+// under the same slug.
+//
+// A path, not a concept the domain needs: a thread is pulled from the tracker
+// and never pushed back, so `internal/issue` has no reason to learn that the
+// file exists. It does not have to — the file is named after the issue, and
+// issue.SlugFiles takes it away when the issue goes.
+//
+// It sits here rather than in either command because `pull` writes it and
+// `comment` rewrites it, and two spellings of one path is how the two come to
+// disagree about where a thread is.
+func commentsSidecarPath(root, id string) string {
+	return filepath.Join(root, id+".comments.md")
+}
+
+// The sync commands all start the same way and must fail the same way.
+//
+// Every one of them needs a store and a client, and the order matters: a
+// command that dialled first would report a network problem for a project that
+// was never initialized, and an operator would go looking at the wrong thing.
+// So the store is resolved before a socket is opened, and each failure names
+// the command that fixes it.
+
+// syncStart resolves the store and builds a client for it.
+//
+// The client is built from the project's own configuration, which is why there
+// is no --login flag anywhere in this tree: which login a project runs under is
+// a fact about the project, stated once by `kettle init`, not a thing a caller
+// gets to differ about per invocation. That the two could disagree is what the
+// Python version needed a PreToolUse hook to police.
+func syncStart(out string) (string, *gitea.Client, error) {
+	root, err := storeRoot(out)
+	if err != nil {
+		return "", nil, err
+	}
+	cfg, err := config.Require("")
+	if err != nil {
+		return "", nil, err
+	}
+	client, err := gitea.New(cfg)
+	if err != nil {
+		return "", nil, err
+	}
+	return root, client, nil
+}
+
+// syncStartExisting is syncStart for the commands that read the store rather
+// than create it: push, comment, close and the sync form of evict all operate
+// on issues that are already on disk, and a missing store is a mistake to
+// report, not a directory to conjure.
+func syncStartExisting(out string) (string, *gitea.Client, error) {
+	root, client, err := syncStart(out)
+	if err != nil {
+		return "", nil, err
+	}
+	if err := issue.RequireStore(root); err != nil {
+		return "", nil, err
+	}
+	return root, client, nil
+}

cli/internal/cmd/sync_pull_test.go

@@ -0,0 +1,664 @@
+package cmd_test
+
+// The transport, end to end: the real binary, run as a subprocess against a
+// throwaway project, talking to an httptest server that speaks enough of the
+// Gitea REST API to answer it.
+//
+// Enough and no more. What is worth proving here is not that JSON round-trips —
+// internal/mapping has tests for that, without a server anywhere — but the two
+// rules that cost work when they are wrong: a confirmed push takes the local file
+// with it, and an unconfirmed one does not touch it.
+//
+// The repository is always owner/repo, and the credentials arrive through
+// KETTLE_URL / KETTLE_TOKEN / KETTLE_REPO, which is also what a CI run does.
+// KETTLE_CONFIG_HOME points at a temp directory so no fixture can read or
+// overwrite the developer's own tokens.
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// --------------------------------------------------------------------------
+// the fake tracker
+// --------------------------------------------------------------------------
+
+// pullFakeGitea is a Gitea instance with one repository in it, owner/repo.
+type pullFakeGitea struct {
+	mu       sync.Mutex
+	issues   map[int]*wire.Issue
+	deps     map[int][]int
+	comments map[int][]wire.Comment
+	labels   map[string]int64
+	next     int
+
+	// writesFail makes every issue create and edit answer 500 — the failure a
+	// push has to survive without losing a file.
+	writesFail bool
+}
+
+func pullNewGitea() *pullFakeGitea {
+	return &pullFakeGitea{
+		issues:   map[int]*wire.Issue{},
+		deps:     map[int][]int{},
+		comments: map[int][]wire.Comment{},
+		labels:   map[string]int64{},
+	}
+}
+
+// pullAdd puts an issue in the tracker the way the web UI would: it is there
+// before this project ever hears about it.
+func (g *pullFakeGitea) pullAdd(p wire.Issue) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if p.State == "" {
+		p.State = "open"
+	}
+	p.HTMLURL = pullURL(p.Number)
+	g.issues[p.Number] = &p
+	if p.Number > g.next {
+		g.next = p.Number
+	}
+}
+
+func (g *pullFakeGitea) pullIssue(n int) wire.Issue {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if p := g.issues[n]; p != nil {
+		return *p
+	}
+	return wire.Issue{}
+}
+
+func (g *pullFakeGitea) pullRetitle(n int, title string) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	g.issues[n].Title = title
+}
+
+func (g *pullFakeGitea) pullBlocks(blocked int, blockers ...int) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	g.deps[blocked] = append(g.deps[blocked], blockers...)
+}
+
+func pullURL(n int) string {
+	return fmt.Sprintf("https://git.example.com/owner/repo/issues/%d", n)
+}
+
+var (
+	pullIssueRoute = regexp.MustCompile(`^issues/(\d+)$`)
+	pullSubRoute   = regexp.MustCompile(`^issues/(\d+)/(dependencies|comments|labels)$`)
+)
+
+func (g *pullFakeGitea) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	path, ok := strings.CutPrefix(r.URL.Path, "/api/v1/repos/owner/repo/")
+	if !ok {
+		http.Error(w, `{"message":"no such repository"}`, http.StatusNotFound)
+		return
+	}
+
+	switch {
+	case path == "labels" && r.Method == http.MethodGet:
+		out := []wire.Label{}
+		for name, id := range g.labels {
+			out = append(out, wire.Label{ID: id, Name: name})
+		}
+		sort.Slice(out, func(i, j int) bool { return out[i].ID < out[j].ID })
+		pullJSON(w, out)
+
+	case path == "labels" && r.Method == http.MethodPost:
+		var req wire.LabelRequest
+		pullDecode(r, &req)
+		id := int64(1000 + len(g.labels))
+		g.labels[req.Name] = id
+		pullJSON(w, wire.Label{ID: id, Name: req.Name, Color: req.Color, Exclusive: req.Exclusive})
+
+	case path == "milestones" && r.Method == http.MethodGet:
+		pullJSON(w, []wire.Milestone{})
+
+	case path == "issues" && r.Method == http.MethodPost:
+		if g.writesFail {
+			http.Error(w, `{"message":"the tracker is having a bad day"}`, http.StatusInternalServerError)
+			return
+		}
+		var req wire.IssueRequest
+		pullDecode(r, &req)
+		g.next++
+		p := &wire.Issue{
+			Number: g.next, Title: pullStr(req.Title), Body: pullStr(req.Body),
+			State: "open", HTMLURL: pullURL(g.next), Labels: g.pullLabelsFor(req.Labels),
+		}
+		g.issues[p.Number] = p
+		pullJSON(w, p)
+
+	case path == "issues" && r.Method == http.MethodGet:
+		g.pullList(w, r)
+
+	case pullIssueRoute.MatchString(path):
+		n := pullNumber(pullIssueRoute, path)
+		p := g.issues[n]
+		if p == nil {
+			http.Error(w, `{"message":"no such issue"}`, http.StatusNotFound)
+			return
+		}
+		if r.Method == http.MethodPatch {
+			if g.writesFail {
+				http.Error(w, `{"message":"the tracker is having a bad day"}`, http.StatusInternalServerError)
+				return
+			}
+			var req wire.IssueRequest
+			pullDecode(r, &req)
+			if req.Title != nil {
+				p.Title = *req.Title
+			}
+			if req.Body != nil {
+				p.Body = *req.Body
+			}
+			if req.State != nil {
+				p.State = *req.State
+			}
+			if req.Labels != nil {
+				p.Labels = g.pullLabelsFor(req.Labels)
+			}
+		}
+		pullJSON(w, p)
+
+	case pullSubRoute.MatchString(path):
+		m := pullSubRoute.FindStringSubmatch(path)
+		n, _ := strconv.Atoi(m[1])
+		switch {
+		case m[2] == "dependencies" && r.Method == http.MethodGet:
+			out := []wire.Issue{}
+			for _, d := range g.deps[n] {
+				if p := g.issues[d]; p != nil {
+					out = append(out, *p)
+				}
+			}
+			pullJSON(w, out)
+		case m[2] == "dependencies" && r.Method == http.MethodPost:
+			var req struct {
+				Index int `json:"index"`
+			}
+			pullDecode(r, &req)
+			g.deps[n] = append(g.deps[n], req.Index)
+			w.WriteHeader(http.StatusCreated)
+		case m[2] == "comments" && r.Method == http.MethodGet:
+			out := g.comments[n]
+			if out == nil {
+				out = []wire.Comment{}
+			}
+			pullJSON(w, out)
+		case m[2] == "labels" && r.Method == http.MethodPut:
+			var req struct {
+				Labels []int64 `json:"labels"`
+			}
+			pullDecode(r, &req)
+			g.issues[n].Labels = g.pullLabelsFor(&req.Labels)
+			pullJSON(w, g.issues[n].Labels)
+		default:
+			http.Error(w, `{"message":"not implemented"}`, http.StatusNotFound)
+		}
+
+	default:
+		http.Error(w, `{"message":"not implemented: `+path+`"}`, http.StatusNotFound)
+	}
+}
+
+// pullList is the filtered listing, paginated the way the client asks for it.
+func (g *pullFakeGitea) pullList(w http.ResponseWriter, r *http.Request) {
+	q := r.URL.Query()
+	state, page, limit := q.Get("state"), 1, 50
+	if v, err := strconv.Atoi(q.Get("page")); err == nil && v > 0 {
+		page = v
+	}
+	if v, err := strconv.Atoi(q.Get("limit")); err == nil && v > 0 {
+		limit = v
+	}
+	var want []string
+	if v := q.Get("labels"); v != "" {
+		want = strings.Split(v, ",")
+	}
+
+	numbers := make([]int, 0, len(g.issues))
+	for n := range g.issues {
+		numbers = append(numbers, n)
+	}
+	sort.Ints(numbers)
+
+	out := []wire.Issue{}
+	for _, n := range numbers {
+		p := g.issues[n]
+		if state != "" && state != "all" && p.State != state {
+			continue
+		}
+		has := map[string]bool{}
+		for _, l := range p.Labels {
+			has[l.Name] = true
+		}
+		missing := false
+		for _, l := range want {
+			missing = missing || !has[l]
+		}
+		if missing {
+			continue
+		}
+		out = append(out, *p)
+	}
+
+	start := (page - 1) * limit
+	if start > len(out) {
+		start = len(out)
+	}
+	end := start + limit
+	if end > len(out) {
+		end = len(out)
+	}
+	pullJSON(w, out[start:end])
+}
+
+func (g *pullFakeGitea) pullLabelsFor(ids *[]int64) []wire.Label {
+	if ids == nil {
+		return nil
+	}
+	byID := map[int64]string{}
+	for name, id := range g.labels {
+		byID[id] = name
+	}
+	var out []wire.Label
+	for _, id := range *ids {
+		if name, ok := byID[id]; ok {
+			out = append(out, wire.Label{ID: id, Name: name})
+		}
+	}
+	return out
+}
+
+func pullNumber(re *regexp.Regexp, path string) int {
+	n, _ := strconv.Atoi(re.FindStringSubmatch(path)[1])
+	return n
+}
+
+func pullDecode(r *http.Request, into any) {
+	_ = json.NewDecoder(r.Body).Decode(into)
+}
+
+func pullJSON(w http.ResponseWriter, v any) {
+	w.Header().Set("Content-Type", "application/json")
+	_ = json.NewEncoder(w).Encode(v)
+}
+
+func pullStr(p *string) string {
+	if p == nil {
+		return ""
+	}
+	return *p
+}
+
+// pullEnv starts the fake and returns the environment that points the binary at
+// it. The credential home is a temp directory: a test run may neither read nor
+// overwrite the developer's own tokens.
+func pullEnv(t *testing.T, g *pullFakeGitea) []string {
+	t.Helper()
+	srv := httptest.NewServer(g)
+	t.Cleanup(srv.Close)
+	return []string{
+		config.EnvURL + "=" + srv.URL,
+		config.EnvToken + "=t0ken",
+		config.EnvRepo + "=owner/repo",
+		config.EnvHome + "=" + t.TempDir(),
+	}
+}
+
+func pullStore(dir string) string { return filepath.Join(dir, ".kettle", "issues") }
+
+func pullRead(t *testing.T, path string) string {
+	t.Helper()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(raw)
+}
+
+func pullExists(path string) bool {
+	_, err := os.Stat(path)
+	return err == nil
+}
+
+// --------------------------------------------------------------------------
+// push
+// --------------------------------------------------------------------------
+
+// The rule the whole design rests on: once the tracker has the issue, the
+// tracker IS the issue, and the local copy goes — sidecars included.
+func TestPushCreatesTheIssueAndTakesTheLocalCopyWithIt(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	env := pullEnv(t, g)
+
+	mustRun(t, dir, "new", "--type", "task", "--title", "Wire sqlc into the appclick layer")
+	const id = "wire-sqlc-into-the-appclick-layer"
+	store := pullStore(dir)
+	sidecar := filepath.Join(store, id+".comments.md")
+	if err := os.WriteFile(sidecar, []byte("# thread\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	r := runWith(t, dir, env, "", "push")
+	if r.code != 0 {
+		t.Fatalf("push exited %d:\n%s", r.code, r.out())
+	}
+	// The number and the URL lead: in a moment they are the only address the
+	// issue has.
+	if !strings.Contains(r.stdout, "created  "+id+"  #1  "+pullURL(1)) {
+		t.Errorf("the receipt does not say where the issue lives now:\n%s", r.stdout)
+	}
+
+	if pullExists(filepath.Join(store, id+".md")) {
+		t.Error("the local file survived a confirmed push — what is in the store is what has not left")
+	}
+	if pullExists(sidecar) {
+		t.Error("the sidecar was left behind; every file under the slug goes")
+	}
+
+	// The ledger is what makes the slug come back, so it has to hold the number.
+	ledger := pullRead(t, filepath.Join(store, ".remote.json"))
+	if !strings.Contains(ledger, `"owner/repo#1": "`+id+`"`) {
+		t.Errorf("the ledger does not index the number:\n%s", ledger)
+	}
+	// And the slug travelled up in the body, which is what survives a lost ledger.
+	if body := g.pullIssue(1).Body; !strings.Contains(body, "<!-- kettle:id "+id+" -->") {
+		t.Errorf("the id marker did not go up with the issue:\n%s", body)
+	}
+	if !strings.HasPrefix(g.pullIssue(1).Body, "<!-- kettle:id") {
+		t.Error("the marker must be the first line of the tracker-side body")
+	}
+}
+
+// Network down, non-2xx, an answer that does not confirm the write: the file
+// stays and the run stops. Nothing is deleted that was not just accepted.
+func TestPushLeavesTheFileWhenTheTrackerRefuses(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	g.writesFail = true
+	env := pullEnv(t, g)
+
+	mustRun(t, dir, "new", "--type", "task", "--title", "Never made it up")
+	const id = "never-made-it-up"
+	path := filepath.Join(pullStore(dir), id+".md")
+	before := pullRead(t, path)
+
+	r := runWith(t, dir, env, "", "push")
+	if r.code == 0 {
+		t.Fatalf("a tracker that refuses the write must fail the run:\n%s", r.out())
+	}
+	if after := pullRead(t, path); after != before {
+		t.Errorf("the file was touched by a push that never landed:\n%s", after)
+	}
+	// The message has to name the file, because "is my only copy still there" is
+	// the question an operator has at that moment.
+	if !strings.Contains(r.stderr, path) {
+		t.Errorf("the failure does not name the file it did not touch:\n%s", r.stderr)
+	}
+	if pullExists(filepath.Join(pullStore(dir), ".remote.json")) {
+		t.Error("a ledger entry was written for an issue the tracker never confirmed")
+	}
+}
+
+// --------------------------------------------------------------------------
+// pull
+// --------------------------------------------------------------------------
+
+// A number is an address, not a query. Only filter mode leaves closed issues out.
+func TestPullByNumberFetchesAClosedIssue(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	g.pullAdd(wire.Issue{
+		Number: 7, Title: "Closed but addressable", State: "closed",
+		Body: "## Summary\nДело сделано.\n", UpdatedAt: "2026-08-01T10:00:00Z",
+	})
+	env := pullEnv(t, g)
+
+	r := runWith(t, dir, env, "", "pull", "7")
+	if r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	file := pullRead(t, filepath.Join(pullStore(dir), "closed-but-addressable.md"))
+	if !strings.Contains(file, "state: closed") {
+		t.Errorf("the closed state did not land on disk:\n%s", file)
+	}
+	if !strings.Contains(file, "gitea: owner/repo#7") {
+		t.Errorf("the cross-repo handle is missing:\n%s", file)
+	}
+	if !strings.Contains(file, "origin: gitea") {
+		t.Errorf("the issue does not say it exists elsewhere:\n%s", file)
+	}
+}
+
+// A pull answers with the unit of work — the issue and what blocks it — and
+// --no-deps is how you ask for one row of it.
+func TestPullBringsTheBlockerDownWithIt(t *testing.T) {
+	g := pullNewGitea()
+	g.pullAdd(wire.Issue{Number: 1, Title: "Migrate the schema", Body: "## Summary\nx\n"})
+	g.pullAdd(wire.Issue{Number: 2, Title: "Wire sqlc into the layer", Body: "## Summary\nx\n"})
+	g.pullBlocks(2, 1)
+	env := pullEnv(t, g)
+
+	t.Run("by default", func(t *testing.T) {
+		dir := newProject(t)
+		if r := runWith(t, dir, env, "", "pull", "2"); r.code != 0 {
+			t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+		}
+		if !pullExists(filepath.Join(pullStore(dir), "migrate-the-schema.md")) {
+			t.Fatal("the blocker did not come down — a pull returns the unit of work")
+		}
+		dependent := pullRead(t, filepath.Join(pullStore(dir), "wire-sqlc-into-the-layer.md"))
+		if !strings.Contains(dependent, "depends: [migrate-the-schema]") {
+			t.Errorf("depends: was not filled from the tracker's own graph:\n%s", dependent)
+		}
+	})
+
+	t.Run("--no-deps", func(t *testing.T) {
+		dir := newProject(t)
+		if r := runWith(t, dir, env, "", "pull", "2", "--no-deps"); r.code != 0 {
+			t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+		}
+		if pullExists(filepath.Join(pullStore(dir), "migrate-the-schema.md")) {
+			t.Error("--no-deps followed a blocker anyway")
+		}
+		dependent := pullRead(t, filepath.Join(pullStore(dir), "wire-sqlc-into-the-layer.md"))
+		if !strings.Contains(dependent, "depends: []") {
+			t.Errorf("--no-deps filled depends: anyway:\n%s", dependent)
+		}
+	})
+}
+
+// The round trip, and the two things that carry the slug through it: the ledger,
+// and — when the ledger is gone, as it is in a fresh clone — the marker in the
+// body. A rename in the web UI changes neither.
+func TestAPushedIssueComesBackUnderItsOriginalSlug(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	env := pullEnv(t, g)
+
+	mustRun(t, dir, "new", "--type", "task", "--title", "Wire sqlc into the appclick layer")
+	const id = "wire-sqlc-into-the-appclick-layer"
+	store := pullStore(dir)
+
+	if r := runWith(t, dir, env, "", "push"); r.code != 0 {
+		t.Fatalf("push exited %d:\n%s", r.code, r.out())
+	}
+	if pullExists(filepath.Join(store, id+".md")) {
+		t.Fatal("push did not drop the local copy")
+	}
+	g.pullRetitle(1, "Somebody retitled this in the web UI")
+
+	// The ledger knows the number, so it wins.
+	if r := runWith(t, dir, env, "", "pull", "1"); r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	file := pullRead(t, filepath.Join(store, id+".md"))
+	if !strings.Contains(file, "# Somebody retitled this in the web UI") {
+		t.Errorf("the new title did not come down:\n%s", file)
+	}
+	// The marker is transport bookkeeping and never reaches the store.
+	if strings.Contains(file, "kettle:id") {
+		t.Errorf("the id marker was written into the local file:\n%s", file)
+	}
+
+	// Now lose both the file and the ledger, the way a fresh clone has neither.
+	// The marker in the body is all that is left, and it is enough.
+	for _, p := range []string{filepath.Join(store, id+".md"), filepath.Join(store, ".remote.json")} {
+		if err := os.Remove(p); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if r := runWith(t, dir, env, "", "pull", "1"); r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	if !pullExists(filepath.Join(store, id+".md")) {
+		names, _ := os.ReadDir(store)
+		var have []string
+		for _, e := range names {
+			have = append(have, e.Name())
+		}
+		t.Fatalf("the issue came back under another name — every depends: pointing at it now "+
+			"dangles; the store holds: %s", strings.Join(have, ", "))
+	}
+}
+
+// A closed issue is not a unit of work, so a FILTER enumerates it and leaves it
+// out — the exact opposite of what a key does, and only --state closed changes
+// it.
+func TestPullFilterModeLeavesClosedIssuesOut(t *testing.T) {
+	g := pullNewGitea()
+	bug := []wire.Label{{ID: 1, Name: "type/bug"}}
+	g.pullAdd(wire.Issue{Number: 1, Title: "Still broken", Body: "## Summary\nx\n", Labels: bug})
+	g.pullAdd(wire.Issue{Number: 2, Title: "Fixed last week", State: "closed",
+		Body: "## Summary\nx\n", Labels: bug})
+	env := pullEnv(t, g)
+
+	dir := newProject(t)
+	r := runWith(t, dir, env, "", "pull", "--label", "type/bug", "--state", "all")
+	if r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	if pullExists(filepath.Join(pullStore(dir), "fixed-last-week.md")) {
+		t.Error("a filter stored a closed issue")
+	}
+	// Nothing is dropped in silence.
+	if !strings.Contains(r.stderr, "1 closed issue(s) enumerated, not stored") {
+		t.Errorf("the closed issue went out without a word:\n%s", r.stderr)
+	}
+
+	// Naming the state is how you ask for one.
+	if r := runWith(t, dir, env, "", "pull", "--label", "type/bug", "--state", "closed"); r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	if !pullExists(filepath.Join(pullStore(dir), "fixed-last-week.md")) {
+		t.Error("--state closed did not store the closed issue")
+	}
+}
+
+// ONE RULE, NO EXCEPTION: a PATCH is a push, and it drops the local copy too.
+func TestPushUpdateDropsTheLocalCopyAsWell(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	g.pullAdd(wire.Issue{
+		Number: 3, Title: "Came down and went back up",
+		Body:   "## Summary\nx\n\n## Spec\nnone\n\n## Acceptance criteria\n- [ ] x\n",
+		Labels: []wire.Label{{ID: 1, Name: "type/task"}},
+	})
+	env := pullEnv(t, g)
+
+	if r := runWith(t, dir, env, "", "pull", "3"); r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	const id = "came-down-and-went-back-up"
+	path := filepath.Join(pullStore(dir), id+".md")
+	if !pullExists(path) {
+		t.Fatal("the issue did not arrive")
+	}
+
+	r := runWith(t, dir, env, "", "push", "--update", id)
+	if r.code != 0 {
+		t.Fatalf("push --update exited %d:\n%s", r.code, r.out())
+	}
+	if !strings.Contains(r.stdout, "updated  "+id+"  #3") {
+		t.Errorf("the receipt does not report the PATCH:\n%s", r.stdout)
+	}
+	if pullExists(path) {
+		t.Error("--update kept the local file — two rules would put back the question " +
+			"push exists to remove")
+	}
+}
+
+// A dry run makes no request, so it must not need a credential to say what it
+// would do — no URL, no token, no repository in the environment at all.
+func TestPushDryRunNeedsNoCredential(t *testing.T) {
+	dir := newProject(t)
+	mustRun(t, dir, "new", "--type", "task", "--title", "Planned but not sent")
+
+	r := run(t, dir, "push", "--dry-run")
+	if r.code != 0 {
+		t.Fatalf("a dry run must not need a tracker:\n%s", r.out())
+	}
+	if !strings.Contains(r.stdout, "ok  planned-but-not-sent") ||
+		!strings.Contains(r.stdout, "1 issue(s) would be created") {
+		t.Errorf("the plan was not printed:\n%s", r.stdout)
+	}
+	if !pullExists(filepath.Join(pullStore(dir), "planned-but-not-sent.md")) {
+		t.Error("a dry run deleted the issue")
+	}
+}
+
+// --------------------------------------------------------------------------
+// remote
+// --------------------------------------------------------------------------
+
+// Discovery writes nothing: the store is a store, not a search-results folder.
+func TestRemoteListsWithoutWritingAnything(t *testing.T) {
+	dir := newProject(t)
+	g := pullNewGitea()
+	g.pullAdd(wire.Issue{Number: 4, Title: "Something open", Body: "x"})
+	g.pullAdd(wire.Issue{Number: 5, Title: "Something closed", State: "closed", Body: "x"})
+	env := pullEnv(t, g)
+
+	r := runWith(t, dir, env, "", "remote")
+	if r.code != 0 {
+		t.Fatalf("remote exited %d:\n%s", r.code, r.out())
+	}
+	if !strings.Contains(r.stdout, "#4") || strings.Contains(r.stdout, "#5") {
+		t.Errorf("the default listing is the open issues:\n%s", r.stdout)
+	}
+	if entries, err := os.ReadDir(pullStore(dir)); err != nil || len(entries) != 0 {
+		t.Errorf("a listing left files in the store: %v", entries)
+	}
+
+	// A number the store already knows about says so, so it is obvious what a
+	// pull would refresh and what it would add.
+	if r := runWith(t, dir, env, "", "pull", "4"); r.code != 0 {
+		t.Fatalf("pull exited %d:\n%s", r.code, r.out())
+	}
+	again := runWith(t, dir, env, "", "remote")
+	if !strings.Contains(again.stdout, "└─ local: something-open") {
+		t.Errorf("the local slug was not reported:\n%s", again.stdout)
+	}
+}

cli/internal/cmd/sync_write_test.go

@@ -0,0 +1,631 @@
+package cmd_test
+
+// The four commands that WRITE — comment, close, labels, sync-evict — end to
+// end: the real binary, in a throwaway project, against an httptest server
+// speaking enough of the Gitea API to answer them.
+//
+// A fake tracker rather than a mocked client, because what these commands are
+// trusted to get right is exactly the part a mock would stand in for: what goes
+// out, and what is believed about the answer. `sync-evict` deletes files on the
+// strength of a payload, so the payload has to come off a socket.
+//
+// Every helper here is named `wr…` so it cannot collide with the read side's.
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"regexp"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+const (
+	wrRepo  = "kettle/tests"
+	wrToken = "s3cr3t-token"
+	wrWhen  = "2026-08-11T12:00:00Z"
+)
+
+// --------------------------------------------------------------------------
+// the fake tracker
+// --------------------------------------------------------------------------
+
+type wrLabel struct {
+	ID          int64  `json:"id"`
+	Name        string `json:"name"`
+	Color       string `json:"color"`
+	Description string `json:"description"`
+	Exclusive   bool   `json:"exclusive"`
+}
+
+type wrIssue struct {
+	Number    int    `json:"number"`
+	Title     string `json:"title"`
+	Body      string `json:"body"`
+	State     string `json:"state"`
+	HTMLURL   string `json:"html_url"`
+	UpdatedAt string `json:"updated_at"`
+}
+
+type wrUser struct {
+	Login string `json:"login"`
+}
+
+type wrComment struct {
+	ID        int64  `json:"id"`
+	Body      string `json:"body"`
+	HTMLURL   string `json:"html_url"`
+	User      wrUser `json:"user"`
+	CreatedAt string `json:"created_at"`
+}
+
+var (
+	wrIssuePath   = regexp.MustCompile(`^issues/(\d+)$`)
+	wrCommentPath = regexp.MustCompile(`^issues/(\d+)/comments$`)
+	wrLabelPath   = regexp.MustCompile(`^labels/(\d+)$`)
+)
+
+// wrTracker is one repository on a pretend Gitea. It records every call, so a
+// test can assert that a dry run sent nothing and that a second bootstrap wrote
+// nothing.
+type wrTracker struct {
+	mu       sync.Mutex
+	labels   []wrLabel
+	issues   map[int]*wrIssue
+	comments map[int][]wrComment
+	broken   map[int]bool // numbers whose GET answers 500
+	calls    []string
+	next     int64
+}
+
+func wrNewTracker() *wrTracker {
+	return &wrTracker{
+		issues:   map[int]*wrIssue{},
+		comments: map[int][]wrComment{},
+		broken:   map[int]bool{},
+		next:     100,
+	}
+}
+
+func (tr *wrTracker) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	tr.calls = append(tr.calls, r.Method+" "+r.URL.Path)
+
+	// The scheme Gitea uses and the client sends: the word `token`.
+	if r.Header.Get("Authorization") != "token "+wrToken {
+		http.Error(w, `{"message":"token required"}`, http.StatusUnauthorized)
+		return
+	}
+	path, ok := strings.CutPrefix(r.URL.Path, "/api/v1/repos/"+wrRepo+"/")
+	if !ok {
+		http.Error(w, `{"message":"no such repository"}`, http.StatusNotFound)
+		return
+	}
+
+	switch {
+	case path == "labels" && r.Method == http.MethodGet:
+		wrJSON(w, tr.labels)
+
+	case path == "labels" && r.Method == http.MethodPost:
+		var req wrLabel
+		wrDecode(r, &req)
+		tr.next++
+		req.ID = tr.next
+		tr.labels = append(tr.labels, req)
+		wrJSON(w, req)
+
+	case r.Method == http.MethodPatch && wrLabelPath.MatchString(path):
+		id, _ := strconv.ParseInt(wrLabelPath.FindStringSubmatch(path)[1], 10, 64)
+		var req wrLabel
+		wrDecode(r, &req)
+		for i := range tr.labels {
+			if tr.labels[i].ID == id {
+				req.ID = id
+				tr.labels[i] = req
+				wrJSON(w, req)
+				return
+			}
+		}
+		http.Error(w, `{"message":"no such label"}`, http.StatusNotFound)
+
+	case wrIssuePath.MatchString(path):
+		n, _ := strconv.Atoi(wrIssuePath.FindStringSubmatch(path)[1])
+		if tr.broken[n] {
+			http.Error(w, `{"message":"the tracker is having a bad day"}`, http.StatusInternalServerError)
+			return
+		}
+		got := tr.issues[n]
+		if got == nil {
+			http.Error(w, `{"message":"no such issue"}`, http.StatusNotFound)
+			return
+		}
+		if r.Method == http.MethodPatch {
+			var req struct {
+				State *string `json:"state"`
+				Title *string `json:"title"`
+			}
+			wrDecode(r, &req)
+			// A close is state and nothing else; a title arriving here would be
+			// the command editing an issue it was only asked to close.
+			if req.Title != nil {
+				http.Error(w, `{"message":"close sent a title"}`, http.StatusUnprocessableEntity)
+				return
+			}
+			if req.State != nil {
+				got.State = *req.State
+			}
+			got.UpdatedAt = wrWhen
+		}
+		wrJSON(w, got)
+
+	case wrCommentPath.MatchString(path):
+		n, _ := strconv.Atoi(wrCommentPath.FindStringSubmatch(path)[1])
+		if tr.issues[n] == nil {
+			http.Error(w, `{"message":"no such issue"}`, http.StatusNotFound)
+			return
+		}
+		if r.Method == http.MethodPost {
+			var req struct {
+				Body string `json:"body"`
+			}
+			wrDecode(r, &req)
+			tr.next++
+			c := wrComment{
+				ID:        tr.next,
+				Body:      req.Body,
+				HTMLURL:   fmt.Sprintf("https://tracker.example/%s/issues/%d#issuecomment-%d", wrRepo, n, tr.next),
+				User:      wrUser{Login: "tester"},
+				CreatedAt: wrWhen,
+			}
+			tr.comments[n] = append(tr.comments[n], c)
+			wrJSON(w, c)
+			return
+		}
+		wrJSON(w, tr.comments[n])
+
+	default:
+		http.Error(w, `{"message":"not implemented: `+path+`"}`, http.StatusNotFound)
+	}
+}
+
+func wrJSON(w http.ResponseWriter, v any) {
+	w.Header().Set("Content-Type", "application/json")
+	_ = json.NewEncoder(w).Encode(v)
+}
+
+func wrDecode(r *http.Request, v any) {
+	_ = json.NewDecoder(r.Body).Decode(v)
+}
+
+// --- what the tracker holds, for a test to arrange and to read back ---------
+
+func (tr *wrTracker) add(number int, title, state string) {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	tr.issues[number] = &wrIssue{
+		Number:    number,
+		Title:     title,
+		State:     state,
+		HTMLURL:   fmt.Sprintf("https://tracker.example/%s/issues/%d", wrRepo, number),
+		UpdatedAt: wrWhen,
+	}
+}
+
+func (tr *wrTracker) breaks(number int) {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	tr.broken[number] = true
+}
+
+func (tr *wrTracker) state(number int) string {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	if got := tr.issues[number]; got != nil {
+		return got.State
+	}
+	return ""
+}
+
+func (tr *wrTracker) label(name string) *wrLabel {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	for i := range tr.labels {
+		if tr.labels[i].Name == name {
+			out := tr.labels[i]
+			return &out
+		}
+	}
+	return nil
+}
+
+func (tr *wrTracker) labelCount() int {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	return len(tr.labels)
+}
+
+func (tr *wrTracker) thread(number int) []wrComment {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	return append([]wrComment{}, tr.comments[number]...)
+}
+
+// mark is where the log has got to, so a test can ask what one run sent.
+func (tr *wrTracker) mark() int {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	return len(tr.calls)
+}
+
+func (tr *wrTracker) since(mark int) []string {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	return append([]string{}, tr.calls[mark:]...)
+}
+
+func (tr *wrTracker) count(method string) int {
+	tr.mu.Lock()
+	defer tr.mu.Unlock()
+	n := 0
+	for _, c := range tr.calls {
+		if strings.HasPrefix(c, method+" ") {
+			n++
+		}
+	}
+	return n
+}
+
+// --------------------------------------------------------------------------
+// the fixture
+// --------------------------------------------------------------------------
+
+// wrProject is an initialized project pointed at a fake tracker.
+//
+// The credentials arrive through the environment, which is what they are there
+// for — and KETTLE_CONFIG_HOME goes at a temp directory so a run can neither
+// read nor overwrite the developer's own tokens. KETTLE_LOGIN is cleared for the
+// same reason: a value in the developer's shell would send every fixture
+// looking for a login that is not in the temp file.
+func wrProject(t *testing.T) (dir string, tr *wrTracker, env []string) {
+	t.Helper()
+	dir = newProject(t)
+	tr = wrNewTracker()
+	srv := httptest.NewServer(tr)
+	t.Cleanup(srv.Close)
+	return dir, tr, []string{
+		"KETTLE_URL=" + srv.URL,
+		"KETTLE_TOKEN=" + wrToken,
+		"KETTLE_REPO=" + wrRepo,
+		"KETTLE_CONFIG_HOME=" + t.TempDir(),
+		"KETTLE_LOGIN=",
+	}
+}
+
+func wrStore(dir string) string { return filepath.Join(dir, ".kettle", "issues") }
+
+// wrTracked writes an issue the tracker also holds — a working copy, as a pull
+// would have left it.
+func wrTracked(t *testing.T, dir, id string, number int, state string) string {
+	t.Helper()
+	return wrWrite(t, dir, id, fmt.Sprintf(
+		"---\nid: %s\nstate: %s\nlabels: [type/task]\nassignees: []\nmilestone: none\ndepends: []\n"+
+			"origin: gitea\ngitea: %s#%d\nsynced: 2026-01-01T00:00:00Z\nurl: https://tracker.example/%s/issues/%d\n---\n"+
+			"# %s\n\n## Summary\nчто-то\n\n## Spec\nnone\n\n## Acceptance criteria\n- [ ] сделано\n",
+		id, state, wrRepo, number, wrRepo, number, id))
+}
+
+// wrLocal writes an `origin: local` issue — the only copy of that work.
+func wrLocal(t *testing.T, dir, id, state string) string {
+	t.Helper()
+	return wrWrite(t, dir, id, fmt.Sprintf(
+		"---\nid: %s\nstate: %s\nlabels: [type/task]\nassignees: []\nmilestone: none\ndepends: []\n"+
+			"origin: local\n---\n# %s\n\n## Summary\nчто-то\n\n## Spec\nnone\n\n## Acceptance criteria\n- [ ] сделано\n",
+		id, state, id))
+}
+
+func wrWrite(t *testing.T, dir, id, text string) string {
+	t.Helper()
+	path := filepath.Join(wrStore(dir), id+".md")
+	if err := os.WriteFile(path, []byte(text), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return path
+}
+
+func wrRead(t *testing.T, path string) string {
+	t.Helper()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(raw)
+}
+
+func wrGone(t *testing.T, path, why string) {
+	t.Helper()
+	if _, err := os.Stat(path); err == nil {
+		t.Errorf("%s is still there — %s", filepath.Base(path), why)
+	}
+}
+
+func wrThere(t *testing.T, path, why string) {
+	t.Helper()
+	if _, err := os.Stat(path); err != nil {
+		t.Fatalf("%s is gone — %s", filepath.Base(path), why)
+	}
+}
+
+// --------------------------------------------------------------------------
+// labels
+// --------------------------------------------------------------------------
+
+func TestLabelsCreatesTheCanonicalSetAndThenChangesNothing(t *testing.T) {
+	dir, tr, env := wrProject(t)
+
+	r := runWith(t, dir, env, "", "labels")
+	if r.code != 0 {
+		t.Fatalf("labels exited %d:\n%s", r.code, r.out())
+	}
+
+	// The set is the domain's, name for name: nothing is spelled out in the
+	// command, so adding a type over there is what adds it here.
+	want := issue.CanonicalLabels()
+	if got := tr.labelCount(); got != len(want) {
+		t.Fatalf("the repository holds %d label(s), want %d:\n%s", got, len(want), r.out())
+	}
+	for _, name := range want {
+		l := tr.label(name)
+		if l == nil {
+			t.Fatalf("%s was not created:\n%s", name, r.out())
+		}
+		if l.Color == "" {
+			t.Errorf("%s was created with no colour", name)
+		}
+		// `exclusive` is the flag no tracker CLI could set, and the whole reason
+		// label creation goes through the API.
+		if !l.Exclusive {
+			t.Errorf("%s is not exclusive", name)
+		}
+	}
+	if !strings.Contains(r.stdout, "created  type/bug") {
+		t.Errorf("the receipt does not name what it created:\n%s", r.stdout)
+	}
+
+	// A label belongs to the repository, not to any issue: this must not have
+	// touched the store.
+	if _, err := os.Stat(filepath.Join(wrStore(dir), "INDEX.md")); err == nil {
+		t.Error("a label bootstrap wrote into the issue store")
+	}
+
+	mark := tr.mark()
+	again := runWith(t, dir, env, "", "labels")
+	if again.code != 0 {
+		t.Fatalf("the second run exited %d:\n%s", again.code, again.out())
+	}
+	for _, c := range tr.since(mark) {
+		if !strings.HasPrefix(c, "GET ") {
+			t.Errorf("the second run wrote: %s", c)
+		}
+	}
+	if !strings.Contains(again.stdout, "present  type/bug") ||
+		!strings.Contains(again.stdout, "0 created") {
+		t.Errorf("a second run must be a no-op and say so:\n%s", again.stdout)
+	}
+}
+
+// --------------------------------------------------------------------------
+// close
+// --------------------------------------------------------------------------
+
+func TestCloseChangesTheStateOnTheTrackerAndOnDisk(t *testing.T) {
+	dir, tr, env := wrProject(t)
+	tr.add(42, "Done and elsewhere", "open")
+	tr.add(99, "Never seen here", "open")
+	path := wrTracked(t, dir, "done-and-elsewhere", 42, "open")
+
+	dry := runWith(t, dir, env, "", "close", "--dry-run", "done-and-elsewhere")
+	if dry.code != 0 || !strings.Contains(dry.stdout, "would close") {
+		t.Fatalf("the dry run said nothing:\n%s", dry.out())
+	}
+	if n := tr.count("PATCH"); n != 0 {
+		t.Errorf("a dry run sent %d write(s) — it must make no request at all", n)
+	}
+	if !strings.Contains(wrRead(t, path), "state: open") {
+		t.Error("a dry run wrote to the local file")
+	}
+
+	r := runWith(t, dir, env, "", "close", "done-and-elsewhere")
+	if r.code != 0 {
+		t.Fatalf("close exited %d:\n%s", r.code, r.out())
+	}
+	if got := tr.state(42); got != "closed" {
+		t.Errorf("the tracker says %q, want closed", got)
+	}
+	local := wrRead(t, path)
+	if !strings.Contains(local, "state: closed") {
+		t.Errorf("the local copy was not brought along:\n%s", local)
+	}
+	// The answer that authorized the write is also the newest thing the tracker
+	// has said, so the freshness fields are stamped from it.
+	if !strings.Contains(local, "remote-updated: "+wrWhen) || !strings.Contains(local, "synced: 20") {
+		t.Errorf("the freshness fields were not stamped:\n%s", local)
+	}
+	index := filepath.Join(wrStore(dir), "INDEX.md")
+	if !strings.Contains(wrRead(t, index), "closed") {
+		t.Error("INDEX.md was not rebuilt from what is now on disk")
+	}
+
+	// A number this machine has never seen: closed in the tracker, nothing
+	// written here, and the receipt says which is which.
+	byNumber := runWith(t, dir, env, "", "close", "99")
+	if byNumber.code != 0 {
+		t.Fatalf("closing by number exited %d:\n%s", byNumber.code, byNumber.out())
+	}
+	if got := tr.state(99); got != "closed" {
+		t.Errorf("#99 says %q, want closed", got)
+	}
+	if !strings.Contains(byNumber.stdout, "no local copy") {
+		t.Errorf("the receipt hid that there was nothing to write:\n%s", byNumber.stdout)
+	}
+
+	// A number resolves through the file that carries the handle, so the local
+	// copy of #42 is kept honest even when it was named by number.
+	back := runWith(t, dir, env, "", "close", "--reopen", "42")
+	if back.code != 0 {
+		t.Fatalf("reopening by number exited %d:\n%s", back.code, back.out())
+	}
+	if got := tr.state(42); got != "open" {
+		t.Errorf("#42 says %q, want open", got)
+	}
+	if !strings.Contains(wrRead(t, path), "state: open") {
+		t.Errorf("a number named the tracker but not the local copy holding its handle:\n%s", wrRead(t, path))
+	}
+
+	// An issue that has never left this machine has no state in the tracker to
+	// change, and saying so beats editing one field of a local file.
+	wrLocal(t, dir, "never-left-here", "open")
+	refused := runWith(t, dir, env, "", "close", "never-left-here")
+	if refused.code == 0 || !strings.Contains(refused.stderr, "push") {
+		t.Errorf("closing a local issue must stop and say why:\n%s", refused.out())
+	}
+}
+
+// --------------------------------------------------------------------------
+// comment
+// --------------------------------------------------------------------------
+
+func TestCommentPostsAndTheThreadLandsBesideTheIssue(t *testing.T) {
+	dir, tr, env := wrProject(t)
+	tr.add(42, "Talk about it", "open")
+	wrTracked(t, dir, "talk-about-it", 42, "open")
+	wrLocal(t, dir, "never-left-here", "open")
+
+	const said = "готово, задеплоено"
+	r := runWith(t, dir, env, "", "comment", "talk-about-it", "--body", said)
+	if r.code != 0 {
+		t.Fatalf("comment exited %d:\n%s", r.code, r.out())
+	}
+	thread := tr.thread(42)
+	if len(thread) != 1 || thread[0].Body != said {
+		t.Fatalf("the tracker holds %v", thread)
+	}
+
+	sidecar := filepath.Join(wrStore(dir), "talk-about-it.comments.md")
+	got := wrRead(t, sidecar)
+	if !strings.Contains(got, said) || !strings.Contains(got, "## comment ") {
+		t.Errorf("the thread did not land beside the issue:\n%s", got)
+	}
+	if !strings.Contains(r.stdout, "posted comment") || !strings.Contains(r.stdout, "thread:") {
+		t.Errorf("the receipt does not say what happened:\n%s", r.stdout)
+	}
+
+	// The target is a local id resolved through the `gitea:` handle, so an issue
+	// that carries none cannot be commented on at all.
+	refused := runWith(t, dir, env, "", "comment", "never-left-here", "--body", "x")
+	if refused.code == 0 || !strings.Contains(refused.stderr, "push") {
+		t.Errorf("commenting on a local-only issue must stop and say why:\n%s", refused.out())
+	}
+	if n := tr.count("POST"); n != 1 {
+		t.Errorf("%d comment(s) went out, want 1 — the refused one was sent anyway", n)
+	}
+}
+
+// --------------------------------------------------------------------------
+// sync-evict
+// --------------------------------------------------------------------------
+
+// The one thing this command adds to the offline evict: a `state:` that is not
+// stale. The file says open, the tracker says closed, and the tracker is right.
+func TestSyncEvictRefreshesTheStateBeforeItDecides(t *testing.T) {
+	dir, tr, env := wrProject(t)
+	tr.add(42, "Closed in the web ui", "closed")
+	path := wrTracked(t, dir, "closed-in-the-web-ui", 42, "open")
+
+	offline := runWith(t, dir, env, "", "evict")
+	if offline.code != 0 || !strings.Contains(offline.stdout, "0 issue(s) evicted") {
+		t.Fatalf("the offline evict must keep an issue whose file reads open:\n%s", offline.out())
+	}
+	wrThere(t, path, "the offline evict asks the file, and the file says open")
+
+	// A dry run asks, reports, and touches nothing.
+	dry := runWith(t, dir, env, "", "sync-evict", "--dry-run")
+	if dry.code != 0 || !strings.Contains(dry.stdout, "would evict") {
+		t.Fatalf("the dry run said nothing:\n%s", dry.out())
+	}
+	wrThere(t, path, "a dry run deleted the issue")
+	if !strings.Contains(wrRead(t, path), "state: open") {
+		t.Error("a dry run wrote the refreshed state to disk")
+	}
+
+	r := runWith(t, dir, env, "", "sync-evict")
+	if r.code != 0 {
+		t.Fatalf("sync-evict exited %d:\n%s", r.code, r.out())
+	}
+	if !strings.Contains(r.stdout, "open -> closed") {
+		t.Errorf("the refresh was not reported:\n%s", r.stdout)
+	}
+	wrGone(t, path, "the tracker said it was closed")
+	if index := wrRead(t, filepath.Join(wrStore(dir), "INDEX.md")); strings.Contains(index, "closed-in-the-web-ui") {
+		t.Errorf("INDEX.md still lists the evicted issue:\n%s", index)
+	}
+}
+
+func TestSyncEvictKeepsALocalIssueTheTrackerNeverHeardOf(t *testing.T) {
+	dir, tr, env := wrProject(t)
+	tr.add(42, "Done elsewhere", "closed")
+	tracked := wrTracked(t, dir, "done-elsewhere", 42, "closed")
+	local := wrLocal(t, dir, "only-copy-there-is", "closed")
+
+	r := runWith(t, dir, env, "", "sync-evict")
+	if r.code != 0 {
+		t.Fatalf("sync-evict exited %d:\n%s", r.code, r.out())
+	}
+	wrThere(t, local, "a closed origin: local issue was deleted, and that file IS the work")
+	wrGone(t, tracked, "it is closed and the tracker has it")
+	// It was never asked about either: an issue that has never left this machine
+	// is not a question the tracker has an answer to.
+	if n := tr.count("GET"); n != 1 {
+		t.Errorf("%d issue(s) were asked about, want 1", n)
+	}
+
+	// Naming it explicitly does not make deleting it safe, and the reason is
+	// said out loud rather than left to be inferred from silence.
+	named := runWith(t, dir, env, "", "sync-evict", "only-copy-there-is")
+	if named.code != 0 {
+		t.Fatalf("naming a local issue exited %d:\n%s", named.code, named.out())
+	}
+	if !strings.Contains(named.stdout, "kept") || !strings.Contains(named.stdout, "IS the issue") {
+		t.Errorf("keeping it must be said out loud:\n%s", named.out())
+	}
+	wrThere(t, local, "naming it on the command line deleted it")
+}
+
+// A failed answer evicts NOTHING AT ALL — not even the issues whose answers had
+// already arrived. There is no ordering constraint between evictions, so there
+// is no reason to start before every answer is in.
+func TestATrackerFailureDuringSyncEvictEvictsNothing(t *testing.T) {
+	dir, tr, env := wrProject(t)
+	tr.add(42, "First answer", "closed")
+	tr.add(43, "Second answer", "closed")
+	tr.breaks(43)
+	answered := wrTracked(t, dir, "aaa-answered", 42, "closed")
+	unanswered := wrTracked(t, dir, "bbb-unanswered", 43, "closed")
+
+	r := runWith(t, dir, env, "", "sync-evict")
+	if r.code == 0 {
+		t.Fatalf("a tracker failure must stop the run:\n%s", r.out())
+	}
+	if !strings.Contains(r.stderr, "Nothing was evicted") {
+		t.Errorf("the failure must say what it did not do:\n%s", r.stderr)
+	}
+	wrThere(t, answered, "its answer arrived, but another one did not")
+	wrThere(t, unanswered, "the tracker never answered for it")
+}

cli/internal/cmd/tree.go

@@ -0,0 +1,164 @@
+package cmd
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+func init() {
+	register(&Command{
+		Name:  "tree",
+		Group: GroupIssue,
+		Args:  "[<id>…]",
+		Short: "draw the dependency graph of the local store",
+		Long: `Edges come from the ` + "`depends:`" + ` metadata, which is the authoritative edge list;
+prose in the body is never walked. Because the graph is slugs all the way down,
+this works identically for issues that were never pushed anywhere.
+
+Downwards is what this draws — what an issue depends on. The other direction is
+a grep, not a flag:
+
+    grep -ln 'depends:.*migrate-schema' .tea/issues/*.md`,
+		Examples: []Example{
+			{"kettle tree", "every root (nothing depends on it)"},
+			{"kettle tree wire-sqlc-appclick", "one subtree"},
+			{"kettle tree --depth 2 --write", "shallow, and saved beside the issues"},
+		},
+		Setup: func(fs *flag.FlagSet) func([]string) error {
+			depth := fs.Int("depth", 6, "maximum depth")
+			write := fs.Bool("write", false, "also write <store>/tree-<slug>.md")
+			out := storeFlag(fs)
+
+			return func(args []string) error {
+				root, err := storeRoot(*out)
+				if err != nil {
+					return err
+				}
+				if err := issue.StoreError(root); err != nil {
+					return err
+				}
+				issues, err := issue.LoadAll(root)
+				if err != nil {
+					return err
+				}
+				edges := issue.Graph(issues)
+
+				roots := args
+				for _, r := range roots {
+					if _, ok := issues[r]; !ok {
+						return Fail("no issue %q in %s", r, root)
+					}
+				}
+				if len(roots) == 0 {
+					dependedOn := map[string]bool{}
+					for _, deps := range edges {
+						for _, d := range deps {
+							dependedOn[d] = true
+						}
+					}
+					for id := range issues {
+						if !dependedOn[id] {
+							roots = append(roots, id)
+						}
+					}
+					if len(roots) == 0 { // every issue is somebody's dependency
+						for id := range issues {
+							roots = append(roots, id)
+						}
+					}
+					sort.Strings(roots)
+				}
+
+				text := renderTree(roots, issues, edges, *depth)
+				fmt.Print(text)
+				if *write {
+					slug := "all"
+					if len(roots) == 1 {
+						slug = roots[0]
+					}
+					path := filepath.Join(root, "tree-"+slug+".md")
+					if err := os.WriteFile(path, []byte(text), 0o644); err != nil {
+						return err
+					}
+					fmt.Printf("written: %s\n", path)
+				}
+				return nil
+			}
+		},
+	})
+}
+
+func renderTree(roots []string, issues map[string]*issue.Issue, edges map[string][]string, depth int) string {
+	var lines []string
+	seen := map[string]bool{}
+
+	var walk func(id, prefix string, isLast, isRoot bool, level int)
+	walk = func(id, prefix string, isLast, isRoot bool, level int) {
+		connector := ""
+		if !isRoot {
+			connector = "├── "
+			if isLast {
+				connector = "└── "
+			}
+		}
+		lines = append(lines, prefix+connector+treeLabel(id, issues, seen, edges))
+		if seen[id] || level >= depth {
+			return
+		}
+		seen[id] = true
+		kids := edges[id]
+		childPrefix := prefix
+		if !isRoot {
+			childPrefix = prefix + "│   "
+			if isLast {
+				childPrefix = prefix + "    "
+			}
+		}
+		for i, k := range kids {
+			walk(k, childPrefix, i == len(kids)-1, false, level+1)
+		}
+	}
+
+	for _, r := range roots {
+		if seen[r] {
+			continue // already drawn as somebody's child — one tree, not two
+		}
+		walk(r, "", true, true, 0)
+		lines = append(lines, "")
+	}
+
+	head := fmt.Sprintf("%d root(s)", len(roots))
+	if len(roots) == 1 {
+		head = roots[0]
+	}
+	out := fmt.Sprintf("# Dependency tree — %s\n\n```\n%s```\n", head, strings.Join(lines, "\n"))
+	if cycles := issue.FindCycles(edges); len(cycles) > 0 {
+		out += "\n## Cycles\n\n"
+		for _, c := range cycles {
+			out += "- " + strings.Join(c, " -> ") + "\n"
+		}
+	}
+	return out
+}
+
+func treeLabel(id string, issues map[string]*issue.Issue, seen map[string]bool, edges map[string][]string) string {
+	i, ok := issues[id]
+	if !ok {
+		return id + " (not in the store)"
+	}
+	tail := ""
+	if seen[id] && len(edges[id]) > 0 {
+		tail = " (see above)"
+	}
+	typ := i.Type()
+	if typ == "" {
+		typ = "-"
+	}
+	return fmt.Sprintf("%s [%s] %s — %s    %s.md%s", id, typ, i.Title, i.State, id, tail)
+}

cli/internal/config/config.go

@@ -0,0 +1,326 @@
+// Package config holds the two files kettle reads: what this project is, and
+// who this machine is.
+//
+// The split is the whole design. `<project>/.kettle/config.yaml` says which
+// tracker repository the issues belong to and which login to reach it under —
+// facts about the project, written by `kettle init`. The credentials themselves
+// live in one file per machine, outside any repository, mode 0600.
+//
+// A token in a file inside a working tree ends up in a commit. Not always, not
+// immediately, and not by anyone careless — but a project config is exactly the
+// file somebody eventually decides to share, and a secret that has ever been
+// pushed is a secret that has to be rotated. So the project pins a login by
+// NAME and the name is worth nothing on its own.
+package config
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gopkg.in/yaml.v3"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+)
+
+// Environment overrides, each winning over the file it shadows. They exist for
+// CI, for a one-off run against another instance, and for anyone who would
+// rather not have a token on disk at all.
+const (
+	EnvLogin = "KETTLE_LOGIN"
+	EnvURL   = "KETTLE_URL"
+	EnvToken = "KETTLE_TOKEN"
+	EnvRepo  = "KETTLE_REPO"
+	// EnvHome relocates the machine-wide login file; the test suite sets it so
+	// a run can never read or write the developer's own.
+	EnvHome = "KETTLE_CONFIG_HOME"
+)
+
+const projectHeader = `# kettle — project configuration
+#
+# login  the name of a login in the machine-wide file, NOT a credential.
+#        Manage those with ` + "`kettle auth`" + `; they live outside this tree.
+# repo   the tracker repository these issues belong to, as owner/name.
+#
+# Overrides, when you need one: ` + EnvLogin + `, ` + EnvRepo + `, ` + EnvURL + `, ` + EnvToken + `.
+`
+
+// Project is `<project>/.kettle/config.yaml`.
+type Project struct {
+	// Login names an entry in the machine-wide login file. Never a token.
+	Login string `yaml:"login"`
+	// Repo is the tracker repository, as owner/name.
+	Repo string `yaml:"repo"`
+}
+
+// Login is one set of credentials for one Gitea instance.
+type Login struct {
+	Name  string `yaml:"name"`
+	URL   string `yaml:"url"`
+	User  string `yaml:"user,omitempty"`
+	Token string `yaml:"token"`
+}
+
+// Logins is the machine-wide file.
+type Logins struct {
+	Logins []Login `yaml:"logins"`
+}
+
+// ErrNoConfig means the project has no config.yaml yet.
+var ErrNoConfig = errors.New("no project configuration")
+
+// ProjectPath is where this project's config.yaml is, or "" with no project.
+func ProjectPath(start string) string { return project.ConfigPath(start) }
+
+// LoadProject reads the project configuration.
+//
+// A missing file is ErrNoConfig, not an empty config: "this project has not
+// been told which tracker it belongs to" and "it belongs to no tracker" are
+// different answers and only one of them is fixable by running init.
+func LoadProject(start string) (*Project, error) {
+	path := ProjectPath(start)
+	if path == "" {
+		return nil, project.NotFoundError(start)
+	}
+	raw, err := os.ReadFile(path)
+	if os.IsNotExist(err) {
+		return nil, fmt.Errorf("%w at %s — run `kettle init` there", ErrNoConfig, path)
+	}
+	if err != nil {
+		return nil, err
+	}
+	var p Project
+	if err := strictUnmarshal(raw, &p); err != nil {
+		return nil, fmt.Errorf("%s: %w", path, err)
+	}
+	return &p, nil
+}
+
+// ReadProjectFile reads a config.yaml at a path already known, reporting
+// whether the file was there.
+//
+// LoadProject resolves the path by walking for a marker, which is the right
+// thing everywhere except inside `kettle init` — the command that is creating
+// the marker, and on a dry run may not have created it at all.
+func ReadProjectFile(path string) (*Project, bool, error) {
+	raw, err := os.ReadFile(path)
+	if os.IsNotExist(err) {
+		return &Project{}, false, nil
+	}
+	if err != nil {
+		return nil, false, err
+	}
+	var p Project
+	if err := strictUnmarshal(raw, &p); err != nil {
+		return nil, true, fmt.Errorf("%s: %w", path, err)
+	}
+	return &p, true, nil
+}
+
+// SaveProject writes the project configuration, header comment and all.
+func SaveProject(path string, p *Project) error {
+	body, err := yaml.Marshal(p)
+	if err != nil {
+		return err
+	}
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		return err
+	}
+	return os.WriteFile(path, append([]byte(projectHeader+"\n"), body...), 0o644)
+}
+
+// LoginsPath is the machine-wide login file.
+//
+// One file per machine, deliberately outside every working tree: which tokens
+// this computer holds is a fact about the computer, the way which issues a tree
+// holds is a fact about the tree.
+func LoginsPath() string {
+	if h := os.Getenv(EnvHome); h != "" {
+		return filepath.Join(h, "logins.yaml")
+	}
+	if x := os.Getenv("XDG_CONFIG_HOME"); x != "" {
+		return filepath.Join(x, "kettle", "logins.yaml")
+	}
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return ""
+	}
+	return filepath.Join(home, ".config", "kettle", "logins.yaml")
+}
+
+// LoadLogins reads the machine-wide login file. A missing file is an empty
+// list, not an error: a machine with no logins yet is an ordinary machine.
+func LoadLogins() (*Logins, error) {
+	path := LoginsPath()
+	if path == "" {
+		return &Logins{}, nil
+	}
+	raw, err := os.ReadFile(path)
+	if os.IsNotExist(err) {
+		return &Logins{}, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	var l Logins
+	if err := strictUnmarshal(raw, &l); err != nil {
+		return nil, fmt.Errorf("%s: %w", path, err)
+	}
+	return &l, nil
+}
+
+// SaveLogins writes the machine-wide login file with 0600, and creates its
+// directory with 0700. The file holds bearer tokens; nothing else on the
+// machine has any business reading it.
+func SaveLogins(l *Logins) error {
+	path := LoginsPath()
+	if path == "" {
+		return errors.New("cannot locate a home directory for the login file — set " + EnvHome)
+	}
+	if err := os.MkdirAll(filepath.Dir(path), 0o700); err != nil {
+		return err
+	}
+	body, err := yaml.Marshal(l)
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(path, body, 0o600)
+}
+
+// Find returns the login with this name.
+func (l *Logins) Find(name string) *Login {
+	for i := range l.Logins {
+		if l.Logins[i].Name == name {
+			return &l.Logins[i]
+		}
+	}
+	return nil
+}
+
+// Names lists every login on this machine, for an error message that can
+// actually be acted on.
+func (l *Logins) Names() []string {
+	out := make([]string, 0, len(l.Logins))
+	for _, e := range l.Logins {
+		out = append(out, e.Name)
+	}
+	return out
+}
+
+// Resolved is everything the transport needs, with every override applied.
+type Resolved struct {
+	Login string
+	URL   string
+	Token string
+	Owner string
+	Repo  string
+}
+
+// Slug is owner/name, the way a tracker writes it.
+func (r *Resolved) Slug() string { return r.Owner + "/" + r.Repo }
+
+// Redacted is the same thing with the token replaced, for printing.
+func (r *Resolved) Redacted() Resolved {
+	out := *r
+	if out.Token != "" {
+		out.Token = "(set)"
+	}
+	return out
+}
+
+// Resolve merges the project config, the machine's login file, and the
+// environment into what the transport needs.
+//
+// Every failure names the file it read and the command that fixes it. "401
+// Unauthorized" is what happens when this function is allowed to return a
+// half-filled struct.
+func Resolve(start string) (*Resolved, error) {
+	var p Project
+	if loaded, err := LoadProject(start); err == nil {
+		p = *loaded
+	} else if !errors.Is(err, ErrNoConfig) {
+		return nil, err
+	}
+
+	out := &Resolved{Login: p.Login}
+	if v := os.Getenv(EnvLogin); v != "" {
+		out.Login = v
+	}
+
+	repo := p.Repo
+	if v := os.Getenv(EnvRepo); v != "" {
+		repo = v
+	}
+	if repo != "" {
+		owner, name, ok := strings.Cut(repo, "/")
+		if !ok || owner == "" || name == "" {
+			return nil, fmt.Errorf("repo %q is not owner/name", repo)
+		}
+		out.Owner, out.Repo = owner, name
+	}
+
+	if out.Login != "" {
+		logins, err := LoadLogins()
+		if err != nil {
+			return nil, err
+		}
+		entry := logins.Find(out.Login)
+		if entry == nil {
+			known := "none on this machine"
+			if names := logins.Names(); len(names) > 0 {
+				known = strings.Join(names, ", ")
+			}
+			return nil, fmt.Errorf("no login %q in %s — known: %s; add one with `kettle auth add`",
+				out.Login, LoginsPath(), known)
+		}
+		out.URL, out.Token = entry.URL, entry.Token
+	}
+
+	if v := os.Getenv(EnvURL); v != "" {
+		out.URL = v
+	}
+	if v := os.Getenv(EnvToken); v != "" {
+		out.Token = v
+	}
+	out.URL = strings.TrimRight(out.URL, "/")
+	return out, nil
+}
+
+// Require is Resolve plus the assertion that the result can actually reach a
+// tracker.
+func Require(start string) (*Resolved, error) {
+	r, err := Resolve(start)
+	if err != nil {
+		return nil, err
+	}
+	var missing []string
+	if r.URL == "" {
+		missing = append(missing, "a URL (pin a login with `kettle init --login`, or set "+EnvURL+")")
+	}
+	if r.Token == "" {
+		missing = append(missing, "a token (`kettle auth add`, or set "+EnvToken+")")
+	}
+	if r.Owner == "" {
+		missing = append(missing, "a repository (`kettle init --repo owner/name`, or set "+EnvRepo+")")
+	}
+	if len(missing) > 0 {
+		return nil, fmt.Errorf("this project has no %s", strings.Join(missing, ", and no "))
+	}
+	return r, nil
+}
+
+// strictUnmarshal refuses keys the struct does not know.
+//
+// The alternative is silence: an older binary reading a newer config would drop
+// the setting it did not recognize, and rewriting the file would delete it.
+// Being told "unknown field" beats finding out later.
+func strictUnmarshal(raw []byte, out any) error {
+	dec := yaml.NewDecoder(strings.NewReader(string(raw)))
+	dec.KnownFields(true)
+	if err := dec.Decode(out); err != nil && err.Error() != "EOF" {
+		return err
+	}
+	return nil
+}

cli/internal/gitea/client.go

@@ -0,0 +1,381 @@
+// Package gitea is the transport: everything that talks to a tracker, and
+// nothing else.
+//
+// It knows numbers, logins, HTTP verbs, pagination and JSON. It does not know
+// what an issue IS — no sections, no acceptance criteria, no type taxonomy —
+// and the import graph says so in both directions: this package may not reach
+// into internal/issue, and internal/issue may not reach in here. A tracker
+// number is not a domain concept and a checkbox is not a transport one.
+// Translating between the two is a layer of its own — internal/mapping — and
+// that layer is not imported here either: it sits above this package, not
+// beside it.
+//
+// The JSON shapes and the issue keys are internal/wire's. They are not this
+// package's to own, because the bridge needs exactly the same vocabulary and
+// cannot import a transport to get it; a copy on each side is two structs that
+// drift and a command that copies fields between them by hand.
+//
+// Every request goes through Call. One place sets the header, one place reads
+// a status code, one place files the request body. When this was a Python
+// module shelling out to `tea api`, "why did that fail" meant reading a
+// subprocess's stderr and guessing; here a failure is an *APIError carrying the
+// status AND the body the server actually sent, because "500" on its own has
+// never helped anybody.
+package gitea
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+const (
+	// apiPrefix is where every Gitea instance puts its REST API.
+	apiPrefix = "/api/v1"
+	// userAgent names this binary in the server's log. A tracker admin looking
+	// at a burst of requests should be able to tell what made them.
+	userAgent = "kettle"
+	// requestTimeout bounds a single call. A hung tracker must not hang a push
+	// half way through a set of issues.
+	requestTimeout = 30 * time.Second
+	// maxErrorBody caps what an error quotes back. A server having a bad day
+	// answers with an HTML page, and an error message is not a place to paste
+	// one.
+	maxErrorBody = 2000
+)
+
+// Client talks to one repository on one Gitea instance.
+type Client struct {
+	// HTTP is the transport, exported so a caller can change the timeout or
+	// hand in an instrumented one. Never nil after New.
+	HTTP *http.Client
+
+	base  string // instance URL with the API prefix, no trailing slash
+	token string
+	repo  wire.Repo
+
+	// payloadRoot is resolved once, by New, and is never taken from a caller.
+	// The one time where a request body lands was an argument, it got pointed
+	// at the issue store — see writePayload.
+	payloadRoot string
+}
+
+// New builds a client for the repository this project points at.
+//
+// It refuses a half-filled configuration instead of letting the first call come
+// back 401 or 404: those answers name nothing an operator can act on, and every
+// field missing here has exactly one command that supplies it.
+func New(cfg *config.Resolved) (*Client, error) {
+	if cfg == nil {
+		return nil, errors.New("no resolved configuration — call config.Require first")
+	}
+	var missing []string
+	if cfg.URL == "" {
+		missing = append(missing, "a URL (pin a login with `kettle init --login`, or set "+config.EnvURL+")")
+	}
+	if cfg.Token == "" {
+		missing = append(missing, "a token (`kettle auth add`, or set "+config.EnvToken+")")
+	}
+	if cfg.Owner == "" || cfg.Repo == "" {
+		missing = append(missing, "a repository (`kettle init --repo owner/name`, or set "+config.EnvRepo+")")
+	}
+	if len(missing) > 0 {
+		return nil, fmt.Errorf("this project has no %s", strings.Join(missing, ", and no "))
+	}
+	return &Client{
+		HTTP:        &http.Client{Timeout: requestTimeout},
+		base:        strings.TrimRight(cfg.URL, "/") + apiPrefix,
+		token:       cfg.Token,
+		repo:        wire.Repo{Owner: cfg.Owner, Name: cfg.Repo},
+		payloadRoot: project.PayloadRoot(""),
+	}, nil
+}
+
+// Repo is the repository every path is built against.
+func (c *Client) Repo() wire.Repo { return c.repo }
+
+// For returns a copy of this client pointed at another repository, for the run
+// that was given an explicit owner/name. The credentials and the scratchpad
+// come along; only the paths change.
+func (c *Client) For(r wire.Repo) *Client {
+	out := *c
+	out.repo = r
+	return &out
+}
+
+// Body is a request payload and the name its dump is filed under.
+//
+// The name is the caller's label for this call, not a path: it becomes
+// `<name>.json` in the scratchpad, and something that identifies the call in a
+// post-mortem — an issue's slug, a label's name — is worth more there than a
+// serial number.
+type Body struct {
+	Name string
+	Data any
+}
+
+// Call makes one request and decodes the answer into out, which may be nil when
+// there is nothing to read.
+//
+// body may be nil. When it is not, its Data is marshalled once: the bytes filed
+// in the scratchpad and the bytes on the wire are the same bytes, so a retry
+// from the file sends what this call sent.
+//
+// An empty response body leaves out untouched — a 204 from a PATCH is a
+// success, not a decode failure.
+func (c *Client) Call(method, path string, body *Body, out any) error {
+	var payload []byte
+	if body != nil {
+		var err error
+		if payload, err = c.writePayload(body); err != nil {
+			return err
+		}
+	}
+
+	endpoint := c.base + "/" + strings.TrimLeft(path, "/")
+	var reader io.Reader
+	if payload != nil {
+		reader = bytes.NewReader(payload)
+	}
+	req, err := http.NewRequest(method, endpoint, reader)
+	if err != nil {
+		return fmt.Errorf("%s %s: %w", method, endpoint, err)
+	}
+	// Gitea's own scheme, and what the `tea` CLI this replaces sent: the word
+	// `token`, not `Bearer`. An instance answers 401 to the other spelling.
+	req.Header.Set("Authorization", "token "+c.token)
+	req.Header.Set("Accept", "application/json")
+	req.Header.Set("User-Agent", userAgent)
+	if payload != nil {
+		req.Header.Set("Content-Type", "application/json")
+	}
+
+	resp, err := c.HTTP.Do(req)
+	if err != nil {
+		// The token travels in a header and never in the URL, so an error is
+		// free to quote the URL in full.
+		return fmt.Errorf("%s %s: %w", method, endpoint, err)
+	}
+	defer resp.Body.Close()
+
+	raw, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return fmt.Errorf("%s %s: reading the response: %w", method, endpoint, err)
+	}
+	if resp.StatusCode < 200 || resp.StatusCode > 299 {
+		return &APIError{Method: method, URL: endpoint, Status: resp.StatusCode, Body: string(raw)}
+	}
+	if out == nil || len(bytes.TrimSpace(raw)) == 0 {
+		return nil
+	}
+	if err := json.Unmarshal(raw, out); err != nil {
+		return fmt.Errorf("%s %s: %d answered with what is not the JSON expected (%w): %s",
+			method, endpoint, resp.StatusCode, err, truncate(string(raw)))
+	}
+	return nil
+}
+
+// APIError is a non-2xx answer, carrying both halves of what happened.
+//
+// The status on its own is not a diagnosis. Gitea answers 422 for a label that
+// already exists, for a milestone id that belongs to another repository, and
+// for a body missing a field, and the three are told apart only by the message
+// sent with them — so the body travels with the code, always.
+type APIError struct {
+	Method string
+	URL    string
+	Status int
+	Body   string
+}
+
+func (e *APIError) Error() string {
+	body := strings.TrimSpace(e.Body)
+	if body == "" {
+		body = "(the response body was empty)"
+	} else {
+		body = truncate(body)
+	}
+	status := http.StatusText(e.Status)
+	if status != "" {
+		status = " " + status
+	}
+	return fmt.Sprintf("%s %s: %d%s: %s", e.Method, e.URL, e.Status, status, body)
+}
+
+// StatusIs reports whether err is an API answer with this status code, for the
+// handful of places where one code means something specific — a 409 from a
+// dependency link that is already there, say.
+func StatusIs(err error, status int) bool {
+	var apiErr *APIError
+	return errors.As(err, &apiErr) && apiErr.Status == status
+}
+
+func truncate(s string) string {
+	if len(s) <= maxErrorBody {
+		return s
+	}
+	cut := s[:maxErrorBody]
+	// Never split a rune: a truncated message that ends in a broken byte is a
+	// message a terminal renders as garbage.
+	for len(cut) > 0 && !utf8.ValidString(cut) {
+		cut = cut[:len(cut)-1]
+	}
+	return fmt.Sprintf("%s… (%d bytes total)", cut, len(s))
+}
+
+// --------------------------------------------------------------------------
+// where request bodies land
+// --------------------------------------------------------------------------
+
+// writePayload marshals a request body, files a copy under `.kettle/payload/`,
+// and returns the bytes to send.
+//
+// The file survives the call, for a retry or a post-mortem.
+//
+// WHERE IT LANDS IS NOT THE CALLER'S BUSINESS, and never was. The directory is
+// this package's scratchpad — a SIBLING of the issue store under the same
+// marker, resolved by the same walk, so which command wrote a body cannot
+// change where it went and the two can never end up in different projects. The
+// one time it was a caller's argument it got pointed at the store, and a label
+// bootstrap that touches no issue at all materialized an issue directory on a
+// fresh checkout: store contents are the thing being tracked, request bodies
+// are debris of the transport, and when they share a path `ls` starts lying
+// about what the project holds.
+//
+// It is created lazily, by the first write of a run and only then, so a dry run
+// or a run with nothing to send leaves no directory behind.
+func (c *Client) writePayload(b *Body) ([]byte, error) {
+	if c.payloadRoot == "" {
+		return nil, project.NotFoundError("")
+	}
+
+	buf := &bytes.Buffer{}
+	enc := json.NewEncoder(buf)
+	enc.SetIndent("", "  ")
+	// An issue body carries `<!-- … -->` markers and prose full of `&`.
+	// Escaping those to < would make the dump unreadable exactly when
+	// somebody is reading it because something went wrong.
+	enc.SetEscapeHTML(false)
+	if err := enc.Encode(b.Data); err != nil {
+		return nil, fmt.Errorf("encoding the %s request body: %w", b.name(), err)
+	}
+	raw := buf.Bytes()
+
+	if err := os.MkdirAll(c.payloadRoot, 0o755); err != nil {
+		return nil, err
+	}
+	path := filepath.Join(c.payloadRoot, b.name()+".json")
+	if err := os.WriteFile(path, raw, 0o644); err != nil {
+		return nil, err
+	}
+	return raw, nil
+}
+
+// name is the file stem, with everything that is not plainly a file name folded
+// away.
+//
+// Sanitizing here rather than trusting callers: label names are namespaced
+// (`type/bug`), and a name passed straight through would write outside the
+// scratchpad — which is the one thing this directory exists to prevent.
+func (b *Body) name() string {
+	if b.Name == "" {
+		return "request"
+	}
+	safe := strings.Map(func(r rune) rune {
+		switch {
+		case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9', r == '-', r == '_':
+			return r
+		}
+		return '-'
+	}, b.Name)
+	if safe = strings.Trim(safe, "-"); safe == "" {
+		return "request"
+	}
+	return safe
+}
+
+// --------------------------------------------------------------------------
+// pagination
+// --------------------------------------------------------------------------
+
+const (
+	// pageLimit is how many rows a list request asks for at a time. Gitea's own
+	// default is smaller and its maximum is larger; 50 is what the Python this
+	// replaces used and what the page-budget arithmetic is written against.
+	pageLimit = 50
+	// maxPages bounds any single listing. A tracker with a runaway number of
+	// rows must not turn one command into an unbounded read.
+	maxPages = 40
+	// PageSlack is how far past the ideal page count a Keep-bounded listing may
+	// scan before it gives up. The ideal is what Limit would need if every
+	// payload counted; the slack pays for the ones that do not. Deliberately
+	// small: "fetch until N are kept" without a bound is "fetch the whole
+	// tracker" on any repository whose filter matches mostly closed issues.
+	PageSlack = 4
+)
+
+// pages GETs a list endpoint page by page and hands each page to each as it
+// arrives, stopping when each returns false, when a short page says the list is
+// exhausted, or when budget pages have been read.
+//
+// A callback rather than a slice, because a caller whose budget is spent on
+// what it KEEPS cannot be served by a function that fetches everything first:
+// the page after the one that completed the budget must never be requested.
+func pages[T any](c *Client, path string, limit, budget int, each func([]T) (bool, error)) error {
+	sep := "?"
+	if strings.Contains(path, "?") {
+		sep = "&"
+	}
+	for page := 1; page <= budget; page++ {
+		var batch []T
+		if err := c.Call(http.MethodGet, fmt.Sprintf("%s%spage=%d&limit=%d", path, sep, page, limit), nil, &batch); err != nil {
+			return err
+		}
+		if len(batch) == 0 {
+			return nil
+		}
+		more, err := each(batch)
+		if err != nil || !more {
+			return err
+		}
+		if len(batch) < limit {
+			return nil // a short page is the last one
+		}
+	}
+	return nil
+}
+
+// paginate follows a list endpoint to exhaustion and returns the whole list.
+func paginate[T any](c *Client, path string, limit int) ([]T, error) {
+	var out []T
+	err := pages(c, path, limit, maxPages, func(batch []T) (bool, error) {
+		out = append(out, batch...)
+		return true, nil
+	})
+	return out, err
+}
+
+// repoPath builds an endpoint under this client's repository. Owner and name
+// are escaped: they arrive from a config file, and a file is a thing people
+// type into.
+func (c *Client) repoPath(suffix string) string {
+	return "repos/" + url.PathEscape(c.repo.Owner) + "/" + url.PathEscape(c.repo.Name) + "/" + suffix
+}
+
+// repoPathf is repoPath with the issue or label number formatted in.
+func (c *Client) repoPathf(format string, args ...any) string {
+	return c.repoPath(fmt.Sprintf(format, args...))
+}

cli/internal/gitea/client_test.go

@@ -0,0 +1,396 @@
+package gitea_test
+
+// The transport is tested against httptest, never against a tracker: a test
+// that needs a server somewhere is a test nobody runs.
+//
+// Every fixture builds a throwaway project in a temp directory and points the
+// project walk at it with CLAUDE_PROJECT_DIR. Without that the walk falls
+// through to the working directory — which during a test run is this repository
+// — and a request dump would land in the developer's own project. Nothing here
+// reads a login file either, but KETTLE_CONFIG_HOME is redirected all the same,
+// so a run can neither read nor overwrite the developer's own tokens.
+
+import (
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/config"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// newProject makes an initialized project and points the walk at it. Returns
+// the project root.
+func newProject(t *testing.T) string {
+	t.Helper()
+	dir := t.TempDir()
+	if err := os.MkdirAll(filepath.Join(dir, ".kettle"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	t.Setenv("CLAUDE_PROJECT_DIR", dir)
+	t.Setenv(config.EnvHome, filepath.Join(dir, "config"))
+	return dir
+}
+
+func newClient(t *testing.T, url string) *gitea.Client {
+	t.Helper()
+	c, err := gitea.New(&config.Resolved{URL: url, Token: "s3cret", Owner: "acme", Repo: "widgets"})
+	if err != nil {
+		t.Fatalf("New: %v", err)
+	}
+	return c
+}
+
+func writeJSON(t *testing.T, w http.ResponseWriter, v any) {
+	t.Helper()
+	w.Header().Set("Content-Type", "application/json")
+	if err := json.NewEncoder(w).Encode(v); err != nil {
+		t.Errorf("encoding the fake response: %v", err)
+	}
+}
+
+// A list endpoint is followed to the last page and no further: the short page
+// ends it, and the page after that is never asked for.
+func TestPaginationFollowsToTheLastPage(t *testing.T) {
+	newProject(t)
+
+	var asked []string
+	var auth string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		asked = append(asked, r.URL.RequestURI())
+		auth = r.Header.Get("Authorization")
+
+		page, _ := strconv.Atoi(r.URL.Query().Get("page"))
+		limit, _ := strconv.Atoi(r.URL.Query().Get("limit"))
+		n := limit
+		if page == 3 {
+			n = 7 // the short page
+		} else if page > 3 {
+			n = 0
+		}
+		out := []map[string]any{}
+		for i := 0; i < n; i++ {
+			out = append(out, map[string]any{"id": (page-1)*limit + i + 1, "body": "hello"})
+		}
+		writeJSON(t, w, out)
+	}))
+	defer srv.Close()
+
+	got, err := newClient(t, srv.URL).ListComments(42)
+	if err != nil {
+		t.Fatalf("ListComments: %v", err)
+	}
+	if len(got) != 107 {
+		t.Errorf("got %d comments, want 107 (50 + 50 + 7)", len(got))
+	}
+	if len(asked) != 3 {
+		t.Errorf("made %d requests (%v), want 3 — a short page is the last one", len(asked), asked)
+	}
+	if got[0].ID != 1 || got[106].ID != 107 {
+		t.Errorf("pages arrived out of order: first %d, last %d", got[0].ID, got[106].ID)
+	}
+	// Gitea's own scheme, and what the CLI this replaces sent.
+	if auth != "token s3cret" {
+		t.Errorf("Authorization was %q, want %q", auth, "token s3cret")
+	}
+	want := "/api/v1/repos/acme/widgets/issues/42/comments?page=1&limit=50"
+	if asked[0] != want {
+		t.Errorf("first request was %s, want %s", asked[0], want)
+	}
+}
+
+// A non-2xx carries the status AND the body, because the status alone has never
+// told anybody which of the four things that answer 422 actually happened.
+func TestErrorNamesTheStatusAndTheBody(t *testing.T) {
+	newProject(t)
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusUnprocessableEntity)
+		io.WriteString(w, `{"message":"label already exists","url":"https://example.test/docs"}`)
+	}))
+	defer srv.Close()
+
+	_, err := newClient(t, srv.URL).GetIssue(7)
+	if err == nil {
+		t.Fatal("a 422 returned no error")
+	}
+	var apiErr *gitea.APIError
+	if !errors.As(err, &apiErr) {
+		t.Fatalf("error is %T, want *gitea.APIError: %v", err, err)
+	}
+	if apiErr.Status != http.StatusUnprocessableEntity {
+		t.Errorf("Status is %d, want 422", apiErr.Status)
+	}
+	if !gitea.StatusIs(err, http.StatusUnprocessableEntity) {
+		t.Error("StatusIs did not recognize its own error")
+	}
+	for _, want := range []string{"422", "label already exists", "GET", "/api/v1/repos/acme/widgets/issues/7"} {
+		if !strings.Contains(err.Error(), want) {
+			t.Errorf("the error does not mention %q:\n%s", want, err)
+		}
+	}
+	// The token is in a header, so quoting the URL is safe — and it had better
+	// stay that way.
+	if strings.Contains(err.Error(), "s3cret") {
+		t.Errorf("the error quotes the token:\n%s", err)
+	}
+}
+
+// A request body is filed in the scratchpad, which is a SIBLING of the store
+// and never inside it. A call that touches no issue must not materialize an
+// issue directory.
+func TestPayloadLandsBesideTheStoreAndNeverInIt(t *testing.T) {
+	root := newProject(t)
+
+	var sent []byte
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		sent, _ = io.ReadAll(r.Body)
+		writeJSON(t, w, map[string]any{
+			"number": 42, "id": 5, "html_url": "https://example.test/acme/widgets/issues/42"})
+	}))
+	defer srv.Close()
+
+	body := "<!-- kettle:id wire-sqlc --> a & b"
+	got, err := newClient(t, srv.URL).CreateIssue(
+		wire.IssueRequest{Title: wire.Set("wire sqlc"), Body: wire.Set(body)}, "issue-wire-sqlc")
+	if err != nil {
+		t.Fatalf("CreateIssue: %v", err)
+	}
+	if got.Number != 42 {
+		t.Errorf("got issue #%d, want #42", got.Number)
+	}
+
+	path := filepath.Join(root, ".kettle", "payload", "issue-wire-sqlc.json")
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("the request body was not filed at %s: %v", path, err)
+	}
+	if string(raw) != string(sent) {
+		t.Errorf("the filed body is not the body that was sent:\nfiled: %s\nsent:  %s", raw, sent)
+	}
+	// A dump escaped to \u003c is unreadable exactly when it is being read.
+	if !strings.Contains(string(raw), "<!-- kettle:id wire-sqlc --> a & b") {
+		t.Errorf("the dump escaped the markup it was meant to preserve:\n%s", raw)
+	}
+	// The whole reason the scratchpad is a sibling.
+	if _, err := os.Stat(filepath.Join(root, ".kettle", "issues")); !os.IsNotExist(err) {
+		t.Errorf("writing a request body materialized the issue store (%v)", err)
+	}
+	// A namespaced name must not climb out of the scratchpad.
+	if _, err := newClient(t, srv.URL).CreateLabel(wire.LabelRequest{Name: "type/bug", Color: "#ee0701"}); err != nil {
+		t.Fatalf("CreateLabel: %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(root, ".kettle", "payload", "label-type-bug.json")); err != nil {
+		t.Errorf("a label request body was not filed under a safe name: %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(root, ".kettle", "payload", "type")); !os.IsNotExist(err) {
+		t.Error("a label name with a slash in it made a directory inside the scratchpad")
+	}
+}
+
+// A run that sends no body leaves no directory behind — the scratchpad is
+// created by the first write and only then.
+func TestAReadOnlyCallCreatesNoScratchpad(t *testing.T) {
+	root := newProject(t)
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		writeJSON(t, w, map[string]any{"number": 42})
+	}))
+	defer srv.Close()
+
+	if _, err := newClient(t, srv.URL).GetIssue(42); err != nil {
+		t.Fatalf("GetIssue: %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(root, ".kettle", "payload")); !os.IsNotExist(err) {
+		t.Errorf("a read created the payload directory (%v)", err)
+	}
+}
+
+// The milestone filter is re-checked on the client, because Gitea silently
+// ignores one it cannot resolve and answers with the whole backlog. Pull
+// requests go the same way.
+func TestListIssuesRechecksWhatTheServerIgnored(t *testing.T) {
+	newProject(t)
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if strings.Contains(r.URL.Path, "/milestones") {
+			writeJSON(t, w, []map[string]any{{"id": 3, "title": "v1"}, {"id": 9, "title": "later"}})
+			return
+		}
+		if r.URL.Query().Get("page") != "1" {
+			writeJSON(t, w, []map[string]any{})
+			return
+		}
+		writeJSON(t, w, []map[string]any{
+			{"number": 1, "title": "in the milestone", "milestone": map[string]any{"id": 3, "title": "v1"}},
+			{"number": 2, "title": "another milestone", "milestone": map[string]any{"id": 9, "title": "later"}},
+			{"number": 3, "title": "no milestone at all"},
+			{"number": 4, "title": "a pull request", "milestone": map[string]any{"id": 3, "title": "v1"},
+				"pull_request": map[string]any{"merged": false}},
+		})
+	}))
+	defer srv.Close()
+
+	got, err := newClient(t, srv.URL).ListIssues(gitea.IssueFilter{Milestone: "v1", Limit: 50})
+	if err != nil {
+		t.Fatalf("ListIssues: %v", err)
+	}
+	if got.Milestone != "v1" {
+		t.Errorf("resolved milestone is %q, want v1", got.Milestone)
+	}
+	if len(got.Issues) != 1 || got.Issues[0].Number != 1 {
+		t.Fatalf("got %d issue(s) %v, want only #1 — the backlog was not re-filtered", len(got.Issues), got.Issues)
+	}
+
+	if _, err := newClient(t, srv.URL).ListIssues(gitea.IssueFilter{Milestone: "typo", Limit: 50}); err == nil {
+		t.Error("an unknown milestone was accepted — that reads as a milestone with the whole backlog in it")
+	} else if !strings.Contains(err.Error(), "have: v1 (id 3)") {
+		t.Errorf("the error does not say what the repo actually has: %v", err)
+	}
+}
+
+// A Keep predicate that rejects everything must not turn a bounded read into a
+// walk of the whole tracker, and coming up short is reported rather than
+// answered in silence.
+func TestListIssuesStopsAtThePageBudget(t *testing.T) {
+	newProject(t)
+
+	pages := 0
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		pages++
+		limit, _ := strconv.Atoi(r.URL.Query().Get("limit"))
+		out := []map[string]any{}
+		for i := 0; i < limit; i++ {
+			out = append(out, map[string]any{"number": pages*100 + i, "state": "closed"})
+		}
+		writeJSON(t, w, out)
+	}))
+	defer srv.Close()
+
+	got, err := newClient(t, srv.URL).ListIssues(gitea.IssueFilter{
+		Limit: 2,
+		Keep:  func(i *wire.Issue) bool { return i.State == "open" },
+	})
+	if err != nil {
+		t.Fatalf("ListIssues: %v", err)
+	}
+	if pages != gitea.PageSlack {
+		t.Errorf("read %d page(s), want %d — one ideal page times the slack", pages, gitea.PageSlack)
+	}
+	if got.Warning == "" {
+		t.Error("stopped short of the limit and said nothing about it")
+	}
+	// Everything enumerated comes back even though none of it counted: a caller
+	// with something to say about the ones that did not still can.
+	if len(got.Issues) != gitea.PageSlack*2 {
+		t.Errorf("got %d issue(s), want every payload that was enumerated", len(got.Issues))
+	}
+}
+
+// A Keep-bounded read stops the moment the budget is full: the page after the
+// one that completed it is never requested.
+func TestListIssuesStopsAtTheLimit(t *testing.T) {
+	newProject(t)
+
+	pages := 0
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		pages++
+		limit, _ := strconv.Atoi(r.URL.Query().Get("limit"))
+		out := []map[string]any{}
+		for i := 0; i < limit; i++ {
+			out = append(out, map[string]any{"number": pages*100 + i, "state": "open"})
+		}
+		writeJSON(t, w, out)
+	}))
+	defer srv.Close()
+
+	got, err := newClient(t, srv.URL).ListIssues(gitea.IssueFilter{
+		Limit: 2,
+		Keep:  func(i *wire.Issue) bool { return i.State == "open" },
+	})
+	if err != nil {
+		t.Fatalf("ListIssues: %v", err)
+	}
+	if pages != 1 {
+		t.Errorf("read %d page(s), want 1 — the budget was full after the first", pages)
+	}
+	if len(got.Issues) != 2 || got.Warning != "" {
+		t.Errorf("got %d issue(s), warning %q; want 2 and no warning", len(got.Issues), got.Warning)
+	}
+}
+
+// A dependency endpoint the instance does not have is "no dependencies", not a
+// failed pull. A dead connection still is one.
+func TestDependenciesToleratesAnInstanceWithoutThem(t *testing.T) {
+	newProject(t)
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		http.Error(w, "not implemented", http.StatusNotImplemented)
+	}))
+	defer srv.Close()
+
+	got, err := newClient(t, srv.URL).DependencyKeys(42)
+	if err != nil || len(got) != 0 {
+		t.Errorf("DependencyKeys = %v, %v; want no keys and no error", got, err)
+	}
+
+	srv.Close()
+	if _, err := newClient(t, srv.URL).Dependencies(42); err == nil {
+		t.Error("a dead connection was reported as an instance without dependency support")
+	}
+}
+
+// A half-filled configuration is refused here rather than at the first 401,
+// because a 401 names nothing an operator can act on.
+func TestNewRefusesAHalfFilledConfiguration(t *testing.T) {
+	for _, tc := range []struct {
+		what string
+		cfg  config.Resolved
+		want string
+	}{
+		{"no url", config.Resolved{Token: "t", Owner: "a", Repo: "b"}, config.EnvURL},
+		{"no token", config.Resolved{URL: "u", Owner: "a", Repo: "b"}, "kettle auth add"},
+		{"no repo", config.Resolved{URL: "u", Token: "t"}, "kettle init --repo"},
+	} {
+		_, err := gitea.New(&tc.cfg)
+		if err == nil {
+			t.Errorf("%s: accepted", tc.what)
+			continue
+		}
+		if !strings.Contains(err.Error(), tc.want) {
+			t.Errorf("%s: the error does not name the fix (%q): %v", tc.what, tc.want, err)
+		}
+	}
+}
+
+// The layering rule, from this side. The transport knows numbers, logins, HTTP
+// and JSON; the domain knows none of those, and neither may reach the other.
+//
+// The bridge is out too, and for a reason of its own: it is the layer that
+// translates between the two, so it sits ABOVE both. A transport that imported
+// it would be a transport that knows what an issue is, one indirection later —
+// and the protocol both of them share, internal/wire, exists precisely so that
+// neither has to reach for the other to name a payload.
+func TestTransportDoesNotImportTheDomain(t *testing.T) {
+	out, err := exec.Command("go", "list", "-deps", ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	for _, dep := range strings.Fields(string(out)) {
+		switch {
+		case strings.HasSuffix(dep, "/internal/issue"):
+			t.Errorf("the transport imports %s — what an issue IS is not a transport concept", dep)
+		case strings.HasSuffix(dep, "/internal/mapping"):
+			t.Errorf("the transport imports %s — translating is a layer of its own, and it sits above this one", dep)
+		}
+	}
+}

cli/internal/gitea/issues.go

@@ -0,0 +1,332 @@
+package gitea
+
+import (
+	"errors"
+	"fmt"
+	"net/http"
+	"net/url"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// GetIssue fetches one issue by number.
+//
+// A number is an address, not a query: this answers for a closed issue exactly
+// as it does for an open one.
+func (c *Client) GetIssue(number int) (*wire.Issue, error) {
+	var got wire.Issue
+	if err := c.Call(http.MethodGet, c.repoPathf("issues/%d", number), nil, &got); err != nil {
+		return nil, err
+	}
+	// A 200 that carries no number is not this issue. Gitea has answered that
+	// way for a repository whose issue tracker is disabled.
+	if got.Number == 0 {
+		return nil, fmt.Errorf("issue #%d not found in %s", number, c.repo)
+	}
+	return &got, nil
+}
+
+// CreateIssue files a new issue. name labels the request body in the
+// scratchpad; the issue's slug is what makes that dump worth keeping.
+func (c *Client) CreateIssue(req wire.IssueRequest, name string) (*wire.Issue, error) {
+	var got wire.Issue
+	body := &Body{Name: name, Data: req}
+	if err := c.Call(http.MethodPost, c.repoPath("issues"), body, &got); err != nil {
+		return nil, err
+	}
+	return &got, nil
+}
+
+// EditIssue patches an existing issue. Only the fields set on req are sent.
+func (c *Client) EditIssue(number int, req wire.IssueRequest, name string) (*wire.Issue, error) {
+	var got wire.Issue
+	body := &Body{Name: name, Data: req}
+	if err := c.Call(http.MethodPatch, c.repoPathf("issues/%d", number), body, &got); err != nil {
+		return nil, err
+	}
+	return &got, nil
+}
+
+// SetLabels replaces an issue's labels with exactly these ids.
+//
+// It exists because Gitea occasionally drops labels handed to it on create, and
+// the answer to that is to re-apply them rather than to trust the echo.
+func (c *Client) SetLabels(number int, ids []int64, name string) ([]wire.Label, error) {
+	if ids == nil {
+		ids = []int64{}
+	}
+	var got []wire.Label
+	body := &Body{Name: name, Data: struct {
+		Labels []int64 `json:"labels"`
+	}{ids}}
+	if err := c.Call(http.MethodPut, c.repoPathf("issues/%d/labels", number), body, &got); err != nil {
+		return nil, err
+	}
+	return got, nil
+}
+
+// ListComments is an issue's whole thread, every page of it.
+func (c *Client) ListComments(number int) ([]wire.Comment, error) {
+	return paginate[wire.Comment](c, c.repoPathf("issues/%d/comments", number), pageLimit)
+}
+
+// CreateComment posts a comment on an issue.
+func (c *Client) CreateComment(number int, text, name string) (*wire.Comment, error) {
+	var got wire.Comment
+	body := &Body{Name: name, Data: commentBody{Body: text}}
+	if err := c.Call(http.MethodPost, c.repoPathf("issues/%d/comments", number), body, &got); err != nil {
+		return nil, err
+	}
+	return &got, nil
+}
+
+// EditComment rewrites one comment, addressed by its own id and not by the
+// issue it is on — which is how Gitea addresses it.
+func (c *Client) EditComment(id int64, text, name string) (*wire.Comment, error) {
+	var got wire.Comment
+	body := &Body{Name: name, Data: commentBody{Body: text}}
+	if err := c.Call(http.MethodPatch, c.repoPathf("issues/comments/%d", id), body, &got); err != nil {
+		return nil, err
+	}
+	return &got, nil
+}
+
+type commentBody struct {
+	Body string `json:"body"`
+}
+
+// --------------------------------------------------------------------------
+// listing, and the filter the server does not honour
+// --------------------------------------------------------------------------
+
+// IssueFilter is what a listing asks for.
+type IssueFilter struct {
+	// State is open (the default), closed, or all.
+	State string
+	// Labels are label names; an issue must carry all of them.
+	Labels []string
+	// Query is Gitea's keyword search over title and body.
+	Query string
+	// Milestone is an id or a title. It is resolved against the repository
+	// before it is trusted — see ResolveMilestone.
+	Milestone string
+	// Limit counts the payloads the CALLER cares about, not the ones the server
+	// returned. Must be 1 or more.
+	Limit int
+	// Keep says whether a payload counts against Limit. Without it every
+	// payload counts and a listing behaves as any other. With it, pages keep
+	// coming until Limit have counted, and the returned list carries the ones
+	// that did not count too — they were enumerated, and a caller with
+	// something to say about them ("11 closed, not stored") still can.
+	//
+	// What Keep means is the caller's business; this package only counts.
+	Keep func(*wire.Issue) bool
+}
+
+// IssueListing is what a filtered read found.
+type IssueListing struct {
+	// Issues are every payload that passed the filter, kept or not.
+	Issues []wire.Issue
+	// Milestone is the resolved milestone title, for a receipt.
+	Milestone string
+	// Warning is set when a Keep-bounded read ran out of page budget with the
+	// budget unfilled. Returned rather than printed: the transport does not own
+	// the operator's terminal, and a caller that is rendering JSON needs it as
+	// data.
+	Warning string
+}
+
+// ListIssues reads filtered issue payloads.
+//
+// One request per page, and a payload already carries the issue body — a whole
+// milestone costs one call per page, not one per issue.
+//
+// Two boundaries hold whatever Keep decides:
+//
+//   - Stop at the limit. The page after the one that completed the budget is
+//     never requested.
+//   - Stop at the page budget. A predicate that rejects everything must not turn
+//     a bounded read into a walk of the whole tracker, so a filtered read scans
+//     at most PageSlack times the pages Limit would need if every payload
+//     counted. Hitting that with the budget unfilled sets Warning rather than
+//     answering short in silence: the caller asked for N and is told it got
+//     fewer.
+func (c *Client) ListIssues(f IssueFilter) (*IssueListing, error) {
+	if f.Limit < 1 {
+		return nil, fmt.Errorf("a listing limit must be 1 or more, got %d", f.Limit)
+	}
+
+	out := &IssueListing{}
+	var milestoneID int64
+	if f.Milestone != "" {
+		ms, err := c.ResolveMilestone(f.Milestone)
+		if err != nil {
+			return nil, err
+		}
+		milestoneID, out.Milestone = ms.ID, ms.Title
+	}
+
+	params := url.Values{}
+	state := f.State
+	if state == "" {
+		state = "open"
+	}
+	params.Set("state", state)
+	params.Set("type", "issues")
+	if len(f.Labels) > 0 {
+		params.Set("labels", strings.Join(f.Labels, ","))
+	}
+	if f.Query != "" {
+		params.Set("q", f.Query)
+	}
+	if out.Milestone != "" {
+		params.Set("milestones", out.Milestone)
+	}
+	path := c.repoPath("issues?" + params.Encode())
+
+	perPage := min(f.Limit, pageLimit)
+	ideal := max(1, (f.Limit+perPage-1)/perPage)
+	budget := ideal
+	if f.Keep != nil {
+		budget = ideal * PageSlack
+	}
+
+	kept, seen, lastFull := 0, 0, false
+	err := pages(c, path, perPage, budget, func(batch []wire.Issue) (bool, error) {
+		seen++
+		lastFull = len(batch) == perPage
+		for i := range batch {
+			p := &batch[i]
+			if !matches(p, milestoneID, f.Labels) {
+				continue
+			}
+			out.Issues = append(out.Issues, *p)
+			if f.Keep == nil || f.Keep(p) {
+				kept++
+				if kept >= f.Limit {
+					return false, nil
+				}
+			}
+		}
+		return true, nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	if f.Keep != nil && seen >= budget && lastFull {
+		out.Warning = fmt.Sprintf("scanned %d page(s) and stopped %d short of the limit of %d"+
+			" — there may be more; narrow the filter or raise the limit", budget, f.Limit-kept, f.Limit)
+	}
+	return out, nil
+}
+
+// matches re-checks on the client what the server was already asked for.
+//
+// Not paranoia: Gitea silently IGNORES a `milestones=` value it cannot resolve
+// and answers with the whole backlog, which is why the milestone is resolved to
+// an id first and every payload is checked against that id here. The same
+// re-check on labels costs nothing, and `pull_request` is the one filter that
+// matters most — a pull request rendered as a unit of work is not a bug the
+// operator can see until it is in the store.
+//
+// A function and not a method: the payload is the protocol's, and re-checking a
+// filter the server ignored is this package's business, not the protocol's.
+func matches(i *wire.Issue, milestoneID int64, labels []string) bool {
+	if i.IsPullRequest() {
+		return false
+	}
+	if milestoneID != 0 && (i.Milestone == nil || i.Milestone.ID != milestoneID) {
+		return false
+	}
+	have := make(map[string]bool, len(i.Labels))
+	for _, l := range i.Labels {
+		have[l.Name] = true
+	}
+	for _, want := range labels {
+		if !have[want] {
+			return false
+		}
+	}
+	return true
+}
+
+// --------------------------------------------------------------------------
+// dependencies
+// --------------------------------------------------------------------------
+
+// issueMeta is Gitea's IssueMeta: how a dependency names another issue.
+type issueMeta struct {
+	Index int    `json:"index"`
+	Owner string `json:"owner"`
+	Repo  string `json:"repo"`
+}
+
+// Dependencies are the issues that block this one — Gitea's own dependency
+// links, read in the direction AddDependency writes them.
+//
+// An instance that does not have the endpoint, or has dependencies turned off
+// for this repository, answers with a status rather than a list. That is
+// reported as "no dependencies" and not as a failure: a pull must still bring
+// the issue itself back from a tracker whose dependency support is off.
+//
+// Deliberately narrower than the Python it replaces, which swallowed every
+// failure here including a dead connection. "The server said no" and "there was
+// no server" are different answers, and only the first one means the feature is
+// missing.
+func (c *Client) Dependencies(number int) ([]wire.Issue, error) {
+	var got []wire.Issue
+	err := c.Call(http.MethodGet, c.repoPathf("issues/%d/dependencies", number), nil, &got)
+	var apiErr *APIError
+	if errors.As(err, &apiErr) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	return got, nil
+}
+
+// DependencyKeys is the same links as cross-repo handles — what a repeat push
+// compares against so it does not POST a link the tracker already has.
+//
+// A bare number is ambiguous the moment a dependency lives in another
+// repository, and Gitea lets it, so the repository travels with it.
+func (c *Client) DependencyKeys(number int) ([]wire.Key, error) {
+	deps, err := c.Dependencies(number)
+	if err != nil {
+		return nil, err
+	}
+	out := make([]wire.Key, 0, len(deps))
+	for i := range deps {
+		out = append(out, deps[i].KeyIn(c.repo))
+	}
+	return out, nil
+}
+
+// AddDependency makes issue number depend on dep.
+//
+// Confirmed against an instance's own swagger.v1.json (Gitea 1.26.1):
+//
+//	POST /repos/{owner}/{repo}/issues/{index}/dependencies
+//	body: IssueMeta — {"index": <int>, "owner": "<owner>", "repo": "<name>"}
+//	"Make the issue in the url depend on the issue in the form."
+//
+// So the URL names the blocked issue and the body the blocker, which is the
+// direction Dependencies reads back. A link that already exists answers 409, so
+// callers pre-filter with DependencyKeys and treat a failure here as a note
+// rather than an abort: one missing cross-link must not undo a push that has
+// already created issues.
+func (c *Client) AddDependency(number int, dep wire.Key) error {
+	if dep.Repo.Zero() {
+		return fmt.Errorf("dependency %s names no repository — a link needs owner/repo#number", dep)
+	}
+	if dep.Number < 1 {
+		return fmt.Errorf("dependency %s names no issue number", dep)
+	}
+	body := &Body{
+		Name: fmt.Sprintf("dep-%d-%d", number, dep.Number),
+		Data: issueMeta{Index: dep.Number, Owner: dep.Repo.Owner, Repo: dep.Repo.Name},
+	}
+	return c.Call(http.MethodPost, c.repoPathf("issues/%d/dependencies", number), body, nil)
+}

cli/internal/gitea/labels.go

@@ -0,0 +1,108 @@
+package gitea
+
+import (
+	"fmt"
+	"net/http"
+	"strconv"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// ListLabels is every label in the repository, every page of it.
+//
+// A bootstrap decides its plan against this and never against a cache: a cache
+// answers "what did we create last time", and the question is "what does the
+// repository have right now".
+func (c *Client) ListLabels() ([]wire.Label, error) {
+	return paginate[wire.Label](c, c.repoPath("labels"), 100)
+}
+
+// CreateLabel adds a label to the repository.
+//
+// Through the API rather than through any CLI wrapper, because `exclusive` —
+// the flag that makes `type/*` behave like a single choice — is not something
+// the `tea` client could set.
+//
+// What a label MEANS is not decided here either: this creates what it is
+// handed.
+func (c *Client) CreateLabel(req wire.LabelRequest) (*wire.Label, error) {
+	var got wire.Label
+	body := &Body{Name: "label-" + req.Name, Data: req}
+	if err := c.Call(http.MethodPost, c.repoPath("labels"), body, &got); err != nil {
+		return nil, err
+	}
+	if got.ID == 0 {
+		return nil, fmt.Errorf("creating label %q: the tracker's answer carries no id", req.Name)
+	}
+	return &got, nil
+}
+
+// EditLabel patches an existing label by id.
+func (c *Client) EditLabel(id int64, req wire.LabelRequest) (*wire.Label, error) {
+	var got wire.Label
+	body := &Body{Name: "label-" + req.Name, Data: req}
+	if err := c.Call(http.MethodPatch, c.repoPathf("labels/%d", id), body, &got); err != nil {
+		return nil, err
+	}
+	return &got, nil
+}
+
+// ListMilestones is every milestone in the repository, open and closed.
+//
+// Both states, always: a milestone is closed the moment its work is done, and a
+// listing that hid those would fail to resolve exactly the filter somebody
+// types when they want to see what was in it.
+func (c *Client) ListMilestones() ([]wire.Milestone, error) {
+	return paginate[wire.Milestone](c, c.repoPath("milestones?state=all"), 100)
+}
+
+// ResolveMilestone finds a milestone by id or by title, and fails when there is
+// none.
+//
+// It fails LOUDLY, and that is the whole point of resolving before filtering:
+// Gitea silently ignores a `milestones=` filter it cannot resolve and answers
+// with the entire backlog. A typo in a milestone name would otherwise read as
+// "your milestone has 300 issues in it".
+func (c *Client) ResolveMilestone(value string) (*wire.Milestone, error) {
+	got, err := c.ListMilestones()
+	if err != nil {
+		return nil, err
+	}
+	for i := range got {
+		if got[i].Title == value || strconv.FormatInt(got[i].ID, 10) == value {
+			return &got[i], nil
+		}
+	}
+	have := make([]string, 0, len(got))
+	for _, m := range got {
+		have = append(have, fmt.Sprintf("%s (id %d)", m.Title, m.ID))
+	}
+	if len(have) == 0 {
+		have = []string{"none"}
+	}
+	return nil, fmt.Errorf("no milestone %q in %s — have: %s", value, c.repo, strings.Join(have, ", "))
+}
+
+// FindMilestone is the milestone with this title, or nil when the repository
+// has no such milestone.
+//
+// The quiet counterpart of ResolveMilestone, for a push: an issue naming a
+// milestone the tracker does not have is filed without one, because refusing
+// the whole push over a field the tracker will happily accept as empty helps
+// nobody. "none" and "" are both "no milestone".
+func (c *Client) FindMilestone(title string) (*wire.Milestone, error) {
+	if title == "" || title == "none" {
+		return nil, nil
+	}
+	got, err := c.ListMilestones()
+	if err != nil {
+		return nil, err
+	}
+	for i := range got {
+		if got[i].Title == title {
+			return &got[i], nil
+		}
+	}
+	return nil, nil
+}

cli/internal/gitea/remotemap.go

@@ -0,0 +1,88 @@
+package gitea
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// RemoteMapName is the ledger's file name, beside the issues it indexes.
+const RemoteMapName = ".remote.json"
+
+// RemoteMap is the number -> slug ledger: {"owner/repo#42": "wire-sqlc-appclick"}.
+//
+// ITS ENTRIES OUTLIVE THE FILES THEY NAME, and that is deliberate rather than a
+// leak. A push deletes an issue's file the moment the tracker confirms the
+// write, and the entry left behind is what makes the next pull of that number
+// land on the same slug — so every `depends:` that pointed at it still
+// resolves. Nothing prunes them, because "no file" no longer means "no such
+// issue"; eviction does not prune it either, for the same reason a push does
+// not. A stale entry costs one line of JSON and is corrected the next time that
+// number is pulled.
+//
+// It is a cache, not a record. The slug also travels tracker-side, in the issue
+// body, so losing this file costs a re-pull and not information — which is why
+// Load never fails and why a rebuild is a MERGE and never a replacement. The
+// order of authority:
+//
+//	the tracker      the issue, and the marker naming its slug
+//	.remote.json     a local number -> slug ledger, a cache of that marker
+//	the store        whatever happens to be checked out right now
+//
+// The store is a subset of what the ledger knows, so a rebuild that started
+// from the files alone would throw away every entry it cannot see. Start from
+// Load, add what the files say, Save.
+type RemoteMap map[string]string
+
+// RemoteMapPath is where the ledger lives: inside the issue store, beside the
+// issues. root is the STORE, not the payload scratchpad — this file is
+// bookkeeping about issues and belongs where they are.
+func RemoteMapPath(root string) string { return filepath.Join(root, RemoteMapName) }
+
+// LoadRemoteMap reads the ledger.
+//
+// A missing, unreadable or malformed file is an empty ledger and never an
+// error. The ledger is a cache of markers the tracker holds, so refusing to run
+// because it cannot be parsed would block the very pull that would rebuild it —
+// and the cost of starting empty is one re-pull, never a lost issue.
+func LoadRemoteMap(root string) RemoteMap {
+	raw, err := os.ReadFile(RemoteMapPath(root))
+	if err != nil {
+		return RemoteMap{}
+	}
+	var got RemoteMap
+	if err := json.Unmarshal(raw, &got); err != nil || got == nil {
+		return RemoteMap{}
+	}
+	return got
+}
+
+// Save writes the ledger, creating the directory if it is not there.
+//
+// The one write in this package allowed to create the store, and only because
+// of when it happens: the ledger is written the instant the tracker confirms a
+// push and BEFORE the local file is deleted, so failing it over a missing
+// directory would lose the slug at exactly the moment the local copy stops
+// being the record.
+//
+// Indented and key-sorted — encoding/json sorts map keys for us — because this
+// file is read by people and diffed by git as often as it is read by the
+// binary.
+func (m RemoteMap) Save(root string) error {
+	if err := os.MkdirAll(root, 0o755); err != nil {
+		return err
+	}
+	raw, err := json.MarshalIndent(m, "", "  ")
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(RemoteMapPath(root), append(raw, '\n'), 0o644)
+}
+
+// Slug is the local name recorded for a key, or "".
+func (m RemoteMap) Slug(k wire.Key) string { return m[k.String()] }
+
+// Set records that a key is known locally under this slug.
+func (m RemoteMap) Set(k wire.Key, slug string) { m[k.String()] = slug }

cli/internal/gitea/remotemap_test.go

@@ -0,0 +1,80 @@
+package gitea_test
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func TestRemoteMapRoundTrips(t *testing.T) {
+	root := filepath.Join(t.TempDir(), "issues")
+	key := wire.Key{Repo: wire.Repo{Owner: "acme", Name: "widgets"}, Number: 42}
+
+	m := gitea.RemoteMap{}
+	m.Set(key, "wire-sqlc-appclick")
+	if err := m.Save(root); err != nil {
+		t.Fatalf("Save: %v", err)
+	}
+
+	path := gitea.RemoteMapPath(root)
+	if want := filepath.Join(root, ".remote.json"); path != want {
+		t.Errorf("the ledger is at %s, want %s — beside the issues it indexes", path, want)
+	}
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("reading the ledger: %v", err)
+	}
+	if !strings.Contains(string(raw), `"acme/widgets#42": "wire-sqlc-appclick"`) {
+		t.Errorf("the ledger is not readable by a human:\n%s", raw)
+	}
+
+	back := gitea.LoadRemoteMap(root)
+	if got := back.Slug(key); got != "wire-sqlc-appclick" {
+		t.Errorf("the key came back as %q, want wire-sqlc-appclick", got)
+	}
+	if got := back.Slug(wire.Key{Repo: key.Repo, Number: 7}); got != "" {
+		t.Errorf("an unrecorded key answered %q", got)
+	}
+
+	// A rebuild is a merge and never a replacement: what is already recorded
+	// survives an entry added on top of it. This is what makes a pull of a
+	// number whose file was deleted by a push land on the same slug.
+	second := wire.Key{Repo: key.Repo, Number: 43}
+	back.Set(second, "drop-the-wiki")
+	if err := back.Save(root); err != nil {
+		t.Fatalf("Save: %v", err)
+	}
+	again := gitea.LoadRemoteMap(root)
+	if again.Slug(key) != "wire-sqlc-appclick" || again.Slug(second) != "drop-the-wiki" {
+		t.Errorf("a second save lost an entry: %v", again)
+	}
+}
+
+// The ledger is a cache of markers the tracker holds, so an unreadable one must
+// not stop the pull that would rebuild it.
+func TestRemoteMapSurvivesAMissingOrMangledFile(t *testing.T) {
+	root := t.TempDir()
+
+	if got := gitea.LoadRemoteMap(filepath.Join(root, "nowhere")); len(got) != 0 {
+		t.Errorf("a missing ledger loaded as %v", got)
+	}
+	if err := os.WriteFile(gitea.RemoteMapPath(root), []byte("{ not json at all"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	got := gitea.LoadRemoteMap(root)
+	if len(got) != 0 {
+		t.Errorf("a mangled ledger loaded as %v", got)
+	}
+	// Still writable afterwards: an unreadable ledger costs a re-pull, not a run.
+	got.Set(wire.Key{Repo: wire.Repo{Owner: "acme", Name: "widgets"}, Number: 1}, "first")
+	if err := got.Save(root); err != nil {
+		t.Fatalf("Save over a mangled ledger: %v", err)
+	}
+	if gitea.LoadRemoteMap(root).Slug(wire.Key{Repo: wire.Repo{Owner: "acme", Name: "widgets"}, Number: 1}) != "first" {
+		t.Error("the ledger did not come back after being rewritten")
+	}
+}

cli/internal/issue/checkbox.go

@@ -0,0 +1,185 @@
+package issue
+
+import (
+	"fmt"
+	"regexp"
+	"strings"
+)
+
+// A checkbox is the one part of a body that is *state* and not prose, so the
+// format gives it markup of its own. It is item markup, not a property of one
+// section: `## Acceptance criteria` is the usual home, but a type/feature
+// keeps its children as checkboxes under `## Issues`. The scan is therefore
+// over the whole text and the heading is only recorded, never required.
+var (
+	// The trailing group stands in for a lookahead RE2 does not have: after
+	// the bracket there is either whitespace and then anything, or end of line.
+	checkboxRe = regexp.MustCompile(
+		`^(?P<indent>[ \t]*)(?P<marker>[-*+]|\d+[.)])[ \t]+` +
+			`\[(?P<box>[ xX])\](?P<text>[ \t].*|)$`)
+	// Any list item — a sibling ends the item above it, checkbox or not.
+	listItemRe = regexp.MustCompile(`^[ \t]*([-*+]|\d+[.)])([ \t]|$)`)
+	fenceRe    = regexp.MustCompile("^[ \t]{0,3}(`{3,}|~{3,})")
+)
+
+// Checkbox is one checkbox item found in a text.
+type Checkbox struct {
+	// Index is the 1-based position in the list — what a user types to pick it.
+	Index int
+	// Line is the 1-based line of the `- [ ]` marker, in the text given.
+	Line int
+	// EndLine is the 1-based last line of the item, continuations included.
+	EndLine int
+	// Checked is true for [x] / [X].
+	Checked bool
+	// Text is the item's text; continuation lines joined with one space.
+	Text string
+	// Section is the nearest preceding `## ` heading, "" above the first one.
+	Section string
+}
+
+// Checkboxes returns every checkbox item in text, in document order.
+//
+// A pure function of the string it is given — no I/O, no store, no tracker.
+// Pass an issue body to get body-relative line numbers, or a whole file to get
+// file-relative ones; nothing else changes.
+//
+// Rules:
+//
+//   - Only a line matching checkboxRe opens an item. A wrapped ("continuation")
+//     line is part of the item above it, never an item of its own; the item runs
+//     to the next blank line, heading, code fence, or list marker.
+//   - Fenced code blocks are skipped whole: `- [ ]` inside a fence is an example
+//     of the markup, not a box anybody may tick.
+//   - `-`, `*`, `+` and `1.` markers all count, at any indentation, so nested
+//     lists are seen too.
+func Checkboxes(text string) []Checkbox {
+	lines := splitLines(text)
+	var items []Checkbox
+	section, fence := "", ""
+
+	for n, line := range lines {
+		if m := fenceRe.FindStringSubmatch(line); m != nil {
+			tok := m[1]
+			switch {
+			case fence == "":
+				fence = tok
+			case tok[0] == fence[0] && len(tok) >= len(fence):
+				fence = ""
+			}
+			continue
+		}
+		if fence != "" {
+			continue
+		}
+		if strings.HasPrefix(line, "## ") {
+			section = strings.TrimSpace(line)
+			continue
+		}
+		if strings.HasPrefix(line, "# ") {
+			section = ""
+			continue
+		}
+		m := checkboxRe.FindStringSubmatch(line)
+		if m == nil {
+			continue
+		}
+		end := n + 1
+		parts := []string{strings.TrimSpace(m[4])}
+		for k := n + 1; k < len(lines); k++ {
+			next := lines[k]
+			if strings.TrimSpace(next) == "" || strings.HasPrefix(next, "#") ||
+				fenceRe.MatchString(next) || listItemRe.MatchString(next) {
+				break
+			}
+			end = k + 1
+			parts = append(parts, strings.TrimSpace(next))
+		}
+		var kept []string
+		for _, p := range parts {
+			if p != "" {
+				kept = append(kept, p)
+			}
+		}
+		items = append(items, Checkbox{
+			Index:   len(items) + 1,
+			Line:    n + 1,
+			EndLine: end,
+			Checked: m[3] != " ",
+			Text:    strings.Join(kept, " "),
+			Section: section,
+		})
+	}
+	return items
+}
+
+// SetCheckbox returns text with the checkbox on the given 1-based line set to
+// checked.
+//
+// Pure, and deliberately surgical: exactly one byte of the input changes — the
+// one between the brackets. Everything else, including trailing whitespace and
+// the item's own wording, comes back byte for byte. That is the whole point:
+// ticking a box must not produce a diff wider than the state that changed.
+//
+// Already in the requested state is a no-op — text comes back unchanged, and
+// an existing [X] keeps its capital.
+func SetCheckbox(text string, line int, checked bool) (string, error) {
+	off := 0
+	for n := 1; off <= len(text); n++ {
+		nl := strings.IndexByte(text[off:], '\n')
+		var raw string
+		if nl == -1 {
+			raw = text[off:]
+		} else {
+			raw = text[off : off+nl]
+		}
+		if n == line {
+			m := checkboxRe.FindStringSubmatchIndex(strings.TrimRight(raw, "\r"))
+			if m == nil {
+				return "", fmt.Errorf("line %d is not a checkbox item", line)
+			}
+			box := off + m[6] // group 3: box
+			if (text[box] != ' ') == checked {
+				return text, nil
+			}
+			c := byte(' ')
+			if checked {
+				c = 'x'
+			}
+			return text[:box] + string(c) + text[box+1:], nil
+		}
+		if nl == -1 {
+			break
+		}
+		off += nl + 1
+	}
+	return "", fmt.Errorf("line %d is past the end of the text", line)
+}
+
+// CheckboxProgress is (done, total) over every checkbox in text; (0, 0) when
+// it has none.
+//
+// Computed on the fly, on purpose. Progress is not a metadata field: it is the
+// body read back, and the body is the only place the state lives.
+func CheckboxProgress(text string) (done, total int) {
+	items := Checkboxes(text)
+	for _, c := range items {
+		if c.Checked {
+			done++
+		}
+	}
+	return done, len(items)
+}
+
+// splitLines is strings.Split minus the phantom final element a trailing
+// newline produces, matching Python's str.splitlines().
+func splitLines(text string) []string {
+	if text == "" {
+		return nil
+	}
+	lines := strings.Split(strings.ReplaceAll(text, "\r\n", "\n"), "\n")
+	if last := len(lines) - 1; lines[last] == "" {
+		lines = lines[:last]
+	}
+	return lines
+}

cli/internal/issue/checkbox_test.go

@@ -0,0 +1,117 @@
+package issue
+
+import "testing"
+
+const boxes = `## Acceptance criteria
+- [ ] первый пункт
+- [x] второй пункт, который
+  переносится на вторую строку
+* [X] третий
+1. [ ] четвёртый
+
+## Notes
+` + "```" + `
+- [ ] это пример разметки, а не состояние
+` + "```" + `
+`
+
+func TestCheckboxesReadTheWholeBody(t *testing.T) {
+	items := Checkboxes(boxes)
+	if len(items) != 4 {
+		t.Fatalf("found %d items, want 4: %+v", len(items), items)
+	}
+	if items[1].Text != "второй пункт, который переносится на вторую строку" {
+		t.Errorf("continuation not joined: %q", items[1].Text)
+	}
+	if items[1].EndLine != 4 {
+		t.Errorf("end line = %d, want 4", items[1].EndLine)
+	}
+	if !items[2].Checked {
+		t.Error("[X] must read as checked")
+	}
+	if items[3].Section != ACSection {
+		t.Errorf("section = %q", items[3].Section)
+	}
+	for _, c := range items {
+		if c.Section == "## Notes" {
+			t.Error("a checkbox inside a code fence was counted")
+		}
+	}
+}
+
+func TestCheckboxProgressIsCountedOffTheBody(t *testing.T) {
+	done, total := CheckboxProgress(boxes)
+	if done != 2 || total != 4 {
+		t.Errorf("progress = %d/%d, want 2/4", done, total)
+	}
+}
+
+func TestSetCheckboxChangesExactlyOneByte(t *testing.T) {
+	items := Checkboxes(boxes)
+	got, err := SetCheckbox(boxes, items[0].Line, true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(got) != len(boxes) {
+		t.Fatalf("length changed: %d -> %d", len(boxes), len(got))
+	}
+	diff := 0
+	for i := range got {
+		if got[i] != boxes[i] {
+			diff++
+		}
+	}
+	if diff != 1 {
+		t.Errorf("%d bytes changed, want 1", diff)
+	}
+}
+
+func TestSetCheckboxIsANoOpWhenAlreadyInState(t *testing.T) {
+	items := Checkboxes(boxes)
+	// [X] keeps its capital: the state already matches, so nothing is rewritten.
+	got, err := SetCheckbox(boxes, items[2].Line, true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got != boxes {
+		t.Error("an already-checked box was rewritten")
+	}
+}
+
+func TestSetCheckboxRefusesALineThatIsNotOne(t *testing.T) {
+	if _, err := SetCheckbox(boxes, 1, true); err == nil {
+		t.Error("ticking a heading must fail")
+	}
+	if _, err := SetCheckbox(boxes, 9999, true); err == nil {
+		t.Error("ticking past the end must fail")
+	}
+}
+
+func TestBodyDepRefsOnlyReadTheDepSections(t *testing.T) {
+	body := `## Summary
+смотри также some-other-issue, который не зависимость
+
+## Depends on
+- migrate-schema — нужна схема
+- add-pool-cfg
+
+## Issues
+- [ ] wire-sqlc-appclick — часть
+- [ ] #42
+`
+	got := BodyDepRefs(body)
+	want := []DepRef{
+		{DependsSection, "migrate-schema"},
+		{DependsSection, "add-pool-cfg"},
+		{IssuesSection, "wire-sqlc-appclick"},
+		{IssuesSection, "#42"},
+	}
+	if len(got) != len(want) {
+		t.Fatalf("got %+v, want %+v", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Errorf("ref %d = %+v, want %+v", i, got[i], want[i])
+		}
+	}
+}

cli/internal/issue/depsection.go

@@ -0,0 +1,59 @@
+package issue
+
+import (
+	"regexp"
+	"strings"
+)
+
+// A reference is a slug, or `#N` on an issue that came from a tracker.
+var depRefRe = regexp.MustCompile(`#(\d+)|\b([a-z0-9]+(?:-[a-z0-9]+)+)\b`)
+
+// DepRef is one dependency reference written in the body prose, carried out
+// with the section it was found in.
+//
+// The section travels with the reference so a caller can name the one the
+// reader actually has in front of them: a container's children come from
+// `## Issues`, and pointing at `## Depends on` would name a section that is not
+// in the file.
+type DepRef struct {
+	Section string
+	Ref     string
+}
+
+// BodyDepRefs returns every reference under one of DepSections, deduplicated
+// on first sight, in order of first appearance.
+//
+// Never from prose elsewhere, or a graph walk would drag in half the backlog.
+func BodyDepRefs(body string) []DepRef {
+	var out []DepRef
+	seen := map[string]bool{}
+	section := ""
+
+	for _, line := range splitLines(body) {
+		if strings.HasPrefix(line, "## ") {
+			head := strings.TrimSpace(line)
+			section = ""
+			for _, s := range DepSections {
+				if head == s {
+					section = head
+					break
+				}
+			}
+			continue
+		}
+		if section == "" {
+			continue
+		}
+		for _, tok := range depRefRe.FindAllStringSubmatch(line, -1) {
+			ref := tok[2]
+			if tok[1] != "" {
+				ref = "#" + tok[1]
+			}
+			if !seen[ref] {
+				seen[ref] = true
+				out = append(out, DepRef{Section: section, Ref: ref})
+			}
+		}
+	}
+	return out
+}

cli/internal/issue/evict.go

@@ -0,0 +1,155 @@
+package issue
+
+import (
+	"os"
+	"sort"
+)
+
+// Closed issues leave the store. The store is a working set, not an archive.
+//
+// WHAT IS EVICTED, and it is two conditions, both read off the file:
+//
+//	state: closed      the work is done
+//	origin: <tracker>  the work is somewhere else too
+//
+// THE SECOND CONDITION IS THE WHOLE SAFETY ARGUMENT. `origin: local` means this
+// file IS the issue — there is no other copy and deleting it deletes the work.
+// It is therefore never evicted, in any state, not even when named explicitly:
+// a closed local issue is reported and kept. The only files that go are ones
+// whose own metadata says the work can be fetched back, which is the same trade
+// a push makes when it drops a file the tracker has just confirmed.
+//
+// That parallel is exact except for where the confirmation comes from. Push has
+// to ask the tracker, because it is the tracker that just changed. Eviction asks
+// the file, because state and origin are domain fields and the answer is already
+// in the store — which is why this lives in the domain and needs no network, no
+// login, and no tracker. The sync layer's variant refreshes state from the
+// tracker first and then calls Evict, so there is exactly one implementation of
+// "what may be evicted" and it is this one.
+//
+// NOT A ONE-OFF MIGRATION. A pull by number fetches an issue in any state — a
+// number is an address, not a query — so a closed issue pulled after an eviction
+// lands on disk again. That is the tracker being asked a direct question, not a
+// regression; evict it again when you are done with it.
+//
+// `.remote.json` is deliberately NOT pruned. It is the local number -> slug
+// ledger, its entries outlive the files they name, and an evicted issue is in
+// exactly that state. INDEX.md is rebuilt, because it IS a view of the
+// directory.
+
+const closed = "closed"
+
+// LocalReason is printed whether or not the issue was named, because "this
+// closed thing is still here" needs an answer every time.
+const LocalReason = "origin: " + Local + " — this file IS the issue"
+
+// Evicted is one issue that left the store, with every file that went with it.
+type Evicted struct {
+	ID    string
+	Paths []string
+}
+
+// Kept is one issue that was considered and stayed, with the reason.
+type Kept struct {
+	ID   string
+	Why  string
+	Open bool // true when it is simply not closed yet — the normal case
+}
+
+// EvictReport is what a run did, or would have done.
+type EvictReport struct {
+	Evicted    []Evicted
+	Kept       []Kept
+	DryRun     bool
+	IndexPath  string
+	IndexCount int
+}
+
+// Classify splits the store into what may be evicted, what is protected, and
+// what is still open.
+//
+// Pure — it reads the loaded issues and decides; nothing here touches disk.
+// ids restricts the question to those issues; empty considers the whole store.
+// A protected issue is returned as such even when it was named explicitly:
+// naming a file does not make deleting it safe.
+func Classify(issues map[string]*Issue, ids []string) (evict, protected, stillOpen []string) {
+	chosen := ids
+	if len(chosen) == 0 {
+		for id := range issues {
+			chosen = append(chosen, id)
+		}
+		sort.Strings(chosen)
+	}
+	for _, id := range chosen {
+		i, ok := issues[id]
+		if !ok {
+			continue
+		}
+		switch {
+		case i.State != closed:
+			stillOpen = append(stillOpen, id)
+		case i.IsLocal():
+			protected = append(protected, id)
+		default:
+			evict = append(evict, id)
+		}
+	}
+	return evict, protected, stillOpen
+}
+
+// Remove deletes everything the store holds under one slug and returns the
+// paths that went.
+//
+// Deliberately dumb: it takes an id, not a decision. Whether an issue may go is
+// settled by Classify before this is reached, so the dangerous half of the
+// operation has no branches in it at all.
+func Remove(root, id string) ([]string, error) {
+	var gone []string
+	for _, p := range SlugFiles(root, id) {
+		if err := os.Remove(p); err != nil {
+			return gone, err
+		}
+		gone = append(gone, p)
+	}
+	return gone, nil
+}
+
+// Evict classifies, removes, and rebuilds the index. The one implementation,
+// called both by the offline command and by the sync layer — which does nothing
+// to this decision except hand over issues whose state it has just refreshed
+// from the tracker.
+func Evict(root string, issues map[string]*Issue, ids []string, dryRun bool) (*EvictReport, error) {
+	evict, protected, stillOpen := Classify(issues, ids)
+	rep := &EvictReport{DryRun: dryRun}
+
+	for _, id := range evict {
+		var paths []string
+		if dryRun {
+			paths = SlugFiles(root, id)
+		} else {
+			var err error
+			if paths, err = Remove(root, id); err != nil {
+				return rep, err
+			}
+		}
+		rep.Evicted = append(rep.Evicted, Evicted{ID: id, Paths: paths})
+	}
+	for _, id := range protected {
+		rep.Kept = append(rep.Kept, Kept{ID: id, Why: LocalReason})
+	}
+	for _, id := range stillOpen {
+		rep.Kept = append(rep.Kept, Kept{ID: id, Why: "state: " + issues[id].State, Open: true})
+	}
+
+	// Only when something actually went: the index is a view of the directory,
+	// and rewriting it after a run that changed nothing is a write nobody asked
+	// for.
+	if !dryRun && len(rep.Evicted) > 0 {
+		path, n, err := BuildIndex(root)
+		if err != nil {
+			return rep, err
+		}
+		rep.IndexPath, rep.IndexCount = path, n
+	}
+	return rep, nil
+}

cli/internal/issue/graph.go

@@ -0,0 +1,106 @@
+package issue
+
+import "sort"
+
+// Graph is the edge list read off the `depends:` metadata — the authoritative
+// one. Body prose is never walked.
+func Graph(issues map[string]*Issue) map[string][]string {
+	out := make(map[string][]string, len(issues))
+	for id, i := range issues {
+		out[id] = append([]string{}, i.Depends...)
+	}
+	return out
+}
+
+// Dependents lists who depends on id — the upward direction.
+func Dependents(issues map[string]*Issue, id string) []string {
+	var out []string
+	for other, i := range issues {
+		if contains(i.Depends, id) {
+			out = append(out, other)
+		}
+	}
+	sort.Strings(out)
+	return out
+}
+
+// TopoOrder puts dependencies first.
+//
+// Cycles are broken deterministically rather than raising: a cycle is a data
+// problem for the caller to report, not a reason to refuse to order the rest.
+func TopoOrder(ids []string, edges map[string][]string) []string {
+	const (
+		open = 1
+		done = 2
+	)
+	state := map[string]int{}
+	var order []string
+
+	var visit func(string)
+	visit = func(n string) {
+		switch state[n] {
+		case done, open: // open = a back edge; leave it unresolved
+			return
+		}
+		state[n] = open
+		for _, d := range edges[n] {
+			if _, ok := edges[d]; ok {
+				visit(d)
+			}
+		}
+		state[n] = done
+		order = append(order, n)
+	}
+
+	for _, n := range ids {
+		visit(n)
+	}
+	return order
+}
+
+// FindCycles returns one id list per cycle. Empty when the graph is a DAG.
+func FindCycles(edges map[string][]string) [][]string {
+	const (
+		open = 1
+		done = 2
+	)
+	state := map[string]int{}
+	var stack []string
+	var cycles [][]string
+
+	var visit func(string)
+	visit = func(n string) {
+		state[n] = open
+		stack = append(stack, n)
+		for _, d := range edges[n] {
+			if _, ok := edges[d]; !ok {
+				continue
+			}
+			if state[d] == open {
+				for i, s := range stack {
+					if s == d {
+						cycles = append(cycles, append(append([]string{}, stack[i:]...), d))
+						break
+					}
+				}
+			} else if state[d] == 0 {
+				visit(d)
+			}
+		}
+		stack = stack[:len(stack)-1]
+		state[n] = done
+	}
+
+	// Sorted so the report is the same on every run; Go map order is not.
+	ids := make([]string, 0, len(edges))
+	for n := range edges {
+		ids = append(ids, n)
+	}
+	sort.Strings(ids)
+	for _, n := range ids {
+		if state[n] == 0 {
+			visit(n)
+		}
+	}
+	return cycles
+}

cli/internal/issue/index.go

@@ -0,0 +1,128 @@
+package issue
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strings"
+)
+
+var treeFileRe = regexp.MustCompile(`^tree-.+\.md$`)
+
+const indexPreamble = "Every issue this project knows about. `origin: local` means it " +
+	"exists nowhere else — a complete state, not a pending one. Any other value " +
+	"names the tracker it also lives in; the handle is in the file. `progress` " +
+	"counts the body's checkboxes, ticked over total, and is blank for an issue " +
+	"that has none — read off the body at build time, stored nowhere. Rebuild " +
+	"with `kettle index`; tick a box with `kettle ac`."
+
+// BuildIndex rewrites INDEX.md from what is on disk and returns its path and
+// the number of issues in it.
+//
+// An index of a store that is not there is not an empty index, it is a bad
+// path: failing beats writing INDEX.md into a directory nobody asked for. An
+// existing store with nothing in it is a legitimate thing to index and gets an
+// "_empty_" table.
+func BuildIndex(root string) (string, int, error) {
+	if err := RequireStore(root); err != nil {
+		return "", 0, err
+	}
+	issues, err := LoadAll(root)
+	if err != nil {
+		return "", 0, err
+	}
+
+	ids := make([]string, 0, len(issues))
+	for id := range issues {
+		ids = append(ids, id)
+	}
+	sort.Strings(ids)
+
+	out := []string{"# Issue store", "", indexPreamble, ""}
+	if len(ids) > 0 {
+		out = append(out,
+			"| id | state | progress | type | labels | title | milestone | depends | origin |",
+			"|---|---|---|---|---|---|---|---|---|")
+		for _, id := range ids {
+			i := issues[id]
+			var rest []string
+			for _, l := range i.Labels {
+				if !strings.HasPrefix(l, "type/") {
+					rest = append(rest, l)
+				}
+			}
+			out = append(out, fmt.Sprintf("| [%s](%s.md) | %s | %s | %s | %s | %s | %s | %s | %s |",
+				id, id, cell(i.State), progress(i.Body), cell(i.Type()),
+				cellList(rest), cell(i.Title), cell(i.Milestone),
+				cellList(i.Depends), cell(i.Origin)))
+		}
+	} else {
+		out = append(out, "_empty_")
+	}
+
+	if trees := treeFiles(root); len(trees) > 0 {
+		out = append(out, "", "## Dependency trees", "")
+		for _, t := range trees {
+			out = append(out, fmt.Sprintf("- [%s](%s)", t, t))
+		}
+	}
+
+	if cycles := FindCycles(Graph(issues)); len(cycles) > 0 {
+		out = append(out, "", "## Dependency cycles", "")
+		for _, c := range cycles {
+			out = append(out, "- "+strings.Join(c, " -> "))
+		}
+	}
+	out = append(out, "")
+
+	path := filepath.Join(root, "INDEX.md")
+	if err := os.WriteFile(path, []byte(strings.Join(out, "\n")), 0o644); err != nil {
+		return "", 0, err
+	}
+	return path, len(ids), nil
+}
+
+// progress is `3/7` for a body with checkboxes, "" for one without.
+//
+// Counted from the body every time the index is built and stored nowhere — the
+// boxes are the state, and a second copy of it in a metadata field would be
+// wrong by the next edit.
+func progress(body string) string {
+	done, total := CheckboxProgress(body)
+	if total == 0 {
+		return ""
+	}
+	return fmt.Sprintf("%d/%d", done, total)
+}
+
+func cell(v string) string {
+	v = strings.TrimSpace(v)
+	if v == "" {
+		return "—"
+	}
+	return strings.ReplaceAll(v, "|", `\|`)
+}
+
+func cellList(xs []string) string {
+	if len(xs) == 0 {
+		return "—"
+	}
+	return strings.Join(xs, ", ")
+}
+
+func treeFiles(root string) []string {
+	entries, err := os.ReadDir(root)
+	if err != nil {
+		return nil
+	}
+	var out []string
+	for _, e := range entries {
+		if treeFileRe.MatchString(e.Name()) {
+			out = append(out, e.Name())
+		}
+	}
+	sort.Strings(out)
+	return out
+}

cli/internal/issue/issue.go

@@ -0,0 +1,229 @@
+// Package issue is what an issue IS. The domain layer.
+//
+// It knows the canonical markdown format, the label taxonomy, validation, and
+// the dependency graph. It knows NOTHING about any tracker: no Gitea, no
+// logins, no HTTP, no issue numbers. The layering rule is mechanically checked
+// — see TestDomainImportsNothing, which walks this package's transitive
+// dependencies and fails on anything outside the standard library and
+// internal/project.
+//
+// Delete the transport entirely and this layer keeps working: issues that live
+// only on this machine are first-class, not drafts on their way somewhere.
+//
+// Identity is a slug derived from the title, and it is the only identity the
+// domain has. The file name is the id:
+//
+//	.tea/issues/wire-sqlc-appclick.md
+//
+//	---
+//	id: wire-sqlc-appclick
+//	state: open
+//	labels: [type/task, tech/sql]
+//	assignees: [naudachu]
+//	milestone: v0.2
+//	depends: [migrate-schema]
+//	origin: gitea
+//	gitea: owner/repo#42
+//	synced: 2026-08-07T18:40:00Z
+//	---
+//	# Wire sqlc into the appclick repo layer
+//
+//	## Summary
+//	...
+//
+// Keys down to origin are owned here. Everything below is written by the sync
+// layer; this package carries those keys through load/save verbatim in Extra
+// and never reads them. That passthrough is what lets one file represent both
+// a local issue and a synced one without the domain learning a second
+// vocabulary.
+//
+// Every metadata field is one line and lists are inline, so plain grep works
+// without a parser:
+//
+//	grep -l 'labels:.*type/bug' .tea/issues/*.md
+//	grep -ln 'depends:.*migrate-schema' .tea/issues/*.md   # who depends on it
+package issue
+
+import (
+	"fmt"
+	"regexp"
+	"strings"
+)
+
+// Origin is "does this issue exist anywhere but here" — a fact about the work,
+// so it is owned here. Its value is Local or a tracker's name; what that name
+// means, and the handle that goes with it (gitea: owner/repo#42), stay foreign
+// keys this layer carries but never reads.
+const Local = "local"
+
+// DomainKeys are the metadata fields this layer owns, in render order. Foreign
+// keys render after these, sorted, so the sync layer can add fields without
+// touching this list.
+var DomainKeys = []string{"id", "state", "labels", "assignees", "milestone",
+	"depends", "origin"}
+
+var listKeys = map[string]bool{"labels": true, "assignees": true, "depends": true}
+
+// States an issue may be in.
+var States = []string{"open", "closed"}
+
+var slugOK = regexp.MustCompile(`^[a-z0-9]+(-[a-z0-9]+)*$`)
+
+var slugPunct = regexp.MustCompile(`[^a-z0-9]+`)
+
+// Issue is one unit of work. Extra holds metadata this layer does not own.
+type Issue struct {
+	ID        string
+	Title     string
+	Body      string
+	State     string
+	Labels    []string
+	Assignees []string
+	Milestone string
+	Depends   []string
+	Origin    string
+	Extra     map[string]string
+}
+
+// IsLocal reports whether this issue exists nowhere but here.
+//
+// A complete state, not a pending one — and the state in which this file is
+// the only copy of the work. An issue whose Origin names somewhere else can be
+// fetched from there again; this one cannot.
+func (i *Issue) IsLocal() bool { return i.Origin == Local }
+
+// Type is the value of the mandatory, exclusive type/* label.
+func (i *Issue) Type() string { return i.namespaced("type/") }
+
+// Severity is the value of the optional, exclusive severity/* label.
+func (i *Issue) Severity() string { return i.namespaced("severity/") }
+
+func (i *Issue) namespaced(prefix string) string {
+	for _, l := range i.Labels {
+		if v, ok := strings.CutPrefix(l, prefix); ok {
+			return v
+		}
+	}
+	return ""
+}
+
+// FromText parses a stored issue. A non-empty id overrides the one in the
+// metadata block, which is how the store makes the file name authoritative.
+func FromText(text, id string) *Issue {
+	meta, title, body := ParseMeta(text)
+
+	extra := map[string]string{}
+	for k, v := range meta {
+		if !isDomainKey(k) {
+			extra[k] = v
+		}
+	}
+
+	if id == "" {
+		id = meta["id"]
+	}
+	milestone := meta["milestone"]
+	if milestone == "none" {
+		milestone = ""
+	}
+	state := meta["state"]
+	if state == "" {
+		state = "open"
+	}
+	origin := meta["origin"]
+	if origin == "" {
+		origin = Local
+	}
+
+	return &Issue{
+		ID:        id,
+		Title:     title,
+		Body:      strings.TrimSpace(body),
+		State:     state,
+		Labels:    splitList(meta["labels"]),
+		Assignees: splitList(meta["assignees"]),
+		Milestone: milestone,
+		Depends:   splitList(meta["depends"]),
+		Origin:    origin,
+		Extra:     extra,
+	}
+}
+
+// Text renders the issue back to its canonical file form.
+func (i *Issue) Text() string {
+	meta := map[string]string{}
+	for k, v := range i.Extra {
+		meta[k] = v
+	}
+	milestone := i.Milestone
+	if milestone == "" {
+		milestone = "none"
+	}
+	meta["id"] = i.ID
+	meta["state"] = i.State
+	meta["labels"] = renderList(i.Labels)
+	meta["assignees"] = renderList(i.Assignees)
+	meta["milestone"] = milestone
+	meta["depends"] = renderList(i.Depends)
+	meta["origin"] = i.Origin
+
+	body := strings.TrimSpace(i.Body)
+	if body == "" {
+		body = "(no body)"
+	}
+	return fmt.Sprintf("%s\n# %s\n\n%s\n", RenderMeta(meta), i.Title, body)
+}
+
+// Slugify turns a title into an id. Titles are English by format rule, so
+// ASCII is enough; anything else is dropped rather than transliterated.
+func Slugify(text string, maxLen int) string {
+	if maxLen <= 0 {
+		maxLen = 48
+	}
+	s := strings.Trim(slugPunct.ReplaceAllString(strings.ToLower(text), "-"), "-")
+	if len(s) > maxLen {
+		cut := s[:maxLen]
+		if i := strings.LastIndex(cut, "-"); i > 0 {
+			cut = cut[:i]
+		}
+		s = cut
+	}
+	s = strings.Trim(s, "-")
+	if s == "" {
+		return "issue"
+	}
+	return s
+}
+
+// IsSlug reports whether id is a well-formed identity.
+func IsSlug(id string) bool { return slugOK.MatchString(id) }
+
+// UniqueID is base, or base-2, base-3… when the slug is already used.
+func UniqueID(root, base string, taken []string) (string, error) {
+	used := map[string]bool{}
+	for _, t := range taken {
+		used[t] = true
+	}
+	for _, t := range AllIDs(root) {
+		used[t] = true
+	}
+	if !used[base] {
+		return base, nil
+	}
+	for n := 2; n < 1000; n++ {
+		cand := fmt.Sprintf("%s-%d", base, n)
+		if !used[cand] {
+			return cand, nil
+		}
+	}
+	return "", fmt.Errorf("cannot allocate an id for %q", base)
+}
+
+func isDomainKey(k string) bool {
+	for _, d := range DomainKeys {
+		if d == k {
+			return true
+		}
+	}
+	return false
+}

cli/internal/issue/issue_test.go

@@ -0,0 +1,137 @@
+package issue
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+)
+
+const sample = `---
+id: wire-sqlc-appclick
+state: open
+labels: [type/task, tech/sql]
+assignees: [naudachu]
+milestone: v0.2
+depends: [migrate-schema]
+origin: gitea
+gitea: claude-skills/tea#42
+synced: 2026-08-09T18:40:00Z
+---
+# Wire sqlc into the appclick repo layer
+
+## Summary
+Проводка sqlc.
+
+## Spec
+none
+
+## Motivation
+Ручной SQL расходится со схемой.
+
+## Acceptance criteria
+- [x] сгенерирован код
+- [ ] тесты зелёные
+`
+
+func TestFromTextReadsTheDomainAndCarriesTheRest(t *testing.T) {
+	i := FromText(sample, "")
+
+	if i.ID != "wire-sqlc-appclick" {
+		t.Errorf("id = %q", i.ID)
+	}
+	if i.Title != "Wire sqlc into the appclick repo layer" {
+		t.Errorf("title = %q", i.Title)
+	}
+	if want := []string{"type/task", "tech/sql"}; !reflect.DeepEqual(i.Labels, want) {
+		t.Errorf("labels = %v, want %v", i.Labels, want)
+	}
+	if i.Type() != "task" {
+		t.Errorf("type = %q", i.Type())
+	}
+	if i.IsLocal() {
+		t.Error("origin gitea must not read as local")
+	}
+	if i.Extra["gitea"] != "claude-skills/tea#42" {
+		t.Errorf("foreign key lost: %v", i.Extra)
+	}
+	// The domain carries foreign keys; it must not learn to read them.
+	if _, ok := i.Extra["labels"]; ok {
+		t.Error("a domain key leaked into Extra")
+	}
+	if strings.Contains(i.Body, "# "+i.Title) {
+		t.Error("the title heading was left in the body")
+	}
+	if !strings.HasPrefix(i.Body, SummarySection) {
+		t.Errorf("body does not start at ## Summary: %q", head(i.Body))
+	}
+}
+
+func TestTextRoundTripsByteForByte(t *testing.T) {
+	if got := FromText(sample, "").Text(); got != sample {
+		t.Errorf("round trip changed the file:\n--- got ---\n%s\n--- want ---\n%s", got, sample)
+	}
+}
+
+func TestMilestoneNoneIsTheEmptyMilestone(t *testing.T) {
+	i := FromText("---\nid: x\nmilestone: none\n---\n# T\n\nbody\n", "")
+	if i.Milestone != "" {
+		t.Errorf("milestone = %q, want empty", i.Milestone)
+	}
+	if !strings.Contains(i.Text(), "milestone: none") {
+		t.Error("an empty milestone must render back as none")
+	}
+}
+
+func TestBareListValueIsTheSameStatementAsABracketedOne(t *testing.T) {
+	i := FromText("---\nid: x\nlabels: type/bug\n---\n# T\n\nbody\n", "")
+	if want := []string{"type/bug"}; !reflect.DeepEqual(i.Labels, want) {
+		t.Errorf("labels = %v, want %v", i.Labels, want)
+	}
+}
+
+func TestFileNameWinsOverTheMetadataID(t *testing.T) {
+	// The store names the file after the slug, so a hand-edited `id:` that
+	// disagrees with it is the one that is wrong.
+	if got := FromText(sample, "renamed-by-hand").ID; got != "renamed-by-hand" {
+		t.Errorf("id = %q", got)
+	}
+}
+
+func TestSlugify(t *testing.T) {
+	cases := map[string]string{
+		// Truncation cuts back to the last dash, so a slug never ends in half
+		// a word — even when the limit happened to land on a boundary.
+		"Wire sqlc into the appclick repo layer": "wire-sqlc-into-the-appclick",
+		"Fix tea-guard crash":                    "fix-tea-guard-crash",
+		"  Trailing --- dashes  ":                "trailing-dashes",
+		// Titles are English by format rule; anything else is dropped rather
+		// than transliterated, and an empty result is not an id.
+		"Крашится гвард": "issue",
+		"":               "issue",
+	}
+	for in, want := range cases {
+		if got := Slugify(in, 32); got != want {
+			t.Errorf("Slugify(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+func head(s string) string {
+	if len(s) > 40 {
+		return s[:40]
+	}
+	return s
+}
+
+func TestSectionBodyStopsAtTheNextHeading(t *testing.T) {
+	body := "## Summary\nодин\nдва\n\n## Spec\nnone\n"
+	if got := SectionBody(body, "## Summary"); got != "один\nдва" {
+		t.Errorf("Summary = %q", got)
+	}
+	if got := SectionBody(body, SpecSection); got != "none" {
+		t.Errorf("Spec = %q", got)
+	}
+	if got := SectionBody(body, "## Missing"); got != "" {
+		t.Errorf("missing section = %q, want empty", got)
+	}
+}

cli/internal/issue/layering_test.go

@@ -0,0 +1,55 @@
+package issue
+
+import (
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+// The domain must depend on nothing but the standard library and the one
+// package that answers "which directory is the project".
+//
+// In Python this rule was a grep in a document and a habit; here it is a build
+// graph, and the test fails the moment a tracker concept — an HTTP client, a
+// JSON payload, a login — is imported into the layer that must not know a
+// tracker exists.
+func TestDomainDependsOnNothing(t *testing.T) {
+	const allowed = "git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+
+	out, err := exec.Command("go", "list", "-deps", ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	for _, dep := range strings.Fields(string(out)) {
+		if dep == allowed || dep == "git.noodles.cam/claude-skills/marketplace/cli/internal/issue" {
+			continue
+		}
+		// A standard-library import path has no dot in its first element,
+		// because it has no domain name in front of it.
+		first, _, _ := strings.Cut(dep, "/")
+		if strings.Contains(first, ".") {
+			t.Errorf("the domain imports %s — a tracker concept in the layer that must not know one exists", dep)
+		}
+	}
+}
+
+// The other half of the same rule: net/http and its friends are standard
+// library, so "no third-party imports" would not catch a transport written by
+// hand. Name them.
+func TestDomainDoesNotReachTheNetworkOrTheShell(t *testing.T) {
+	forbidden := []string{"net/http", "net", "os/exec", "encoding/json"}
+
+	out, err := exec.Command("go", "list", "-deps", ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	deps := map[string]bool{}
+	for _, d := range strings.Fields(string(out)) {
+		deps[d] = true
+	}
+	for _, f := range forbidden {
+		if deps[f] {
+			t.Errorf("the domain reaches %s — that belongs in the transport", f)
+		}
+	}
+}

cli/internal/issue/meta.go

@@ -0,0 +1,80 @@
+package issue
+
+import (
+	"regexp"
+	"sort"
+	"strings"
+)
+
+var titleRe = regexp.MustCompile(`^#[ \t]+(.+?)[ \t]*\n`)
+
+// ParseMeta splits a file into its metadata block, title, and body.
+//
+// Values come back as the raw text that followed the colon. Lists are not
+// unpacked here: a foreign key that happens to look like a list must round
+// trip byte for byte, and the domain's own lists are unpacked by their
+// accessors. The title is the first `# ` heading below the block and is
+// stripped out of the body.
+func ParseMeta(text string) (meta map[string]string, title, body string) {
+	meta = map[string]string{}
+	rest := text
+	if strings.HasPrefix(text, "---") {
+		if end := strings.Index(text[3:], "\n---"); end != -1 {
+			end += 3
+			for _, line := range strings.Split(strings.TrimSpace(text[3:end]), "\n") {
+				k, v, ok := strings.Cut(line, ":")
+				if !ok {
+					continue
+				}
+				meta[strings.TrimSpace(k)] = strings.TrimSpace(v)
+			}
+			rest = text[end+4:]
+		}
+	}
+	rest = strings.TrimLeft(rest, "\n")
+
+	if m := titleRe.FindStringSubmatchIndex(rest); m != nil {
+		title = strings.TrimSpace(rest[m[2]:m[3]])
+		rest = strings.TrimLeft(rest[m[1]:], "\n")
+	}
+	return meta, title, rest
+}
+
+// RenderMeta writes the block back: domain keys in DomainKeys order, foreign
+// keys after them, sorted. Lists stay on one line so grep sees them whole.
+func RenderMeta(meta map[string]string) string {
+	var foreign []string
+	for k := range meta {
+		if !isDomainKey(k) {
+			foreign = append(foreign, k)
+		}
+	}
+	sort.Strings(foreign)
+
+	lines := []string{"---"}
+	for _, k := range append(append([]string{}, DomainKeys...), foreign...) {
+		if v, ok := meta[k]; ok {
+			lines = append(lines, k+": "+v)
+		}
+	}
+	return strings.Join(append(lines, "---"), "\n")
+}
+
+// splitList unpacks the inline `[a, b]` form, and a bare comma-separated value
+// too: a hand-written `labels: type/bug` is the same statement as
+// `labels: [type/bug]` and the format does not make an operator care.
+func splitList(v string) []string {
+	v = strings.TrimSpace(v)
+	if strings.HasPrefix(v, "[") && strings.HasSuffix(v, "]") {
+		v = v[1 : len(v)-1]
+	}
+	var out []string
+	for _, part := range strings.Split(v, ",") {
+		if part = strings.TrimSpace(part); part != "" {
+			out = append(out, part)
+		}
+	}
+	return out
+}
+
+func renderList(xs []string) string { return "[" + strings.Join(xs, ", ") + "]" }

cli/internal/issue/store.go

@@ -0,0 +1,209 @@
+package issue
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/project"
+)
+
+// The store holds two kinds of file, and only one of them is a store.
+//
+// An issue whose origin is Local lives here and nowhere else — that file IS the
+// issue, and losing it loses the work. Anything with a tracker origin is a
+// cache: the tracker has it, this copy is a working copy, and it is deleted the
+// moment a push confirms the tracker is up to date.
+
+// Root resolves the issue store for the current project. An explicit out
+// overrides it and is used exactly as typed: a relative out stays relative to
+// the working directory, because that is what the operator asked for.
+func Root(out string) string {
+	if out != "" {
+		return out
+	}
+	return project.StoreRoot("")
+}
+
+// ErrStoreMissing marks the "the store directory is not there" failure.
+//
+// Deliberately a different answer from "the store is empty". One is a path that
+// does not exist, the other is a repository with no issues filed yet, and
+// conflating the two is exactly what made a missed directory look like an empty
+// backlog.
+var ErrStoreMissing = errors.New("store missing")
+
+// StoreExists reports whether root is a directory that can be read as a store.
+func StoreExists(root string) bool {
+	if root == "" {
+		return false
+	}
+	fi, err := os.Stat(root)
+	return err == nil && fi.IsDir()
+}
+
+// RequireStore asserts the store is there before reading or writing it.
+//
+// An empty root means no project was found at all — a different failure from a
+// project whose store has not been created yet, and the message says so.
+func RequireStore(root string) error {
+	if root == "" {
+		return fmt.Errorf("%w: %s", ErrStoreMissing, project.NotFoundError(""))
+	}
+	if !StoreExists(root) {
+		return fmt.Errorf("%w: store %s does not exist", ErrStoreMissing, root)
+	}
+	return nil
+}
+
+// CreateStore creates the store, reporting whether it made the directory.
+//
+// Only the commands that legitimately bootstrap a store call this — `new` and
+// `pull` — and both announce it. Nothing creates a store as a side effect of a
+// write: a missing directory is something to report, not something to conjure.
+// An unresolved root is never conjured either — without a marker there is no
+// project to create a store IN, and guessing one is how a store once ended up
+// inside the plugin.
+func CreateStore(root string) (bool, error) {
+	if root == "" {
+		return false, fmt.Errorf("%w: %s", ErrStoreMissing, project.NotFoundError(""))
+	}
+	if StoreExists(root) {
+		return false, nil
+	}
+	if err := os.MkdirAll(root, 0o755); err != nil {
+		return false, err
+	}
+	return true, nil
+}
+
+// StoreError says why root cannot be read as a store, or nil when it holds
+// issues.
+//
+// The three messages are distinct on purpose — no project at all, a project
+// with no store, and a store with nothing in it are three different things to
+// do next.
+func StoreError(root string) error {
+	switch {
+	case root == "":
+		return project.NotFoundError("")
+	case !StoreExists(root):
+		return fmt.Errorf("store %s does not exist — nothing was created; pass --out to point elsewhere", root)
+	case len(AllIDs(root)) == 0:
+		return fmt.Errorf("store %s exists but is empty", root)
+	}
+	return nil
+}
+
+// PathOf is where the issue with this id lives.
+func PathOf(root, id string) string { return filepath.Join(root, id+".md") }
+
+// AllIDs lists every issue in the store, by slug.
+//
+// An issue file is named by its slug and a slug has no dot in it, so
+// `<id>.comments.md` — the thread the sync layer parks beside an issue — is not
+// one, and neither is anything else that grew a second extension. Without that
+// rule `wire-sqlc.comments` reads as an issue called `wire-sqlc.comments`, and
+// a bare push tries to file the comment thread as a unit of work.
+func AllIDs(root string) []string {
+	if !StoreExists(root) {
+		return nil
+	}
+	entries, err := os.ReadDir(root)
+	if err != nil {
+		return nil
+	}
+	var out []string
+	for _, e := range entries {
+		name := e.Name()
+		if !strings.HasSuffix(name, ".md") {
+			continue
+		}
+		if strings.HasPrefix(name, ".") || strings.HasPrefix(name, "INDEX") ||
+			strings.HasPrefix(name, "tree-") {
+			continue
+		}
+		id := name[:len(name)-3]
+		if strings.Contains(id, ".") {
+			continue
+		}
+		out = append(out, id)
+	}
+	sort.Strings(out)
+	return out
+}
+
+// SlugFiles lists every file the store holds under one slug — the issue and its
+// sidecars.
+//
+// `<id>.md` is the issue. Anything named `<id>.<something>` beside it is a
+// companion another layer parked there (`<id>.comments.md` is the one that
+// exists today). AllIDs already refuses to read those as issues because a slug
+// has no dot in it; this is the same rule read the other way round.
+//
+// Which is how the domain can remove an issue completely without learning what
+// any of those companions are: it does not need to know that a comment thread
+// exists to know that a file named after this issue belongs to it and goes when
+// it goes. The issue's own file comes first — it is the headline of any receipt
+// printed from this list.
+//
+// A missing store is an empty list, not an error: nothing is there to remove.
+func SlugFiles(root, id string) []string {
+	entries, err := os.ReadDir(root)
+	if err != nil {
+		return nil
+	}
+	prefix, own := id+".", id+".md"
+	var self, sidecars []string
+	for _, e := range entries {
+		name := e.Name()
+		if !strings.HasPrefix(name, prefix) || e.IsDir() {
+			continue
+		}
+		p := filepath.Join(root, name)
+		if name == own {
+			self = append(self, p)
+		} else {
+			sidecars = append(sidecars, p)
+		}
+	}
+	sort.Strings(sidecars)
+	return append(self, sidecars...)
+}
+
+// Load reads one issue. The file name wins over the id in the metadata block.
+func Load(root, id string) (*Issue, error) {
+	raw, err := os.ReadFile(PathOf(root, id))
+	if err != nil {
+		return nil, err
+	}
+	return FromText(string(raw), id), nil
+}
+
+// LoadAll reads the whole store.
+func LoadAll(root string) (map[string]*Issue, error) {
+	out := map[string]*Issue{}
+	for _, id := range AllIDs(root) {
+		i, err := Load(root, id)
+		if err != nil {
+			return nil, err
+		}
+		out[id] = i
+	}
+	return out, nil
+}
+
+// Save writes an issue to the store, which must already exist.
+func Save(root string, i *Issue) (string, error) {
+	if err := RequireStore(root); err != nil {
+		return "", err
+	}
+	p := PathOf(root, i.ID)
+	if err := os.WriteFile(p, []byte(i.Text()), 0o644); err != nil {
+		return "", err
+	}
+	return p, nil
+}

cli/internal/issue/taxonomy.go

@@ -0,0 +1,116 @@
+package issue
+
+import "strings"
+
+// Four namespaces classify an issue. type/* is mandatory and exclusive,
+// severity/* is optional and exclusive, tech/* and comp/* are free-form.
+//
+// Colors are NOT here — a hex code is how a tracker paints a chip, which makes
+// it the sync layer's business.
+
+// Types are the kinds of work, and the order is the order they are offered in.
+var Types = []struct{ Name, Meaning string }{
+	{"bug", "Something behaves incorrectly in existing code"},
+	{"task", "Implementation of new functionality"},
+	{"refactor", "Internal restructuring; behavior must not change"},
+	{"test", "Writing or fixing tests"},
+	{"feature", "Container: several issues delivering one unit of business value"},
+	{"draft", "Idea captured for later; not ready for work"},
+}
+
+// Severities are the business-impact levels, ascending.
+var Severities = []string{"low", "medium", "high", "showstopper", "critical"}
+
+// Section headers are fixed English literals in a fixed order; only body prose
+// is Russian.
+const (
+	SummarySection = "## Summary"
+	SpecSection    = "## Spec"
+	ACSection      = "## Acceptance criteria"
+	DependsSection = "## Depends on"
+	IssuesSection  = "## Issues"
+)
+
+// RequiredSections must be present in every type. type/draft is exempt from
+// acceptance criteria and only from that.
+var RequiredSections = []string{SummarySection, SpecSection}
+
+// DepSections both name what an issue depends on, so both are edge sources and
+// both point the same way. In a type/feature that reads container -> child:
+// "the container is closed when its children are closed" IS a dependency.
+// "a child belongs to a feature" is membership, and membership has no place in
+// a dependency graph — which is why a child never names its container back.
+var DepSections = []string{DependsSection, IssuesSection}
+
+// ExpectedSections are the per-type sections from the templates. Absence is a
+// warning, not a stop.
+var ExpectedSections = map[string][]string{
+	"bug":      {"## Steps to reproduce", "## Expected", "## Actual", "## Environment"},
+	"task":     {"## Motivation"},
+	"refactor": {"## Motivation", "## Invariants"},
+	"test":     {"## Motivation", "## Test cases"},
+	"feature":  {"## Motivation", IssuesSection},
+	"draft":    {"## Notes"},
+}
+
+// KnownType reports whether name is one of Types.
+func KnownType(name string) bool {
+	for _, t := range Types {
+		if t.Name == name {
+			return true
+		}
+	}
+	return false
+}
+
+// KnownSeverity reports whether name is one of Severities.
+func KnownSeverity(name string) bool {
+	for _, s := range Severities {
+		if s == name {
+			return true
+		}
+	}
+	return false
+}
+
+// TypeNames lists the type slugs, for error messages and completion.
+func TypeNames() []string {
+	out := make([]string, len(Types))
+	for i, t := range Types {
+		out[i] = t.Name
+	}
+	return out
+}
+
+// CanonicalLabels is the label set a tracker needs before a push can attach
+// anything: the two exclusive namespaces in full. tech/* and comp/* are
+// project-specific and have no preset.
+func CanonicalLabels() []string {
+	out := make([]string, 0, len(Types)+len(Severities))
+	for _, t := range Types {
+		out = append(out, "type/"+t.Name)
+	}
+	for _, s := range Severities {
+		out = append(out, "severity/"+s)
+	}
+	return out
+}
+
+// SectionBody is the text under header, up to the next `## ` heading.
+func SectionBody(body, header string) string {
+	var out []string
+	active := false
+	for _, line := range strings.Split(body, "\n") {
+		if strings.HasPrefix(line, "## ") {
+			if active {
+				break
+			}
+			active = strings.TrimSpace(line) == header
+			continue
+		}
+		if active {
+			out = append(out, line)
+		}
+	}
+	return strings.TrimSpace(strings.Join(out, "\n"))
+}

cli/internal/issue/template.go

@@ -0,0 +1,132 @@
+package issue
+
+import "strings"
+
+// The type templates, verbatim from references/format.md.
+//
+// Section headers are fixed English literals in a fixed order; body prose is
+// Russian. Both halves of that rule are in the strings below, and the format
+// document is the source of truth for them.
+
+const specSection = "## Spec\nnone\n"
+
+var templates = map[string]string{
+	"bug": `## Summary
+Что сломано и где проявляется, одно-два предложения.
+
+` + specSection + `
+## Steps to reproduce
+1. …
+2. …
+
+## Expected
+Что должно было произойти.
+
+## Actual
+Что происходит на самом деле: вывод команды, лог.
+
+## Environment
+Только релевантное: версии, ОС, конфигурация.
+
+## Acceptance criteria
+- [ ] баг не воспроизводится по шагам выше
+- [ ] добавлена проверка на регрессию (если применимо)
+`,
+	"task": `## Summary
+Что нужно сделать, одно-два предложения.
+
+` + specSection + `
+## Motivation
+Какую проблему пользователя/системы это решает.
+
+## Acceptance criteria
+- [ ] проверяемое условие
+- [ ] …
+`,
+	"refactor": `## Summary
+Что перестраиваем и в каких файлах (` + "`path/file:line`" + `).
+
+` + specSection + `
+## Motivation
+Чем плохо текущее состояние: дублирование, связность, читаемость.
+
+## Invariants
+Что НЕ должно измениться: поведение, публичные API, форматы данных.
+
+## Acceptance criteria
+- [ ] проверяемое условие (тесты зелёные, старый путь удалён, …)
+`,
+	"test": `## Summary
+Что покрываем тестами и где (` + "`path/file:line`" + `).
+
+` + specSection + `
+## Motivation
+Зачем: регрессия после бага, пробел в покрытии, флаки-тест.
+
+## Test cases
+- сценарий → ожидаемый результат
+- …
+
+## Acceptance criteria
+- [ ] перечисленные кейсы покрыты и зелёные
+- [ ] тесты проходят в CI
+`,
+	"feature": `## Summary
+Бизнес-ценность одним-двумя предложениями.
+
+` + specSection + `
+## Motivation
+Какую проблему пользователя/системы это решает.
+
+## Issues
+- [ ] slug-дочернего-issue — краткое описание части
+- [ ] …
+
+## Acceptance criteria
+- [ ] все дочерние issues закрыты
+- [ ] проверяемое условие уровня фичи
+`,
+	"draft": `## Summary
+Идея одним-двумя предложениями.
+
+` + specSection + `
+## Notes
+Свободные заметки: что известно, открытые вопросы, варианты.
+`,
+}
+
+// Template is the prefilled body for a type, with `## Depends on` inserted
+// right after `## Spec` when the issue has dependencies.
+func Template(typ string, depends []string) string {
+	return withDepends(templates[typ], depends)
+}
+
+// withDepends places the section where the format says it goes: after
+// `## Spec`, before everything else. Appended at the end only when the
+// template has no third section to sit in front of.
+func withDepends(body string, depends []string) string {
+	if len(depends) == 0 {
+		return body
+	}
+	var b strings.Builder
+	b.WriteString("## Depends on\n")
+	for _, d := range depends {
+		b.WriteString("- " + d + "\n")
+	}
+	block := b.String()
+
+	var out []string
+	placed := false
+	for _, line := range strings.SplitAfter(body, "\n") {
+		if !placed && len(out) > 0 && strings.HasPrefix(line, "## ") &&
+			!strings.HasPrefix(line, SummarySection) && !strings.HasPrefix(line, SpecSection) {
+			out = append(out, block+"\n")
+			placed = true
+		}
+		out = append(out, line)
+	}
+	if !placed {
+		out = append(out, "\n"+block)
+	}
+	return strings.Join(out, "")
+}

cli/internal/issue/validate.go

@@ -0,0 +1,131 @@
+package issue
+
+import (
+	"fmt"
+	"regexp"
+	"strings"
+)
+
+var (
+	titlePrefixRe = regexp.MustCompile(
+		`(?i)^\s*(\[[^\]]+\]|(fix|feat|feature|bug|task|test|chore|refactor)\s*:)`)
+	cyrillicRe = regexp.MustCompile(`(?i)[а-яё]`)
+)
+
+// Validate reports what is wrong with an issue.
+//
+// Errors mean the issue is not well-formed in the canonical format; warnings
+// mean it deviates from its type template. Pass knownIDs to have dependencies
+// resolved against a store; pass nil to skip that check.
+func Validate(i *Issue, knownIDs map[string]bool) (errs, warns []string) {
+	switch {
+	case i.ID == "":
+		errs = append(errs, "no `id:` — the slug is the issue's identity")
+	case !IsSlug(i.ID):
+		errs = append(errs, fmt.Sprintf("id %q is not a slug (lowercase, digits, single dashes)", i.ID))
+	}
+
+	if !contains(States, i.State) {
+		errs = append(errs, fmt.Sprintf("state %q must be one of: %s",
+			i.State, strings.Join(States, ", ")))
+	}
+
+	var types []string
+	severities := 0
+	for _, l := range i.Labels {
+		if strings.HasPrefix(l, "type/") {
+			types = append(types, l)
+		}
+		if strings.HasPrefix(l, "severity/") {
+			severities++
+		}
+	}
+	switch {
+	case len(types) != 1:
+		found := strings.Join(types, ", ")
+		if found == "" {
+			found = "none"
+		}
+		errs = append(errs, fmt.Sprintf("need exactly one type/* label, found %d: %s",
+			len(types), found))
+	case !KnownType(i.Type()):
+		errs = append(errs, fmt.Sprintf("unknown type %q — known: %s",
+			i.Type(), strings.Join(TypeNames(), ", ")))
+	}
+	if severities > 1 {
+		errs = append(errs, "at most one severity/* label")
+	}
+	if s := i.Severity(); s != "" && !KnownSeverity(s) {
+		warns = append(warns, fmt.Sprintf("unknown severity %q", s))
+	}
+
+	if i.Title == "" {
+		errs = append(errs, "no `# Title` heading below the metadata block")
+	} else {
+		if titlePrefixRe.MatchString(i.Title) {
+			head := i.Title
+			if len(head) > 24 {
+				head = head[:24]
+			}
+			errs = append(errs, fmt.Sprintf(
+				"title carries a type prefix (%q) — the type lives in the label", head))
+		}
+		if cyrillicRe.MatchString(i.Title) {
+			errs = append(errs, "title must be English, imperative mood (prose stays Russian)")
+		}
+	}
+
+	for _, h := range RequiredSections {
+		if !strings.Contains(i.Body, h) {
+			errs = append(errs, "missing section "+h)
+		}
+	}
+	if i.Type() != "draft" && !strings.Contains(i.Body, ACSection) {
+		errs = append(errs, "missing section "+ACSection)
+	}
+	if strings.Contains(i.Body, SpecSection) && SectionBody(i.Body, SpecSection) == "" {
+		errs = append(errs, "## Spec is empty — put a repo path, a URL, or the literal `none`")
+	}
+
+	for _, h := range ExpectedSections[i.Type()] {
+		if !strings.Contains(i.Body, h) {
+			warns = append(warns, fmt.Sprintf("type/%s template usually has %s", i.Type(), h))
+		}
+	}
+
+	if contains(i.Depends, i.ID) {
+		errs = append(errs, "depends on itself")
+	}
+	if knownIDs != nil {
+		for _, d := range i.Depends {
+			if !knownIDs[d] {
+				warns = append(warns, fmt.Sprintf("depends on %q, which is not in the store", d))
+			}
+		}
+	}
+
+	// `depends:` is the machine-readable graph; the body section is prose for
+	// humans. They drift silently unless something says so. Name the section
+	// the reference actually came from — for a container that is `## Issues`.
+	for _, r := range BodyDepRefs(i.Body) {
+		if !strings.HasPrefix(r.Ref, "#") && !contains(i.Depends, r.Ref) {
+			warns = append(warns, fmt.Sprintf(
+				"%s mentions %q but `depends:` does not list it", r.Section, r.Ref))
+		}
+	}
+
+	// An unticked checkbox is never a finding — neither an error nor a warning.
+	// `- [ ]` is work not done yet, which is the normal state of a perfectly
+	// well-formed issue. Reading that state is the `ac` command's job.
+
+	return errs, warns
+}
+
+func contains(xs []string, x string) bool {
+	for _, v := range xs {
+		if v == x {
+			return true
+		}
+	}
+	return false
+}

cli/internal/mapping/frompayload.go

@@ -0,0 +1,207 @@
+package mapping
+
+import (
+	"slices"
+	"strconv"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// Gitea -> domain.
+
+// PayloadOptions are the things a caller knows and this package cannot: what
+// the store already holds, what the tracker's numbers mean locally, and what
+// time it is.
+type PayloadOptions struct {
+	// IDForNumber maps a Gitea number to a local slug. A dependency whose
+	// target has not been pulled yet is dropped from `depends:` rather than
+	// invented — the body still names it, so nothing is lost, and a made-up
+	// slug would be an edge to a file that does not exist.
+	IDForNumber map[int]string
+	// ExtraNumbers are dependencies the caller learned somewhere other than the
+	// body, folded in with the ones the body names.
+	ExtraNumbers []int
+	// Synced is the timestamp stamped into `synced:`. The clock belongs to the
+	// caller: a package with a clock in it is not a pure one.
+	Synced string
+	// LocalBody is the body of the copy already in the store, when there is
+	// one. It contributes exactly one thing — its ticked checkboxes survive the
+	// overwrite. Empty is what a first pull passes.
+	LocalBody string
+}
+
+// FromPayload builds a domain issue from a Gitea issue payload, and returns the
+// numbers it could not resolve to a slug.
+//
+// The id marker is stripped before anything else looks at the body: it is
+// transport bookkeeping, and the caller has already read the slug off it to
+// decide which id to pass. Everything downstream — checkboxes, `#N` references,
+// what lands on disk — sees the body the author wrote.
+func FromPayload(p *wire.Issue, id string, repo wire.Repo, opt PayloadOptions) (*issue.Issue, []int) {
+	body := MergeCheckboxState(StripIDMarker(strings.TrimSpace(p.Body)), opt.LocalBody)
+
+	numbers := NumbersInBody(body)
+	for _, n := range opt.ExtraNumbers {
+		if !slices.Contains(numbers, n) {
+			numbers = append(numbers, n)
+		}
+	}
+
+	// A number that resolves to this issue itself is dropped without a word: a
+	// body may well name its own number, and a self-edge is a cycle the graph
+	// would report as an error the author cannot fix.
+	var deps []string
+	var unresolved []int
+	for _, n := range numbers {
+		slug := opt.IDForNumber[n]
+		switch {
+		case slug != "" && slug != id && !slices.Contains(deps, slug):
+			deps = append(deps, slug)
+		case slug == "":
+			unresolved = append(unresolved, n)
+		}
+	}
+
+	// The repository the caller asked for, never the one the payload names: a
+	// dependency listing answers with issues from elsewhere, and this is the
+	// handle for the copy landing in THIS store.
+	extra := map[string]string{
+		GiteaKey:  wire.Key{Repo: repo, Number: p.Number}.String(),
+		URLKey:    p.HTMLURL,
+		SyncedKey: opt.Synced,
+	}
+	if p.Ref != "" {
+		extra[BranchKey] = p.Ref
+	}
+	if p.UpdatedAt != "" {
+		extra[RemoteUpdatedKey] = p.UpdatedAt
+	}
+	// Zero comments is not a fact worth a line in the file — every issue that
+	// has never been discussed would carry one.
+	if p.Comments > 0 {
+		extra[CommentsKey] = strconv.Itoa(p.Comments)
+	}
+
+	state := p.State
+	if state == "" {
+		state = "open"
+	}
+
+	// Appended into nil slices, so an issue with no labels is the same value as
+	// one loaded from a file — the store's own parser yields nothing, not an
+	// empty list, and two spellings of "none" is a comparison bug waiting.
+	var labels []string
+	for _, l := range p.Labels {
+		labels = append(labels, l.Name)
+	}
+	var assignees []string
+	for _, a := range p.Assignees {
+		assignees = append(assignees, a.Login)
+	}
+	milestone := ""
+	if p.Milestone != nil {
+		milestone = p.Milestone.Title
+	}
+
+	return &issue.Issue{
+		ID:        id,
+		Title:     p.Title,
+		Body:      body,
+		State:     state,
+		Labels:    labels,
+		Assignees: assignees,
+		Milestone: milestone,
+		Depends:   deps,
+		Origin:    Origin,
+		Extra:     extra,
+	}, unresolved
+}
+
+// NumbersInBody is every `#N` referenced from the body's dependency sections.
+// Used only to seed `depends:` on the first pull — after that the metadata
+// field is the graph and the prose is prose.
+func NumbersInBody(body string) []int {
+	var out []int
+	for _, ref := range issue.BodyDepRefs(body) {
+		if !strings.HasPrefix(ref.Ref, "#") {
+			continue
+		}
+		if n, err := strconv.Atoi(ref.Ref[1:]); err == nil {
+			out = append(out, n)
+		}
+	}
+	return out
+}
+
+// MergeCheckboxState is the remote body with every tick the local copy already
+// had put back.
+//
+// The one exception to "a pull overwrites the body", and deliberately the
+// narrowest one that works. A tick is MONOTONE — an item only ever travels
+// `[ ]` -> `[x]` — so the two sides are joined by a set union, not reconciled:
+// no base version, no drift tracking, no conflict to resolve. The set is a set
+// of item TEXTS, and an item comes out ticked when either side has it ticked.
+// Everything else in the body is still the remote's word.
+//
+// Matching is on Checkbox.Text, which the domain parser has already stripped
+// and rejoined with single spaces, so rewrapping a long item does not cost it
+// its tick. It is otherwise literal: reword an item and it is a different item
+// — the tick stays with the wording it was put on.
+//
+// THE SAME TEXT MORE THAN ONCE is read as the rule says, as a set: one ticked
+// local item ticks every remote item with that text. The alternative — pairing
+// duplicates up by order — is the reading that can still drop a tick (local
+// `[ ]` then `[x]`, remote a single line: the ticked one pairs with nothing),
+// and dropping a tick is the bug this exists to fix. Two items whose text is
+// identical are the same item to whoever reads them.
+//
+// The price, accepted explicitly: UNticking is not monotone, so a box unticked
+// in the web UI comes back on the next pull. Untick locally, push.
+func MergeCheckboxState(remoteBody, localBody string) string {
+	ticked := map[string]bool{}
+	for _, c := range issue.Checkboxes(localBody) {
+		if c.Checked {
+			ticked[c.Text] = true
+		}
+	}
+	if len(ticked) == 0 {
+		return remoteBody
+	}
+	body := remoteBody
+	// SetCheckbox trades one character for one character, so line numbers read
+	// off remoteBody stay valid against the partially rewritten body.
+	for _, c := range issue.Checkboxes(remoteBody) {
+		if c.Checked || !ticked[c.Text] {
+			continue
+		}
+		// The line was just read off remoteBody by the same parser, so this
+		// cannot fail; if it ever did, one unticked item is a smaller loss than
+		// abandoning the merge and dropping every other tick with it.
+		if next, err := issue.SetCheckbox(body, c.Line, true); err == nil {
+			body = next
+		}
+	}
+	return body
+}
+
+// RenderComments flattens a comment thread to markdown. Read-only: nothing
+// writes it back, which is why it may be as lossy as a reader needs.
+func RenderComments(comments []wire.Comment) string {
+	var out []string
+	for _, c := range comments {
+		day := c.CreatedAt
+		if len(day) > 10 {
+			day = day[:10]
+		}
+		body := strings.TrimSpace(c.Body)
+		if body == "" {
+			body = "(empty)"
+		}
+		out = append(out,
+			"## comment "+strconv.FormatInt(c.ID, 10)+" — "+c.User.Login+" — "+day,
+			"", body, "")
+	}
+	return strings.Join(out, "\n")
+}

cli/internal/mapping/labels.go

@@ -0,0 +1,123 @@
+package mapping
+
+import (
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// How the taxonomy is painted in Gitea's UI. A hex code says nothing about what
+// an issue IS, which is exactly why the table lives here and not in the domain
+// — internal/issue/taxonomy.go says as much where the labels themselves are.
+//
+// The keys are the canonical set and nothing else. TestEveryCanonicalLabelHasA
+// Color walks issue.CanonicalLabels() and fails on a gap, so a type or a
+// severity added over there cannot quietly arrive here as grey.
+var labelColors = map[string]string{
+	"type/bug":             "#ee0701",
+	"type/task":            "#0e8a16",
+	"type/refactor":        "#1d76db",
+	"type/test":            "#fbca04",
+	"type/feature":         "#5319e7",
+	"type/draft":           "#cccccc",
+	"severity/low":         "#c2e0c6",
+	"severity/medium":      "#fbca04",
+	"severity/high":        "#eb6420",
+	"severity/showstopper": "#ee0701",
+	"severity/critical":    "#b60205",
+}
+
+// DefaultColor paints everything outside the canonical set. `tech/*` and
+// `comp/*` are project-specific and have no preset, so guessing a color for one
+// would be inventing a meaning it does not have.
+const DefaultColor = "#ededed"
+
+// LabelColor is the hex code a label is painted with in the tracker.
+func LabelColor(name string) string {
+	if c, ok := labelColors[name]; ok {
+		return c
+	}
+	return DefaultColor
+}
+
+// LabelSpecs is the request body for each name, in the order given.
+//
+// A wire.LabelRequest and not a shape of this package's own: it is field for
+// field what a label create takes, and a second spelling of it would mean the
+// bootstrap command copying four fields across on its way to the transport.
+// Exclusivity and meaning come from the domain taxonomy; only the color is
+// decided here.
+//
+// A slice and not a map: the order is the taxonomy's, and a bootstrap prints
+// its plan in that order — a map would shuffle the plan on every run and make
+// two identical runs look like different ones.
+func LabelSpecs(names []string) []wire.LabelRequest {
+	ns := exclusiveNamespaces()
+	out := make([]wire.LabelRequest, 0, len(names))
+	for _, name := range names {
+		out = append(out, wire.LabelRequest{
+			Name:        name,
+			Color:       LabelColor(name),
+			Description: typeMeaning(name),
+			Exclusive:   hasAnyPrefix(name, ns),
+		})
+	}
+	return out
+}
+
+// CanonicalLabelSpecs is the set a repository needs before a push can attach
+// anything.
+//
+// Derived from the domain's own list rather than restated: add a type over in
+// the taxonomy and the next bootstrap creates it, with no line changing here
+// except the color it is painted with.
+func CanonicalLabelSpecs() []wire.LabelRequest { return LabelSpecs(issue.CanonicalLabels()) }
+
+// exclusiveNamespaces are the namespaces at most one label may come from, read
+// off the canonical set rather than listed again — the domain publishes exactly
+// the exclusive namespaces there, in full, and that is what makes the set
+// canonical.
+//
+// A prefix test and not a membership test, on purpose: a project's own
+// `type/spike` is still exclusive. Being one of a set of alternatives is a
+// property of the namespace, not of the members the taxonomy happens to know.
+func exclusiveNamespaces() []string {
+	var out []string
+	seen := map[string]bool{}
+	for _, name := range issue.CanonicalLabels() {
+		ns, _, ok := strings.Cut(name, "/")
+		if !ok || seen[ns] {
+			continue
+		}
+		seen[ns] = true
+		out = append(out, ns+"/")
+	}
+	return out
+}
+
+// typeMeaning is the description a `type/*` label carries into the tracker, so
+// the meaning a reader needs is on the chip rather than in this repository.
+// Nothing else gets one: a severity explains itself, and a project's own
+// namespaces are not ours to describe.
+func typeMeaning(name string) string {
+	tail, ok := strings.CutPrefix(name, "type/")
+	if !ok {
+		return ""
+	}
+	for _, t := range issue.Types {
+		if t.Name == tail {
+			return t.Meaning
+		}
+	}
+	return ""
+}
+
+func hasAnyPrefix(s string, prefixes []string) bool {
+	for _, p := range prefixes {
+		if strings.HasPrefix(s, p) {
+			return true
+		}
+	}
+	return false
+}

cli/internal/mapping/labels_test.go

@@ -0,0 +1,95 @@
+package mapping
+
+import (
+	"regexp"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+var hexColor = regexp.MustCompile(`^#[0-9a-f]{6}$`)
+
+// The canonical set is the domain's, and every member of it must have a color
+// here. A type added over in the taxonomy that arrived as grey would look like
+// a label somebody created by hand.
+func TestEveryCanonicalLabelHasAColor(t *testing.T) {
+	for _, name := range issue.CanonicalLabels() {
+		color := LabelColor(name)
+		switch {
+		case color == DefaultColor:
+			t.Errorf("%s has no color of its own", name)
+		case !hexColor.MatchString(color):
+			t.Errorf("%s = %q, want #rrggbb in lower case", name, color)
+		}
+	}
+	// And the other direction: a color left behind after a label was retired
+	// paints nothing and is a lie about what the taxonomy holds.
+	if len(labelColors) != len(issue.CanonicalLabels()) {
+		t.Errorf("%d colors for %d canonical labels — one of the two lists moved without the other",
+			len(labelColors), len(issue.CanonicalLabels()))
+	}
+	// Anything outside the set is project-specific and nobody here can guess
+	// what it means.
+	if got := LabelColor("tech/sql"); got != DefaultColor {
+		t.Errorf("LabelColor(tech/sql) = %q, want the default", got)
+	}
+}
+
+func TestLabelSpecs(t *testing.T) {
+	cases := []struct {
+		name        string
+		description string
+		exclusive   bool
+	}{
+		{"type/bug", "Something behaves incorrectly in existing code", true},
+		{"type/draft", "Idea captured for later; not ready for work", true},
+		{"severity/critical", "", true},
+		// Exclusivity is a property of the namespace, not of the members the
+		// taxonomy happens to know.
+		{"type/spike", "", true},
+		{"tech/sql", "", false},
+		{"comp/appclick", "", false},
+	}
+	names := make([]string, len(cases))
+	for i, c := range cases {
+		names[i] = c.name
+	}
+
+	specs := LabelSpecs(names)
+	if len(specs) != len(cases) {
+		t.Fatalf("%d specs for %d names", len(specs), len(cases))
+	}
+	for i, c := range cases {
+		got := specs[i]
+		// The order is the taxonomy's: a bootstrap prints its plan in it, and
+		// two identical runs must not look like different ones.
+		if got.Name != c.name {
+			t.Fatalf("spec %d is %s, want %s", i, got.Name, c.name)
+		}
+		if got.Description != c.description {
+			t.Errorf("%s description = %q, want %q", c.name, got.Description, c.description)
+		}
+		if got.Exclusive != c.exclusive {
+			t.Errorf("%s exclusive = %v, want %v", c.name, got.Exclusive, c.exclusive)
+		}
+		if got.Color != LabelColor(c.name) {
+			t.Errorf("%s color = %q", c.name, got.Color)
+		}
+	}
+}
+
+func TestCanonicalLabelSpecsAreTheDomainsList(t *testing.T) {
+	specs := CanonicalLabelSpecs()
+	want := issue.CanonicalLabels()
+	if len(specs) != len(want) {
+		t.Fatalf("%d specs, want %d", len(specs), len(want))
+	}
+	for i, name := range want {
+		if specs[i].Name != name {
+			t.Errorf("spec %d = %s, want %s", i, specs[i].Name, name)
+		}
+		if !specs[i].Exclusive {
+			t.Errorf("%s must be exclusive — the canonical set IS the exclusive namespaces", name)
+		}
+	}
+}

cli/internal/mapping/layering_test.go

@@ -0,0 +1,46 @@
+package mapping
+
+import (
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+// The bridge translates values and nothing else: no network, no filesystem, no
+// clock, no configuration. Every one of those is a caller's to supply, which is
+// what lets this package be reasoned about and tested without a Gitea anywhere.
+//
+// Two imports and no more: internal/issue for what an issue is, and
+// internal/wire for the shapes on the other side. wire is allowed precisely
+// because it is inert — shapes and identifiers over the standard library, with
+// a layering test of its own — so naming a payload here costs nothing and
+// reaches nowhere.
+//
+// DIRECT imports, not the dependency walk internal/issue does. The domain
+// reaches os through internal/project and that is the domain's business; what
+// this test is about is what this package itself reaches for. A transport that
+// grew a helper here — or a lookup that quietly opened a config file — is what
+// it catches.
+func TestTheBridgeTranslatesAndNothingElse(t *testing.T) {
+	forbidden := map[string]string{
+		"net/http":  "an HTTP call belongs in the transport",
+		"net":       "an HTTP call belongs in the transport",
+		"os":        "a pure function reads no file and no environment",
+		"os/exec":   "nothing here shells out",
+		"io/ioutil": "a pure function reads no file",
+		"time":      "the clock is the caller's; a timestamp arrives as a string",
+		"git.noodles.cam/claude-skills/marketplace/cli/internal/gitea":   "the transport imports this package, never the reverse",
+		"git.noodles.cam/claude-skills/marketplace/cli/internal/config":  "credentials and repositories are the transport's",
+		"git.noodles.cam/claude-skills/marketplace/cli/internal/project": "nothing here resolves a path",
+	}
+
+	out, err := exec.Command("go", "list", "-f", `{{join .Imports "\n"}}`, ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	for _, dep := range strings.Fields(string(out)) {
+		if why, bad := forbidden[dep]; bad {
+			t.Errorf("mapping imports %s — %s", dep, why)
+		}
+	}
+}

cli/internal/mapping/mapping.go

@@ -0,0 +1,106 @@
+// Package mapping is md <-> Gitea JSON. The whole translation, and only the
+// translation.
+//
+// Pure functions: no network, no filesystem, no flags, no clock. Give it a
+// payload and it hands back a domain issue; give it an issue and it hands back
+// a request body. That purity is the point — it can be reasoned about and
+// tested without a Gitea anywhere, and it is the one package to open when the
+// two representations disagree.
+//
+// Direction of knowledge: this package imports the domain and the protocol
+// (internal/wire), and nothing imports it but the command layer. The domain
+// never imports it, and TestDomainDependsOnNothing over in internal/issue fails
+// the moment it does; the transport never imports it either, and
+// TestTransportDoesNotImportTheDomain over in internal/gitea says so. Both
+// sides speak wire's shapes, which is what lets the two meet without either one
+// reaching into the other.
+//
+// What crosses the boundary, and what does not:
+//
+//	domain          Gitea             note
+//	----------------------------------------------------------------------
+//	id (slug)       body marker       <!-- kettle:id … -->, first line of the
+//	                                  tracker-side body; stripped out of the
+//	                                  local copy — see marker.go
+//	title           title             verbatim, both ways
+//	body            body              verbatim up, verbatim down except the
+//	                                  marker and checkbox state
+//	state           state             open/closed, the same vocabulary
+//	labels          labels[]          names both ways; ids only on write
+//	assignees       assignees[]       logins
+//	milestone       milestone.title   resolved to an id on write
+//	depends         —                 slugs; #N is translated at this edge
+//	—               number, html_url  lands in Extra as gitea:/url:
+//	—               ref               Extra as branch:; push fills it from git
+//
+// `depends:` is the authoritative graph and is always slugs. The body's
+// `## Depends on` section is human prose and is passed through UNCHANGED in
+// both directions: a pull seeds `depends:` from the `#N` it finds there, and a
+// push never rewrites what the author wrote. Deliberate — a translator that
+// edits prose churns the body on every round trip.
+//
+// The ONE thing this package adds to a body is the id marker, and it does so
+// because the slug has to survive a push: push deletes the local file, so the
+// tracker has to be the thing that remembers what the issue was called here.
+package mapping
+
+import (
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// Origin is what this bridge writes into the domain's `origin:` field. The
+// domain records that an issue exists somewhere else; only this layer knows
+// where, and what the handle beside it means.
+const Origin = "gitea"
+
+// The sync-owned metadata fields, named once. Every one of them is bookkeeping
+// about a tracker, which is why the domain carries them verbatim in
+// Issue.Extra and never reads them — the format's ownership table draws the
+// same line. A field spelled in three call sites is a field that gets renamed
+// in two.
+const (
+	// GiteaKey is the handle in the tracker: owner/repo#42, a wire.Key written
+	// out. Cross-repo on purpose — a number alone is only unique inside one
+	// repository, and an issue that has been moved, or a store that has ever
+	// pointed at two repositories, needs the answer to say which.
+	GiteaKey = "gitea"
+	// URLKey is the issue's web address, for a receipt a human can click.
+	URLKey = "url"
+	// SyncedKey is when this copy was last written from or to the tracker —
+	// how old the working copy is, and nothing more.
+	SyncedKey = "synced"
+	// RemoteUpdatedKey is the tracker's own updated_at.
+	RemoteUpdatedKey = "remote-updated"
+	// CommentsKey is how many comments the tracker holds, so a reader knows a
+	// thread exists without fetching it.
+	CommentsKey = "comments"
+	// BranchKey is Gitea's `ref` — the branch an issue is pinned to. Its value
+	// is a git branch name and means exactly `ref`, which is what makes it a
+	// sync field rather than a domain one.
+	BranchKey = "branch"
+)
+
+// RemoteKeyOf is the handle an issue carries, and whether it carries one at
+// all.
+//
+// ok is false for anything that is not a handle: an empty field on a
+// never-pushed issue, a line somebody hand-edited, a key written by a format
+// that predates this one — and a bare `#42`, which names a number without the
+// repository that makes it mean something. Callers act on ok rather than on a
+// zero number, because "#0" and "not synced" would otherwise be the same
+// answer.
+func RemoteKeyOf(i *issue.Issue) (key wire.Key, ok bool) {
+	k, err := wire.ParseKey(i.Extra[GiteaKey])
+	if err != nil || k.Repo.Zero() {
+		return wire.Key{}, false
+	}
+	return k, true
+}
+
+// NumberOf is the Gitea number of an already-synced issue; ok is false for one
+// that has never been pushed.
+func NumberOf(i *issue.Issue) (number int, ok bool) {
+	k, ok := RemoteKeyOf(i)
+	return k.Number, ok
+}

cli/internal/mapping/mapping_test.go

@@ -0,0 +1,381 @@
+package mapping
+
+import (
+	"encoding/json"
+	"reflect"
+	"strings"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// The repository the fixtures are pushed to. A wire.Repo and not a string: the
+// handle in `gitea:` is a key, and a key is a repository and a number.
+var tea = wire.Repo{Owner: "claude-skills", Name: "tea"}
+
+// A file exactly as the store holds it: domain fields, then the sync fields the
+// domain carries and never reads.
+const stored = `---
+id: wire-sqlc-appclick
+state: open
+labels: [type/task, tech/sql]
+assignees: [naudachu]
+milestone: v0.2
+depends: [migrate-schema]
+origin: gitea
+branch: feat/wire-sqlc
+gitea: claude-skills/tea#42
+synced: 2026-08-09T18:40:00Z
+---
+# Wire sqlc into the appclick repo layer
+
+## Summary
+Проводка sqlc в слой репозиториев.
+
+## Spec
+none
+
+## Depends on
+- #7 — нужна схема БД из этого issue
+
+## Acceptance criteria
+- [x] сгенерирован код
+- [ ] тесты зелёные
+`
+
+func ptr[T any](v T) *T { return &v }
+
+func roundTripOptions() RequestOptions {
+	return RequestOptions{
+		LabelIDs:     map[string]int64{"type/task": 11, "tech/sql": 12},
+		MilestoneID:  ptr(int64(5)),
+		IncludeState: true,
+	}
+}
+
+// The whole point of the package in one test: everything the format says is
+// preserved comes back, and the body comes back byte for byte.
+func TestRoundTripPreservesEveryFieldTheFormatKeeps(t *testing.T) {
+	local := issue.FromText(stored, "wire-sqlc-appclick")
+	req := ToRequest(local, roundTripOptions())
+
+	if req.Title == nil || *req.Title != local.Title {
+		t.Errorf("title = %v, want %q", req.Title, local.Title)
+	}
+	if req.Body == nil {
+		t.Fatal("the request carries no body — a create would file an empty issue")
+	}
+	if got := IDInBody(*req.Body); got != local.ID {
+		t.Errorf("the request body does not claim the slug: %q", got)
+	}
+	if got := StripIDMarker(*req.Body); got != strings.TrimSpace(local.Body) {
+		t.Errorf("the prose was rewritten on the way up:\n--- got ---\n%s\n--- want ---\n%s",
+			got, strings.TrimSpace(local.Body))
+	}
+	if want := []int64{11, 12}; req.Labels == nil || !reflect.DeepEqual(*req.Labels, want) {
+		t.Errorf("labels = %v, want %v", req.Labels, want)
+	}
+	if want := []string{"naudachu"}; req.Assignees == nil || !reflect.DeepEqual(*req.Assignees, want) {
+		t.Errorf("assignees = %v, want %v", req.Assignees, want)
+	}
+	if req.Milestone == nil || *req.Milestone != 5 {
+		t.Errorf("milestone = %v, want 5", req.Milestone)
+	}
+	if req.State == nil || *req.State != "open" {
+		t.Errorf("state = %v", req.State)
+	}
+	if req.Ref == nil || *req.Ref != "feat/wire-sqlc" {
+		t.Errorf("ref = %v — branch: is a sync field and must ride along", req.Ref)
+	}
+
+	// What the tracker hands back is the body it was given, plus its own
+	// bookkeeping.
+	echo := &wire.Issue{
+		Number:    42,
+		Title:     *req.Title,
+		Body:      *req.Body,
+		State:     "open",
+		HTMLURL:   "https://git.noodles.cam/claude-skills/tea/issues/42",
+		UpdatedAt: "2026-08-09T18:24:01Z",
+		Ref:       *req.Ref,
+		Comments:  3,
+		Labels:    []wire.Label{{Name: "type/task"}, {Name: "tech/sql"}},
+		Assignees: []wire.User{{Login: "naudachu"}},
+		Milestone: &wire.Milestone{ID: 5, Title: "v0.2"},
+	}
+	back, unresolved := FromPayload(echo, local.ID, tea, PayloadOptions{
+		IDForNumber: map[int]string{7: "migrate-schema"},
+		Synced:      "2026-08-09T18:40:00Z",
+	})
+	if len(unresolved) != 0 {
+		t.Errorf("unresolved = %v, want none", unresolved)
+	}
+
+	if back.Body != strings.TrimSpace(local.Body) {
+		t.Errorf("the body did not survive the trip:\n--- got ---\n%s\n--- want ---\n%s",
+			back.Body, strings.TrimSpace(local.Body))
+	}
+	if strings.Contains(back.Body, "kettle:id") || strings.Contains(back.Body, "tea:id") {
+		t.Error("the marker reached the local copy — it is transport bookkeeping and belongs nowhere near disk")
+	}
+	for _, c := range []struct{ name, got, want string }{
+		{"id", back.ID, local.ID},
+		{"title", back.Title, local.Title},
+		{"state", back.State, local.State},
+		{"milestone", back.Milestone, local.Milestone},
+		{"origin", back.Origin, local.Origin},
+		{"gitea", back.Extra[GiteaKey], "claude-skills/tea#42"},
+		{"branch", back.Extra[BranchKey], "feat/wire-sqlc"},
+		{"synced", back.Extra[SyncedKey], "2026-08-09T18:40:00Z"},
+		{"url", back.Extra[URLKey], "https://git.noodles.cam/claude-skills/tea/issues/42"},
+		{"remote-updated", back.Extra[RemoteUpdatedKey], "2026-08-09T18:24:01Z"},
+		{"comments", back.Extra[CommentsKey], "3"},
+	} {
+		if c.got != c.want {
+			t.Errorf("%s = %q, want %q", c.name, c.got, c.want)
+		}
+	}
+	if !reflect.DeepEqual(back.Labels, local.Labels) {
+		t.Errorf("labels = %v, want %v", back.Labels, local.Labels)
+	}
+	if !reflect.DeepEqual(back.Assignees, local.Assignees) {
+		t.Errorf("assignees = %v, want %v", back.Assignees, local.Assignees)
+	}
+	// `depends:` is slugs; the `#7` the prose names is translated at this edge
+	// and the prose itself is left alone.
+	if !reflect.DeepEqual(back.Depends, local.Depends) {
+		t.Errorf("depends = %v, want %v", back.Depends, local.Depends)
+	}
+	if !strings.Contains(back.Body, "- #7 — нужна схема БД из этого issue") {
+		t.Error("the ## Depends on prose was rewritten; it is the author's text and passes through unchanged")
+	}
+
+	// And the strongest form of "no churn": pushing what came back sends
+	// exactly what was sent the first time.
+	if again := ToRequest(back, roundTripOptions()); !reflect.DeepEqual(again, req) {
+		t.Errorf("a second push differs from the first:\n--- again ---\n%+v\n--- first ---\n%+v", again, req)
+	}
+}
+
+// The other shape an issue comes in: nothing scheduled, nobody assigned.
+func TestNoMilestoneAndNoAssignees(t *testing.T) {
+	local := issue.FromText("---\nid: lone\nstate: open\nlabels: [type/task]\n"+
+		"assignees: []\nmilestone: none\ndepends: []\norigin: local\n---\n"+
+		"# A lone issue\n\n## Summary\nОдин.\n", "lone")
+
+	req := ToRequest(local, RequestOptions{LabelIDs: map[string]int64{"type/task": 11}})
+	if req.Assignees != nil {
+		t.Errorf("assignees = %v — an empty list would clear whoever the tracker has", req.Assignees)
+	}
+	if req.Milestone != nil {
+		t.Errorf("milestone = %v — a missing milestone is no opinion, not a detach", req.Milestone)
+	}
+
+	raw, err := json.Marshal(req)
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	body := string(raw)
+	for _, key := range []string{`"assignees"`, `"milestone"`, `"state"`, `"ref"`} {
+		if strings.Contains(body, key) {
+			t.Errorf("%s is in the request body; on a PATCH that overwrites what the tracker holds: %s", key, body)
+		}
+	}
+	// A resolved-but-empty label set is the opposite statement and must be sent.
+	if !strings.Contains(body, `"labels":[11]`) {
+		t.Errorf("labels missing from %s", body)
+	}
+	empty, err := json.Marshal(ToRequest(local, RequestOptions{LabelIDs: map[string]int64{}}))
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	if !strings.Contains(string(empty), `"labels":[]`) {
+		t.Errorf("a resolved label set that matched nothing must still be sent as []: %s", empty)
+	}
+	silent, err := json.Marshal(ToRequest(local, RequestOptions{}))
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	if strings.Contains(string(silent), `"labels"`) {
+		t.Errorf("a caller that resolved no ids must not clear the tracker's labels: %s", silent)
+	}
+
+	back, unresolved := FromPayload(&wire.Issue{
+		Number:  9,
+		Title:   "A lone issue",
+		Body:    WithIDMarker("## Summary\nОдин.", "lone"),
+		State:   "open",
+		HTMLURL: "https://git.noodles.cam/claude-skills/tea/issues/9",
+	}, "lone", tea, PayloadOptions{Synced: "2026-08-11T10:00:00Z"})
+
+	if len(unresolved) != 0 {
+		t.Errorf("unresolved = %v", unresolved)
+	}
+	if back.Milestone != "" || back.Assignees != nil || back.Labels != nil {
+		t.Errorf("empty came back as something: milestone=%q assignees=%v labels=%v",
+			back.Milestone, back.Assignees, back.Labels)
+	}
+	if _, ok := back.Extra[BranchKey]; ok {
+		t.Error("an absent ref must not write an empty branch: field")
+	}
+	if _, ok := back.Extra[CommentsKey]; ok {
+		t.Error("zero comments is not a fact worth a line in the file")
+	}
+	if !strings.Contains(back.Text(), "milestone: none") {
+		t.Error("an empty milestone must render back as none")
+	}
+}
+
+// A dependency whose target is not in the store yet is reported, never invented:
+// a made-up slug is an edge to a file that does not exist.
+func TestUnresolvedNumbersAreReportedNotInvented(t *testing.T) {
+	body := "## Summary\nx\n\n## Depends on\n- #7\n- #8\n"
+	back, unresolved := FromPayload(&wire.Issue{Number: 1, Title: "T", Body: body},
+		"here", wire.Repo{Owner: "o", Name: "r"}, PayloadOptions{IDForNumber: map[int]string{7: "known"}})
+
+	if want := []string{"known"}; !reflect.DeepEqual(back.Depends, want) {
+		t.Errorf("depends = %v, want %v", back.Depends, want)
+	}
+	if want := []int{8}; !reflect.DeepEqual(unresolved, want) {
+		t.Errorf("unresolved = %v, want %v", unresolved, want)
+	}
+	if !strings.Contains(back.Body, "- #8") {
+		t.Error("the body still names it, which is why dropping it from depends: loses nothing")
+	}
+}
+
+func TestExtraNumbersJoinTheOnesTheBodyNames(t *testing.T) {
+	body := "## Summary\nx\n\n## Depends on\n- #7\n"
+	back, _ := FromPayload(&wire.Issue{Number: 1, Title: "T", Body: body}, "here", wire.Repo{Owner: "o", Name: "r"},
+		PayloadOptions{
+			IDForNumber:  map[int]string{7: "seven", 9: "nine"},
+			ExtraNumbers: []int{7, 9},
+		})
+	if want := []string{"seven", "nine"}; !reflect.DeepEqual(back.Depends, want) {
+		t.Errorf("depends = %v, want %v", back.Depends, want)
+	}
+}
+
+// The one exception to "a pull overwrites the body", and the narrowest one that
+// works: a tick only ever travels one way, so the two sides are a set union.
+func TestMergeCheckboxState(t *testing.T) {
+	cases := []struct {
+		name             string
+		remote, local    string
+		want             string
+		wantUnchangedRef bool
+	}{
+		{
+			name:   "a local tick survives the overwrite",
+			remote: "- [ ] один\n- [ ] два\n",
+			local:  "- [x] два\n",
+			want:   "- [ ] один\n- [x] два\n",
+		},
+		{
+			name:   "rewrapping an item does not cost it its tick",
+			remote: "- [ ] очень длинный\n  пункт\n",
+			local:  "- [x] очень длинный пункт\n",
+			want:   "- [x] очень длинный\n  пункт\n",
+		},
+		{
+			name:   "the same text twice is the same item to whoever reads it",
+			remote: "- [ ] дубль\n- [ ] дубль\n",
+			local:  "- [ ] дубль\n- [x] дубль\n",
+			want:   "- [x] дубль\n- [x] дубль\n",
+		},
+		{
+			name:   "a first pull has nothing to merge",
+			remote: "- [ ] один\n",
+			local:  "",
+			want:   "- [ ] один\n",
+		},
+		{
+			name:   "unticking is not monotone, so it does not travel",
+			remote: "- [x] один\n",
+			local:  "- [ ] один\n",
+			want:   "- [x] один\n",
+		},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			if got := MergeCheckboxState(c.remote, c.local); got != c.want {
+				t.Errorf("got:\n%q\nwant:\n%q", got, c.want)
+			}
+		})
+	}
+}
+
+// What `gitea:` holds is a key, and it round-trips through the one parser.
+// Anything that is not a key reads as "not synced" — never as issue #0, and
+// never as the issue -3 a bare strconv.Atoi would have handed back.
+func TestRemoteKeyRoundTrip(t *testing.T) {
+	cases := []struct {
+		key    string
+		repo   string
+		number int
+		ok     bool
+	}{
+		{"claude-skills/tea#42", "claude-skills/tea", 42, true},
+		{"o/r#1", "o/r", 1, true},
+		// Never pushed, hand-edited, or written by a format that predates this
+		// one — all the same answer, and none of them is issue #0. `#42` is in
+		// the list because a handle without a repository addresses nothing.
+		{"", "", 0, false},
+		{"claude-skills/tea", "", 0, false},
+		{"#42", "", 0, false},
+		{"o/r#", "", 0, false},
+		{"o/r#-3", "", 0, false},
+		{"o/r#4x", "", 0, false},
+	}
+	for _, c := range cases {
+		got, ok := RemoteKeyOf(&issue.Issue{Extra: map[string]string{GiteaKey: c.key}})
+		if got.Repo.String() != c.repo || got.Number != c.number || ok != c.ok {
+			t.Errorf("RemoteKeyOf(%q) = (%v, %v), want (%q, %d, %v)",
+				c.key, got, ok, c.repo, c.number, c.ok)
+		}
+		if c.ok && got.String() != c.key {
+			t.Errorf("the key formatted back as %q, want %q", got, c.key)
+		}
+	}
+}
+
+func TestNumberOf(t *testing.T) {
+	synced := &issue.Issue{Extra: map[string]string{GiteaKey: "o/r#42"}}
+	if n, ok := NumberOf(synced); n != 42 || !ok {
+		t.Errorf("NumberOf = (%d, %v), want (42, true)", n, ok)
+	}
+	if _, ok := NumberOf(&issue.Issue{}); ok {
+		t.Error("an issue that has never been pushed has no number")
+	}
+}
+
+func TestApplyRemoteStampsTheSyncFields(t *testing.T) {
+	local := &issue.Issue{ID: "x", Origin: issue.Local}
+	ApplyRemote(local, &wire.Issue{
+		Number:    42,
+		HTMLURL:   "https://git.noodles.cam/o/r/issues/42",
+		UpdatedAt: "2026-08-09T18:24:01Z",
+	}, wire.Repo{Owner: "o", Name: "r"}, "2026-08-11T10:00:00Z")
+
+	if local.IsLocal() {
+		t.Error("origin must move: the work exists somewhere else now")
+	}
+	if local.Extra[GiteaKey] != "o/r#42" || local.Extra[URLKey] == "" ||
+		local.Extra[SyncedKey] != "2026-08-11T10:00:00Z" ||
+		local.Extra[RemoteUpdatedKey] != "2026-08-09T18:24:01Z" {
+		t.Errorf("extra = %v", local.Extra)
+	}
+}
+
+func TestRenderComments(t *testing.T) {
+	got := RenderComments([]wire.Comment{
+		{ID: 1, User: wire.User{Login: "naudachu"}, CreatedAt: "2026-08-09T18:24:01Z", Body: " привет "},
+		{ID: 2, User: wire.User{Login: "bot"}, CreatedAt: "", Body: ""},
+	})
+	want := "## comment 1 — naudachu — 2026-08-09\n\nпривет\n\n" +
+		"## comment 2 — bot — \n\n(empty)\n"
+	if got != want {
+		t.Errorf("got:\n%q\nwant:\n%q", got, want)
+	}
+}

cli/internal/mapping/marker.go

@@ -0,0 +1,110 @@
+package mapping
+
+import (
+	"regexp"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+)
+
+// The id marker: the slug, kept tracker-side.
+//
+// Push deletes the local file once the tracker has confirmed the write, so the
+// slug — the issue's ONLY identity in the domain — cannot live only on this
+// machine any more. It rides up in the body as an HTML comment:
+//
+//	<!-- kettle:id wire-sqlc-appclick -->
+//
+// Why the body and not a local number -> slug ledger: the ledger is a local
+// file, and "the local copy is not the record" is the whole point of deleting
+// one. A marker in the body survives a rename in the web UI, a lost ledger, a
+// fresh clone, and a second machine — none of which the ledger does. Why an
+// HTML comment: Gitea renders markdown, so it is invisible to a human reader,
+// and it comes back verbatim on every API read.
+//
+// WHERE: the first line of the tracker-side body, followed by one blank line.
+// First because it is the one position that does not depend on what sections
+// the issue happens to have, and because a human who does look at the raw
+// markdown finds it before the prose rather than buried in it.
+//
+// WHAT THE LOCAL FILE SEES: nothing. FromPayload strips every marker before the
+// body reaches the store, so `.kettle/issues/<id>.md` holds exactly what the
+// author wrote — checkbox line numbers, `kettle check`, and diffs are all
+// unaffected, and the slug is already the file's name, so a copy of it in the
+// body would be duplicated state.
+//
+// WHY IT CANNOT ACCUMULATE: the two operations are strip-all and
+// strip-all-then-prepend-one. WithIDMarker never appends to what is there, and
+// StripIDMarker removes EVERY marker line, not the first. So a body that
+// somehow gained two (a hand-edit in the web UI, a copy-paste) is cleaned on
+// the next pull and goes back up with exactly one. There is no code path that
+// adds a marker to a body that has not just been stripped.
+//
+// WHY TWO SPELLINGS ARE READ AND ONE IS WRITTEN: this tool was called `tea`
+// and wrote `<!-- tea:id … -->`. Issues pushed under that name are sitting in
+// the tracker right now, and their local files are gone — the marker is the
+// only copy of their slug there is. A rename that stopped reading the old
+// spelling would orphan every one of them: the pull would fall back to the
+// title, allocate a fresh slug, and every `depends:` pointing at the old one
+// would dangle. So the writer moved and the reader did not.
+var markerRe = regexp.MustCompile(
+	`^[ \t]*<!--[ \t]*(?:kettle|tea):id[ \t]+(\S+)[ \t]*-->[ \t]*$`)
+
+// IDMarker is the marker line for a slug. One place formats it, one regex
+// reads it — and what that regex accepts is deliberately wider than this.
+func IDMarker(id string) string { return "<!-- kettle:id " + id + " -->" }
+
+// IDInBody is the slug a tracker-side body claims, or "" when it claims none.
+//
+// The FIRST valid marker wins; a second one is ignored here and removed by
+// StripIDMarker on the way in. The captured text must be a slug by the domain's
+// own rule — a marker holding anything else is not a slug and is treated as if
+// it were not there, so a mangled comment falls back to the title instead of
+// naming a file after garbage.
+func IDInBody(body string) string {
+	for _, line := range strings.Split(body, "\n") {
+		if m := markerRe.FindStringSubmatch(strings.TrimSuffix(line, "\r")); m != nil {
+			if issue.IsSlug(m[1]) {
+				return m[1]
+			}
+		}
+	}
+	return ""
+}
+
+// StripIDMarker is body with every marker line removed, in either spelling.
+// Idempotent.
+//
+// A body that carries no marker is returned byte for byte — the common case (an
+// issue filed in the web UI) costs nothing and is not reformatted. When a marker
+// is removed from the top, the blank line it was written with goes with it, so
+// the round trip is exact: StripIDMarker(WithIDMarker(b, id)) == b.
+func StripIDMarker(body string) string {
+	lines := strings.Split(body, "\n")
+	found := false
+	for _, line := range lines {
+		if markerRe.MatchString(strings.TrimSuffix(line, "\r")) {
+			found = true
+			break
+		}
+	}
+	if !found {
+		return body
+	}
+	kept := make([]string, 0, len(lines))
+	for _, line := range lines {
+		if !markerRe.MatchString(strings.TrimSuffix(line, "\r")) {
+			kept = append(kept, line)
+		}
+	}
+	return strings.TrimLeft(strings.Join(kept, "\n"), "\n")
+}
+
+// WithIDMarker is body with exactly one marker, as its first line.
+//
+// Strip-then-prepend, always — that is the guarantee that a body can never end
+// up with two, however many it arrived with, and it is what quietly rewrites a
+// `tea:id` marker into the current spelling the next time the issue is pushed.
+func WithIDMarker(body, id string) string {
+	return IDMarker(id) + "\n\n" + StripIDMarker(body)
+}

cli/internal/mapping/marker_test.go

@@ -0,0 +1,104 @@
+package mapping
+
+import (
+	"strings"
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func TestIDInBodyReadsBothSpellings(t *testing.T) {
+	cases := []struct {
+		name string
+		body string
+		want string
+	}{
+		{"the spelling this tool writes", "<!-- kettle:id wire-sqlc -->\n\n## Summary\nx", "wire-sqlc"},
+		// The whole reason the reader is wider than the writer.
+		{"the spelling already in the tracker", "<!-- tea:id wire-sqlc -->\n\n## Summary\nx", "wire-sqlc"},
+		{"indented and loosely spaced", "  <!--   tea:id   wire-sqlc   -->  \n", "wire-sqlc"},
+		{"no marker at all", "## Summary\nx", ""},
+		// A mangled comment falls back to the title rather than naming a file
+		// after garbage.
+		{"not a slug", "<!-- kettle:id Wire_SQLC -->\n", ""},
+		{"not on a line of its own", "text <!-- kettle:id wire-sqlc -->\n", ""},
+		{"the first valid marker wins", "<!-- tea:id first-one -->\n<!-- kettle:id second-one -->\n", "first-one"},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			if got := IDInBody(c.body); got != c.want {
+				t.Errorf("IDInBody = %q, want %q", got, c.want)
+			}
+		})
+	}
+}
+
+// An issue pushed under the old name is sitting in the tracker with its local
+// file long since deleted — the marker is the only copy of its slug there is.
+// It has to keep resolving, and it has to come back up in the new spelling.
+func TestAnIssuePushedByTheOldNameStillResolves(t *testing.T) {
+	const inTracker = "<!-- tea:id wire-sqlc-appclick -->\n\n## Summary\nПроводка sqlc.\n"
+
+	id := IDInBody(inTracker)
+	if id != "wire-sqlc-appclick" {
+		t.Fatalf("IDInBody = %q — every issue pushed under the old name would be orphaned", id)
+	}
+
+	iss, _ := FromPayload(&wire.Issue{Number: 42, Title: "Wire sqlc", Body: inTracker},
+		id, tea, PayloadOptions{})
+	if strings.Contains(iss.Body, "tea:id") {
+		t.Errorf("the old marker reached the local copy: %q", iss.Body)
+	}
+	if iss.Body != "## Summary\nПроводка sqlc." {
+		t.Errorf("body = %q", iss.Body)
+	}
+
+	// And the next push rewrites it into the current spelling, without ever
+	// having two.
+	up := *ToRequest(iss, RequestOptions{}).Body
+	if !strings.HasPrefix(up, "<!-- kettle:id wire-sqlc-appclick -->\n\n") {
+		t.Errorf("the marker was not rewritten: %q", up)
+	}
+	if strings.Contains(up, "tea:id") {
+		t.Errorf("both spellings went up: %q", up)
+	}
+	if n := strings.Count(up, ":id "); n != 1 {
+		t.Errorf("%d markers in the body, want 1", n)
+	}
+}
+
+func TestMarkersCannotAccumulate(t *testing.T) {
+	body := "## Summary\nx"
+	// Whatever it arrived with — one, the other, several — it goes up with one.
+	messy := "<!-- tea:id old-one -->\n\n<!-- kettle:id other-one -->\n\n" + body
+	got := WithIDMarker(messy, "real-one")
+
+	if want := IDMarker("real-one") + "\n\n" + body; got != want {
+		t.Errorf("got:\n%q\nwant:\n%q", got, want)
+	}
+	if got := WithIDMarker(got, "real-one"); strings.Count(got, "<!--") != 1 {
+		t.Errorf("a second pass added one: %q", got)
+	}
+}
+
+func TestStripIsTheExactInverseOfWith(t *testing.T) {
+	bodies := []string{
+		"## Summary\nx",
+		"## Summary\nx\n\n## Acceptance criteria\n- [ ] один\n",
+		"",
+	}
+	for _, b := range bodies {
+		if got := StripIDMarker(WithIDMarker(b, "an-id")); got != b {
+			t.Errorf("StripIDMarker(WithIDMarker(%q)) = %q", b, got)
+		}
+	}
+}
+
+// The common case — an issue filed in the web UI — costs nothing and is not
+// reformatted.
+func TestStripLeavesAnUnmarkedBodyByteForByte(t *testing.T) {
+	body := "\n\n## Summary\nx\n\n\n"
+	if got := StripIDMarker(body); got != body {
+		t.Errorf("StripIDMarker rewrote a body with no marker in it: %q", got)
+	}
+}

cli/internal/mapping/torequest.go

@@ -0,0 +1,97 @@
+package mapping
+
+import (
+	"slices"
+	"strings"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/issue"
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+// domain -> Gitea.
+
+// RequestOptions are what the transport resolved before the call: label names
+// are ids by then, and a milestone title is a number.
+//
+// Both are lookups against one repository, which is why they cannot be done
+// here — this package never learns which repository it is translating for
+// beyond the name it is handed.
+type RequestOptions struct {
+	// LabelIDs is name -> id for the labels this repository holds. A nil map
+	// leaves `labels` out of the request; a non-nil one sends the list, empty
+	// included, and a label the repository does not have is silently left off
+	// rather than failing the write — an unknown label is a bootstrap that has
+	// not run, not a reason to lose the issue.
+	LabelIDs map[string]int64
+	// MilestoneID is the resolved milestone. nil leaves the key out, which on
+	// an edit means "leave whatever is attached alone".
+	MilestoneID *int64
+	// IncludeState sends `state`. An edit that means to open or close says so;
+	// a create takes the tracker's default.
+	IncludeState bool
+}
+
+// ToRequest is the request body for creating or editing an issue.
+//
+// The prose is sent verbatim — see the package doc on why slugs in
+// `## Depends on` are not rewritten to `#N`. The one addition is the id marker,
+// prepended (never appended) so the tracker remembers the slug after push has
+// deleted the local file. FromPayload takes it straight back off, so the body
+// still round-trips byte for byte.
+//
+// A create needs a title and a body, so those two are always filled. Every
+// other key is left out unless the caller has an opinion about it: on a PATCH
+// an absent key leaves the tracker's value alone, and a present one overwrites
+// it — see wire.IssueRequest for what each of them clears when it is sent
+// empty.
+func ToRequest(i *issue.Issue, opt RequestOptions) *wire.IssueRequest {
+	r := &wire.IssueRequest{
+		Title: wire.Set(i.Title),
+		Body:  wire.Set(WithIDMarker(strings.TrimSpace(i.Body), i.ID)),
+	}
+	if opt.LabelIDs != nil {
+		ids := []int64{}
+		for _, name := range i.Labels {
+			if id, ok := opt.LabelIDs[name]; ok {
+				ids = append(ids, id)
+			}
+		}
+		r.Labels = &ids
+	}
+	// Copied, so the request body and the issue it came from cannot alias one
+	// slice: whatever a caller does to either afterwards is not a change to
+	// what was sent.
+	if len(i.Assignees) > 0 {
+		r.Assignees = wire.Set(slices.Clone(i.Assignees))
+	}
+	if opt.MilestoneID != nil {
+		r.Milestone = opt.MilestoneID
+	}
+	if opt.IncludeState {
+		r.State = wire.Set(i.State)
+	}
+	// An empty `branch:` is "no opinion", not "no branch": sending ref="" would
+	// clear whatever is set on the Gitea side, so the key is left out instead.
+	if branch := strings.TrimSpace(i.Extra[BranchKey]); branch != "" {
+		r.Ref = wire.Set(branch)
+	}
+	return r
+}
+
+// ApplyRemote stamps the sync-owned fields onto an issue after a successful
+// write. Mutates and returns it; `origin` is the one domain field this touches,
+// and it touches it because "this work exists somewhere else now" is exactly
+// what has just become true.
+func ApplyRemote(i *issue.Issue, p *wire.Issue, repo wire.Repo, synced string) *issue.Issue {
+	if i.Extra == nil {
+		i.Extra = map[string]string{}
+	}
+	i.Origin = Origin
+	i.Extra[GiteaKey] = wire.Key{Repo: repo, Number: p.Number}.String()
+	i.Extra[URLKey] = p.HTMLURL
+	i.Extra[SyncedKey] = synced
+	if p.UpdatedAt != "" {
+		i.Extra[RemoteUpdatedKey] = p.UpdatedAt
+	}
+	return i
+}

cli/internal/project/init.go

@@ -0,0 +1,206 @@
+package project
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+// Initializing is a statement, and the only one that matters here: *this*
+// directory is the project whose issues live in it. It is answered once, by a
+// person, and everything downstream reads the answer instead of guessing.
+//
+// The marker is deliberately something an operator makes, not something
+// inferred from the tree: `.git` is in every clone including this repository's
+// own, so a plugin that inferred its root from one wrote issues into itself.
+
+// Layouts this has been through, migrated in on init in the order listed —
+// oldest first, so a tree that skipped a generation still lands in one place.
+//
+// Each is a move, never a copy: two stores is the state the marker exists to
+// prevent, and a store left behind at an old path is a store somebody will edit
+// by accident months later.
+var legacy = map[string][]string{
+	"issues": {
+		filepath.Join("tmp", "issues"),
+		filepath.Join(".tea", "issues"),
+	},
+	"payload": {
+		filepath.Join("tmp", "payload"),
+		filepath.Join(".tea", "payload"),
+	},
+}
+
+// ClashError reports that a migration found the same name on both sides.
+//
+// Two versions of one issue, and which one survives is not a decision a
+// migration gets to make quietly.
+type ClashError struct {
+	Src, Dst string
+	Names    []string
+}
+
+func (e *ClashError) Error() string {
+	names := e.Names
+	suffix := ""
+	if len(names) > 5 {
+		suffix = fmt.Sprintf(" (+%d more)", len(names)-5)
+		names = names[:5]
+	}
+	return fmt.Sprintf("%s and %s both hold %s%s — move or delete one side first; nothing was changed",
+		e.Src, e.Dst, strings.Join(names, ", "), suffix)
+}
+
+// Init makes root a project. Everything it does is idempotent:
+//
+//   - creates .tea/issues/ and .tea/payload/
+//   - moves an existing tmp/issues/ and tmp/payload/ in, if it finds them
+//   - adds .tea/ to .gitignore
+//
+// The move is the migration off the old layout and it is a move, not a copy:
+// two stores is the state the marker exists to prevent, and a store left behind
+// at the old path is a store somebody will edit by accident.
+//
+// .tea/ is gitignored because an `origin: local` issue is the only copy of that
+// work and the operator, not this command, decides what goes in a shared
+// history. Committing the store is a legitimate choice — drop the line if you
+// make it.
+//
+// Returns one line per thing done, for the receipt.
+func Init(root string, dryRun bool) ([]string, error) {
+	var done []string
+	marker := filepath.Join(root, Marker)
+	fresh := !isDir(marker)
+
+	for _, name := range []string{"issues", "payload"} {
+		d := filepath.Join(marker, name)
+		if isDir(d) {
+			continue
+		}
+		if !dryRun {
+			if err := os.MkdirAll(d, 0o755); err != nil {
+				return done, err
+			}
+		}
+		done = append(done, "created "+filepath.Join(Marker, name))
+	}
+
+	for _, name := range []string{"issues", "payload"} {
+		for _, old := range legacy[name] {
+			src := filepath.Join(root, old)
+			moved, err := migrate(src, filepath.Join(marker, name), dryRun)
+			if err != nil {
+				return done, err
+			}
+			switch {
+			case moved == nil:
+				// nothing there to migrate
+			case len(moved) == 0:
+				done = append(done, old+" was empty — nothing to move")
+			default:
+				done = append(done, fmt.Sprintf("moved %d file(s) from %s to %s",
+					len(moved), old, filepath.Join(Marker, name)))
+			}
+		}
+	}
+	// The old marker goes only when the migration emptied it — anything else
+	// parked in there is somebody's, and this is not the command that decides
+	// what.
+	if !dryRun {
+		os.Remove(filepath.Join(root, ".tea"))
+	}
+
+	added, err := addToGitignore(filepath.Join(root, ".gitignore"), Marker+"/", dryRun)
+	if err != nil {
+		return done, err
+	}
+	if added {
+		done = append(done, "added "+Marker+"/ to .gitignore")
+	}
+
+	switch {
+	case len(done) == 0:
+		done = append(done, "already initialized — nothing to do")
+	case fresh:
+		done = append(done, fmt.Sprintf("%s now tracks issues in %s/issues", root, Marker))
+	}
+	return done, nil
+}
+
+// migrate moves the CONTENTS of src into dst — contents, not the directory, so
+// an already-created destination is not a reason to refuse. Returns the names
+// moved, or nil when there was nothing to migrate.
+func migrate(src, dst string, dryRun bool) ([]string, error) {
+	if !isDir(src) {
+		return nil, nil
+	}
+	entries, err := os.ReadDir(src)
+	if err != nil {
+		return nil, err
+	}
+	names := make([]string, 0, len(entries))
+	for _, e := range entries {
+		names = append(names, e.Name())
+	}
+	sort.Strings(names)
+	if len(names) == 0 {
+		return []string{}, nil
+	}
+
+	var clashes []string
+	for _, n := range names {
+		if _, err := os.Lstat(filepath.Join(dst, n)); err == nil {
+			clashes = append(clashes, n)
+		}
+	}
+	if len(clashes) > 0 {
+		return nil, &ClashError{Src: src, Dst: dst, Names: clashes}
+	}
+	if dryRun {
+		return names, nil
+	}
+	if err := os.MkdirAll(dst, 0o755); err != nil {
+		return nil, err
+	}
+	for _, n := range names {
+		if err := os.Rename(filepath.Join(src, n), filepath.Join(dst, n)); err != nil {
+			return nil, err
+		}
+	}
+	os.Remove(src) // only succeeds when we emptied it, which is the intent
+	return names, nil
+}
+
+// addToGitignore appends entry unless some line already ignores it.
+func addToGitignore(path, entry string, dryRun bool) (bool, error) {
+	var lines []string
+	if raw, err := os.ReadFile(path); err == nil {
+		lines = strings.Split(strings.TrimSuffix(string(raw), "\n"), "\n")
+	} else if !os.IsNotExist(err) {
+		return false, err
+	}
+	want := strings.TrimSuffix(entry, "/")
+	for _, line := range lines {
+		if strings.TrimSuffix(strings.TrimSpace(line), "/") == want {
+			return false, nil
+		}
+	}
+	if dryRun {
+		return true, nil
+	}
+	trailer := "\n"
+	if len(lines) == 0 || lines[len(lines)-1] == "" {
+		trailer = ""
+	}
+	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0o644)
+	if err != nil {
+		return false, err
+	}
+	defer f.Close()
+	if _, err := f.WriteString(trailer + entry + "\n"); err != nil {
+		return false, err
+	}
+	return true, nil
+}

cli/internal/project/project.go

@@ -0,0 +1,245 @@
+// Package project answers one question: which directory is the project.
+//
+// Everything that is a fact about the project — the issue store, the request
+// payload scratchpad, the tracker the issues belong to — is resolved from the
+// answer, and the answer is found by one walk written once. The guard, the
+// transport and the store used to each have their own copy of that walk in
+// Python, and they disagreed: in a linked worktree `tea` worked while every
+// script reported no login pinned.
+//
+// This package depends on nothing but the standard library, and nothing in it
+// resolves from the binary's own location. Where an installation keeps its
+// files is a fact about the installation; which issues a tree has is a fact
+// about the tree, and a binary installed in one place and pointed at another
+// must answer from the one it was pointed at.
+package project
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// Marker is the directory an operator creates to state "this is a project".
+// It is never inferred. `.git` was tried and is in every clone, including this
+// repository's own, so a plugin resolved its store inside itself.
+const Marker = ".kettle"
+
+// Everything under the marker, each resolved by the same walk so that which
+// command wrote a file cannot change where it landed.
+var (
+	storeParts   = []string{Marker, "issues"}
+	payloadParts = []string{Marker, "payload"}
+	configParts  = []string{Marker, "config.yaml"}
+)
+
+// Anchors are the directories a root search starts from, in order, first hit
+// wins: the project the agent harness was opened on, then the working
+// directory. A non-empty start overrides both and exists so resolution can be
+// exercised against a scratch tree.
+func Anchors(start string) []string {
+	if start != "" {
+		abs, err := filepath.Abs(start)
+		if err != nil {
+			return nil
+		}
+		return []string{abs}
+	}
+	var out []string
+	for _, d := range []string{os.Getenv("CLAUDE_PROJECT_DIR"), cwd()} {
+		if d == "" || !isDir(d) {
+			continue
+		}
+		abs, err := filepath.Abs(d)
+		if err != nil {
+			continue
+		}
+		if !contains(out, abs) {
+			out = append(out, abs)
+		}
+	}
+	return out
+}
+
+// Parents yields start and every ancestor of it, up to the filesystem root.
+func Parents(start string) []string {
+	d, err := filepath.Abs(start)
+	if err != nil {
+		return nil
+	}
+	var out []string
+	for {
+		out = append(out, d)
+		parent := filepath.Dir(d)
+		if parent == d {
+			return out
+		}
+		d = parent
+	}
+}
+
+// GitDirOf is the private git directory `d/.git` points at, or "".
+//
+// Only a `.git` FILE is a pointer; in an ordinary clone `.git` is a directory
+// and there is nothing to follow.
+func GitDirOf(d string) string {
+	p := filepath.Join(d, ".git")
+	fi, err := os.Stat(p)
+	if err != nil || fi.IsDir() {
+		return ""
+	}
+	head, err := os.ReadFile(p)
+	if err != nil {
+		return ""
+	}
+	for _, line := range strings.Split(string(head), "\n") {
+		line = strings.TrimSpace(line)
+		target, ok := strings.CutPrefix(line, "gitdir:")
+		if !ok {
+			continue
+		}
+		target = strings.TrimSpace(target)
+		if target == "" {
+			return ""
+		}
+		if !filepath.IsAbs(target) {
+			target = filepath.Join(d, target)
+		}
+		abs, err := filepath.Abs(target)
+		if err != nil {
+			return ""
+		}
+		return abs
+	}
+	return ""
+}
+
+// MainWorktree is the main working tree of d's repository when d is a linked
+// worktree, or "".
+//
+// `<worktree>/.git` -> `<main>/.git/worktrees/<name>`, whose `commondir` file
+// holds a path to `<main>/.git`; the main working tree is its parent. The
+// `.git` basename check keeps this to worktrees: a submodule's `.git` is a
+// pointer too, but it points into `<super>/.git/modules/…`, and the tree it
+// belongs to is already on the parent chain.
+func MainWorktree(d string) string {
+	gitdir := GitDirOf(d)
+	if gitdir == "" || !isDir(gitdir) {
+		return ""
+	}
+	common := gitdir
+	if raw, err := os.ReadFile(filepath.Join(gitdir, "commondir")); err == nil {
+		if rel := strings.TrimSpace(string(raw)); rel != "" {
+			if abs, err := filepath.Abs(filepath.Join(gitdir, rel)); err == nil {
+				common = abs
+			}
+		}
+	}
+	if filepath.Base(common) != ".git" {
+		return ""
+	}
+	root := filepath.Dir(common)
+	abs, err := filepath.Abs(d)
+	if err != nil {
+		return ""
+	}
+	if root != "" && isDir(root) && root != abs {
+		return root
+	}
+	return ""
+}
+
+// Root is the nearest ancestor of an anchor (inclusive) holding the marker, or
+// "" when there is no project.
+//
+// A marker, not a fixed number of `..` hops: how deep a caller sits below the
+// root is an implementation detail of the layout, and the layout is not a
+// promise. Walking up means every command sees one store from anywhere inside
+// the project — including from inside the store itself — while a cd into a
+// DIFFERENT project correctly answers with that project's store.
+//
+// A linked worktree is the same project on another branch, and the marker is
+// gitignored, so it is only ever in the main checkout: the chain is searched
+// first and always wins, then the main working tree of any worktree met on it.
+func Root(start string) string {
+	for _, anchor := range Anchors(start) {
+		var hops []string
+		for _, d := range Parents(anchor) {
+			if isDir(filepath.Join(d, Marker)) {
+				return d
+			}
+			if main := MainWorktree(d); main != "" && !contains(hops, main) {
+				hops = append(hops, main)
+			}
+		}
+		// One level of indirection, never two: a main checkout is not itself a
+		// linked worktree, so this cannot chain and cannot cycle.
+		for _, hop := range hops {
+			for _, d := range Parents(hop) {
+				if isDir(filepath.Join(d, Marker)) {
+					return d
+				}
+			}
+		}
+	}
+	return ""
+}
+
+// StoreRoot is the absolute path of the issue store, or "" with no project.
+func StoreRoot(start string) string { return under(start, storeParts) }
+
+// PayloadRoot is the absolute path of the request-payload scratchpad, or "".
+//
+// A sibling of the store under the same marker, resolved by the same walk, so
+// the scratchpad and the store can never end up in two different projects —
+// and so a scratchpad can never sit INSIDE a store, where a call that touched
+// no issue would still materialize the issue directory.
+func PayloadRoot(start string) string { return under(start, payloadParts) }
+
+// ConfigPath is the absolute path of the project's tracker config, or "".
+func ConfigPath(start string) string { return under(start, configParts) }
+
+func under(start string, parts []string) string {
+	root := Root(start)
+	if root == "" {
+		return ""
+	}
+	return filepath.Join(append([]string{root}, parts...)...)
+}
+
+// NotFoundError explains why no project could be resolved, naming every
+// directory the search began from.
+//
+// The anchors, not the whole chain above them: an operator who sees the two
+// places the search started knows immediately whether it started where they
+// meant it to.
+func NotFoundError(start string) error {
+	dirs := strings.Join(Anchors(start), " and ")
+	if dirs == "" {
+		dirs = "nowhere"
+	}
+	return fmt.Errorf("no %s/ found — searched up from %s. Run `kettle init` in the project you mean to track issues in", Marker, dirs)
+}
+
+func isDir(p string) bool {
+	fi, err := os.Stat(p)
+	return err == nil && fi.IsDir()
+}
+
+func cwd() string {
+	d, err := os.Getwd()
+	if err != nil {
+		return ""
+	}
+	return d
+}
+
+func contains(xs []string, x string) bool {
+	for _, v := range xs {
+		if v == x {
+			return true
+		}
+	}
+	return false
+}

cli/internal/project/project_test.go

@@ -0,0 +1,160 @@
+package project
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// Every test here strips CLAUDE_PROJECT_DIR: it is the first anchor of the
+// walk, so the harness's own value would point every fixture at whatever
+// repository the suite happens to run in.
+func fixture(t *testing.T) string {
+	t.Helper()
+	t.Setenv("CLAUDE_PROJECT_DIR", "")
+	dir := t.TempDir()
+	// macOS hands out /var/… , a symlink to /private/var, and the walk works
+	// in resolved paths.
+	real, err := filepath.EvalSymlinks(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return real
+}
+
+func chdir(t *testing.T, dir string) {
+	t.Helper()
+	t.Chdir(dir)
+}
+
+func TestRootFindsTheNearestMarkerUpFromTheWorkingDirectory(t *testing.T) {
+	root := fixture(t)
+	deep := filepath.Join(root, "a", "b", "c")
+	if err := os.MkdirAll(deep, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.MkdirAll(filepath.Join(root, Marker), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	chdir(t, deep)
+
+	if got := Root(""); got != root {
+		t.Errorf("Root() = %q, want %q", got, root)
+	}
+	if got := StoreRoot(""); got != filepath.Join(root, Marker, "issues") {
+		t.Errorf("StoreRoot() = %q", got)
+	}
+	// The scratchpad is a sibling of the store, never inside it: a call that
+	// touches no issue must not materialize the issue directory.
+	if got := PayloadRoot(""); got != filepath.Join(root, Marker, "payload") {
+		t.Errorf("PayloadRoot() = %q", got)
+	}
+}
+
+func TestNoMarkerIsAnAnswerNotAFallback(t *testing.T) {
+	dir := fixture(t)
+	chdir(t, dir)
+
+	if got := Root(""); got != "" {
+		t.Errorf("Root() = %q, want empty — a plausible-looking directory is the failure this replaces", got)
+	}
+	if got := StoreRoot(""); got != "" {
+		t.Errorf("StoreRoot() = %q, want empty", got)
+	}
+	if err := NotFoundError(""); err == nil {
+		t.Fatal("NotFoundError must explain itself")
+	}
+}
+
+func TestTheNearestMarkerWinsOverAnAncestorOne(t *testing.T) {
+	outer := fixture(t)
+	inner := filepath.Join(outer, "vendored")
+	for _, d := range []string{filepath.Join(outer, Marker), filepath.Join(inner, Marker)} {
+		if err := os.MkdirAll(d, 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	chdir(t, inner)
+
+	if got := Root(""); got != inner {
+		t.Errorf("Root() = %q, want the nearest marker %q", got, inner)
+	}
+}
+
+// A linked worktree is a SIBLING of the main checkout, so the gitignored
+// marker is never on its parent chain. The whole sync layer once died there
+// while the tracker CLI in the same directory worked.
+func TestALinkedWorktreeResolvesToTheMainCheckout(t *testing.T) {
+	base := fixture(t)
+	main := filepath.Join(base, "repo")
+	wt := filepath.Join(base, "repo-feat")
+	gitdir := filepath.Join(main, ".git", "worktrees", "feat")
+
+	for _, d := range []string{filepath.Join(main, Marker), gitdir, wt} {
+		if err := os.MkdirAll(d, 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	// <worktree>/.git points at the private dir; commondir points back at
+	// <main>/.git, whose parent is the main working tree.
+	write(t, filepath.Join(wt, ".git"), "gitdir: "+gitdir+"\n")
+	write(t, filepath.Join(gitdir, "commondir"), "../..\n")
+
+	chdir(t, wt)
+	if got := Root(""); got != main {
+		t.Errorf("Root() = %q, want the main checkout %q", got, main)
+	}
+}
+
+// A submodule's .git is a pointer too, but it points into
+// <super>/.git/modules/…, and the tree it belongs to is already on the parent
+// chain. Following it would be a hop to nowhere.
+func TestASubmoduleIsNotAWorktree(t *testing.T) {
+	base := fixture(t)
+	sub := filepath.Join(base, "super", "sub")
+	gitdir := filepath.Join(base, "super", ".git", "modules", "sub")
+	for _, d := range []string{sub, gitdir} {
+		if err := os.MkdirAll(d, 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	write(t, filepath.Join(sub, ".git"), "gitdir: "+gitdir+"\n")
+
+	if got := MainWorktree(sub); got != "" {
+		t.Errorf("MainWorktree() = %q, want empty", got)
+	}
+}
+
+func TestAnOrdinaryCloneHasNothingToFollow(t *testing.T) {
+	dir := fixture(t)
+	if err := os.MkdirAll(filepath.Join(dir, ".git"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if got := GitDirOf(dir); got != "" {
+		t.Errorf("GitDirOf() = %q — only a .git FILE is a pointer", got)
+	}
+}
+
+func TestClaudeProjectDirIsTheFirstAnchor(t *testing.T) {
+	base := fixture(t)
+	opened := filepath.Join(base, "opened")
+	elsewhere := filepath.Join(base, "elsewhere")
+	for _, d := range []string{filepath.Join(opened, Marker), filepath.Join(elsewhere, Marker)} {
+		if err := os.MkdirAll(d, 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	t.Setenv("CLAUDE_PROJECT_DIR", opened)
+	chdir(t, elsewhere)
+
+	if got := Root(""); got != opened {
+		t.Errorf("Root() = %q, want the opened project %q", got, opened)
+	}
+}
+
+func write(t *testing.T, path, content string) {
+	t.Helper()
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatal(err)
+	}
+}

cli/internal/wire/key.go

@@ -0,0 +1,103 @@
+package wire
+
+import (
+	"fmt"
+	"regexp"
+	"strconv"
+	"strings"
+)
+
+// Repo is one repository, spelled the way a tracker spells it.
+type Repo struct {
+	Owner string
+	Name  string
+}
+
+func (r Repo) String() string {
+	if r.Zero() {
+		return ""
+	}
+	return r.Owner + "/" + r.Name
+}
+
+// Zero reports whether this names no repository. Both halves are required:
+// half a name addresses nothing.
+func (r Repo) Zero() bool { return r.Owner == "" || r.Name == "" }
+
+// ParseRepo reads owner/name.
+func ParseRepo(s string) (Repo, error) {
+	owner, name, ok := strings.Cut(strings.TrimSpace(s), "/")
+	if !ok || owner == "" || name == "" {
+		return Repo{}, fmt.Errorf("repo %q is not owner/name", s)
+	}
+	return Repo{Owner: owner, Name: name}, nil
+}
+
+// Key is a stable cross-repo handle for one issue: owner/repo#42.
+//
+// It is what the ledger is keyed by and what the `gitea:` metadata field holds,
+// so it has to survive being written to a file and read back — which is why it
+// is a repository and a number and not a bare number. A number is ambiguous the
+// moment a dependency lives in another repository, and dependencies are allowed
+// to.
+type Key struct {
+	// Repo is zero when the caller named a number and nothing else, which is
+	// the common case on a command line: "42" means "42 in this project's
+	// repository", and which repository that is, is the client's business.
+	Repo   Repo
+	Number int
+}
+
+func (k Key) String() string {
+	if k.Repo.Zero() {
+		return "#" + strconv.Itoa(k.Number)
+	}
+	return fmt.Sprintf("%s#%d", k.Repo, k.Number)
+}
+
+// In returns this key with r filled in when it names no repository of its own.
+func (k Key) In(r Repo) Key {
+	if k.Repo.Zero() {
+		k.Repo = r
+	}
+	return k
+}
+
+// The four spellings, as patterns.
+//
+// Digits and only digits after the `#`, which is the test strconv.Atoi is too
+// generous to make on its own: it accepts a sign, and `owner/repo#-3` is not a
+// handle anybody ever wrote. Anything that is not a key has to be recognizable
+// as not a key — a hand-edited metadata line and a number are told apart here
+// and nowhere else.
+var (
+	keyURL       = regexp.MustCompile(`^https?://[^/]+/([^/]+)/([^/]+)/issues/(\d+)/?$`)
+	keyQualified = regexp.MustCompile(`^([\w.-]+/[\w.-]+)#(\d+)$`)
+	keyNumber    = regexp.MustCompile(`^#?(\d+)$`)
+)
+
+// ParseKey reads an issue key: 42, #42, owner/repo#42, or the issue's URL.
+//
+// All four spellings because all four are what somebody has in hand — a number
+// from a receipt, a `#42` copied out of a body, a qualified key out of the
+// ledger, a URL pasted from a browser. Refusing three of them buys nothing.
+func ParseKey(s string) (Key, error) {
+	s = strings.TrimSpace(s)
+	if m := keyURL.FindStringSubmatch(s); m != nil {
+		n, _ := strconv.Atoi(m[3])
+		return Key{Repo: Repo{Owner: m[1], Name: m[2]}, Number: n}, nil
+	}
+	if m := keyQualified.FindStringSubmatch(s); m != nil {
+		repo, err := ParseRepo(m[1])
+		if err != nil {
+			return Key{}, err
+		}
+		n, _ := strconv.Atoi(m[2])
+		return Key{Repo: repo, Number: n}, nil
+	}
+	if m := keyNumber.FindStringSubmatch(s); m != nil {
+		n, _ := strconv.Atoi(m[1])
+		return Key{Number: n}, nil
+	}
+	return Key{}, fmt.Errorf("cannot parse issue key %q — want 42, #42, owner/repo#42, or an issue URL", s)
+}

cli/internal/wire/key_test.go

@@ -0,0 +1,47 @@
+package wire_test
+
+import (
+	"testing"
+
+	"git.noodles.cam/claude-skills/marketplace/cli/internal/wire"
+)
+
+func TestParseKey(t *testing.T) {
+	acme := wire.Repo{Owner: "acme", Name: "widgets"}
+	for _, tc := range []struct {
+		in   string
+		want wire.Key
+	}{
+		{"42", wire.Key{Number: 42}},
+		{"#42", wire.Key{Number: 42}},
+		{" acme/widgets#42 ", wire.Key{Repo: acme, Number: 42}},
+		{"https://git.example.test/acme/widgets/issues/42", wire.Key{Repo: acme, Number: 42}},
+		{"https://git.example.test/acme/widgets/issues/42/", wire.Key{Repo: acme, Number: 42}},
+	} {
+		got, err := wire.ParseKey(tc.in)
+		if err != nil {
+			t.Errorf("ParseKey(%q): %v", tc.in, err)
+			continue
+		}
+		if got != tc.want {
+			t.Errorf("ParseKey(%q) = %v, want %v", tc.in, got, tc.want)
+		}
+	}
+
+	// What is not a key has to be refused as one. `o/r#-3` is the case a bare
+	// strconv.Atoi accepts and nobody ever wrote: a key read back out of a
+	// metadata line somebody hand-edited must come back as "not a key", never
+	// as issue -3.
+	for _, bad := range []string{"not an issue", "o/r#-3", "o/r#4x", "o/r#", "o/r", ""} {
+		if got, err := wire.ParseKey(bad); err == nil {
+			t.Errorf("ParseKey(%q) = %v, want a refusal", bad, got)
+		}
+	}
+
+	if got := (wire.Key{Repo: acme, Number: 42}).String(); got != "acme/widgets#42" {
+		t.Errorf("a qualified key formatted as %q", got)
+	}
+	if got := (wire.Key{Number: 42}).In(acme).String(); got != "acme/widgets#42" {
+		t.Errorf("an unqualified key filled in as %q", got)
+	}
+}

cli/internal/wire/layering_test.go

@@ -0,0 +1,61 @@
+package wire
+
+import (
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+// The protocol is shared by two layers that may not import each other, and it
+// can only be shared because it reaches for nothing itself: no domain, no
+// configuration, no path resolution, no third party. One import from any of
+// those would drag every user of this package into that layer — which is the
+// whole reason these shapes were lifted out of the transport rather than left
+// there for the bridge to reimplement.
+//
+// The dependency walk, so a helper pulled in three packages deep is caught as
+// the same violation as one written at the top of a file.
+func TestWireDependsOnNothing(t *testing.T) {
+	out, err := exec.Command("go", "list", "-deps", ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	for _, dep := range strings.Fields(string(out)) {
+		if dep == "git.noodles.cam/claude-skills/marketplace/cli/internal/wire" {
+			continue
+		}
+		// A standard-library import path has no dot in its first element,
+		// because it has no domain name in front of it.
+		first, _, _ := strings.Cut(dep, "/")
+		if strings.Contains(first, ".") {
+			t.Errorf("the protocol imports %s — these are shapes and identifiers, and nothing else belongs here", dep)
+		}
+	}
+}
+
+// The other half: net/http and os are standard library, so "no third-party
+// imports" would not catch a transport or a file read written by hand here.
+// Name them.
+//
+// DIRECT imports, not the dependency walk — fmt reaches os on its own, and the
+// question this asks is what THIS package reaches for.
+func TestWireReachesNeitherTheNetworkNorTheDisk(t *testing.T) {
+	forbidden := map[string]string{
+		"net/http": "an HTTP call belongs in the transport",
+		"net":      "an HTTP call belongs in the transport",
+		"os":       "a shape reads no file and no environment",
+		"os/exec":  "nothing here shells out",
+		"io":       "nothing here is a stream",
+		"time":     "a timestamp crosses as the string the tracker sent",
+	}
+
+	out, err := exec.Command("go", "list", "-f", `{{join .Imports "\n"}}`, ".").Output()
+	if err != nil {
+		t.Fatalf("go list: %v", err)
+	}
+	for _, dep := range strings.Fields(string(out)) {
+		if why, bad := forbidden[dep]; bad {
+			t.Errorf("wire imports %s — %s", dep, why)
+		}
+	}
+}

cli/internal/wire/request.go

@@ -0,0 +1,59 @@
+package wire
+
+// The bodies that go up, and the shorthand that fills them.
+//
+// The omitted keys carry meaning of their own on a PATCH: a key that is absent
+// leaves the tracker's value alone, and a key that is present overwrites it. So
+// "no opinion" and "empty" must not marshal the same way, which is what every
+// pointer and every omitempty below is for.
+
+// IssueRequest is the body of a create or an edit.
+//
+// Every field is a pointer because Gitea reads an absent key as "no opinion"
+// and a present one as "make it this", and the difference is not academic: an
+// empty `ref` CLEARS the branch an issue is pinned to, and an empty `labels`
+// clears its labels. A caller meaning to change only the state would do both by
+// accident with plain zero values. Set fills a field; leaving it nil leaves the
+// tracker's copy alone.
+type IssueRequest struct {
+	Title *string `json:"title,omitempty"`
+	Body  *string `json:"body,omitempty"`
+	// Labels is a pointer because `[]` is a statement — it clears every label
+	// on the issue — while a caller that has not resolved label ids at all has
+	// no business making it. A plain slice with omitempty cannot say both.
+	Labels    *[]int64  `json:"labels,omitempty"`
+	Assignees *[]string `json:"assignees,omitempty"`
+	// Milestone is a pointer for the same reason, and because 0 is Gitea's
+	// "detach from its milestone" — a value somebody may well mean.
+	Milestone *int64  `json:"milestone,omitempty"`
+	State     *string `json:"state,omitempty"`
+	Ref       *string `json:"ref,omitempty"`
+}
+
+// LabelRequest is the body of a label create or edit — everything a repository
+// needs to make one label.
+//
+// Value fields, not pointers, and every one of them is sent: Gitea 1.26 patches
+// only what it is given, but an older server reads an absent field as empty and
+// blanks it. A label edit is rare enough that sending the unchanged name and
+// description along costs nothing and removes a way to lose them.
+//
+// It goes up as a request body of its own because `tea labels create` could not
+// set `exclusive` — the flag that makes `type/*` behave like a single choice —
+// which is the whole reason label creation went through the API rather than a
+// CLI wrapper.
+//
+// What a label MEANS — which namespaces are exclusive, what colour a severity
+// is — is not decided here. This is the shape; the taxonomy is the domain's and
+// the palette is the bridge's.
+type LabelRequest struct {
+	Name        string `json:"name"`
+	Color       string `json:"color"`
+	Description string `json:"description"`
+	Exclusive   bool   `json:"exclusive"`
+}
+
+// Set is a pointer to v, for filling the optional fields of a request. Gitea
+// reads an absent key as "no opinion" and a present one as "make it this", so
+// those fields are pointers and this is the shorthand that fills them.
+func Set[T any](v T) *T { return &v }

cli/internal/wire/wire.go

@@ -0,0 +1,155 @@
+// Package wire is the protocol: the JSON shapes a Gitea instance sends and
+// takes, the identifiers that address them, and nothing else.
+//
+// It is a package because two layers need the same vocabulary and neither may
+// import the other. internal/gitea is the transport — HTTP verbs, pagination,
+// status codes, credentials — and internal/mapping is the bridge — md <-> JSON,
+// pure functions, no network. Both have to name a Gitea issue, and when each
+// named it with a struct of its own, every command written on top of the two
+// would have had to copy a payload field by field from one spelling into the
+// other. Two copies of a shape also drift: the first field only one of them
+// learns is a field the other silently drops.
+//
+// THIS PACKAGE IMPORTS THE STANDARD LIBRARY AND NOTHING ELSE — no HTTP, no
+// filesystem, no configuration, and above all not internal/issue. That is what
+// lets the transport and the bridge share it without either one landing inside
+// the other's layer, and layering_test.go fails the moment it stops being true.
+//
+// Structs and not map[string]any, because the two representations disagreeing
+// is the failure this vocabulary exists to make debuggable: a typo in a key is
+// a compile error here and a silently dropped field there. Anything Gitea sends
+// that is not named below is not read by anybody — decoding is lossy on
+// purpose, since the tracker is not the record for anything the domain owns.
+package wire
+
+// User is whoever wrote or was assigned something.
+//
+// Only the login crosses this boundary — it is the one field of a Gitea user
+// that means anything to a command, it is what `assignees:` holds, and a
+// display name is not an identity anything can be pushed against. A transport
+// that carries the rest invites somebody to use it.
+type User struct {
+	Login string `json:"login"`
+}
+
+// Label as the tracker holds it.
+//
+// Color is hex. Gitea returns it without the leading `#` (`ee0701`) and accepts
+// it either way; both spellings are the same color, so a comparison has to
+// strip before it compares.
+type Label struct {
+	ID          int64  `json:"id"`
+	Name        string `json:"name"`
+	Color       string `json:"color"`
+	Description string `json:"description"`
+	Exclusive   bool   `json:"exclusive"`
+}
+
+// Milestone as the tracker holds it. The domain carries its title; the id
+// exists only long enough to be sent back.
+type Milestone struct {
+	ID          int64  `json:"id"`
+	Title       string `json:"title"`
+	State       string `json:"state"`
+	Description string `json:"description"`
+}
+
+// RepoRef is the repository an issue payload says it belongs to. Present on a
+// dependency listing, where the answer may well be another repository.
+type RepoRef struct {
+	Owner    string `json:"owner"`
+	Name     string `json:"name"`
+	FullName string `json:"full_name"`
+}
+
+// PullRequest is non-nil on a row that is a pull request rather than an issue.
+// Gitea's issue endpoints return both, and `type=issues` is a filter the server
+// has been known to ignore — which is why every listing re-checks it.
+type PullRequest struct {
+	Merged  bool   `json:"merged"`
+	HTMLURL string `json:"html_url"`
+}
+
+// Issue is a tracker row: a Gitea issue as the API reports it.
+//
+// Timestamps stay strings. They are written into an issue's metadata verbatim
+// and compared as opaque values; parsing them here would mean formatting them
+// back, and a round trip through a time package is a chance to hand the store a
+// different string than the tracker sent.
+type Issue struct {
+	Number int    `json:"number"`
+	Title  string `json:"title"`
+	Body   string `json:"body"`
+	State  string `json:"state"`
+	// Ref is the branch the issue is pinned to.
+	Ref string `json:"ref"`
+	// HTMLURL and UpdatedAt are the tracker's own bookkeeping and land in the
+	// domain's Extra untouched.
+	HTMLURL string `json:"html_url"`
+	// Comments is a count, not a thread: the thread is fetched separately and
+	// parked beside the issue as a sidecar.
+	Comments    int          `json:"comments"`
+	Labels      []Label      `json:"labels"`
+	Assignees   []User       `json:"assignees"`
+	Milestone   *Milestone   `json:"milestone"`
+	Repository  *RepoRef     `json:"repository"`
+	PullRequest *PullRequest `json:"pull_request"`
+	CreatedAt   string       `json:"created_at"`
+	UpdatedAt   string       `json:"updated_at"`
+}
+
+// IsPullRequest reports whether this row is a pull request.
+func (i *Issue) IsPullRequest() bool { return i.PullRequest != nil }
+
+// LabelNames are the label names, in the order the tracker listed them.
+func (i *Issue) LabelNames() []string {
+	out := make([]string, 0, len(i.Labels))
+	for _, l := range i.Labels {
+		out = append(out, l.Name)
+	}
+	return out
+}
+
+// AssigneeLogins are the assignees, as logins.
+func (i *Issue) AssigneeLogins() []string {
+	out := make([]string, 0, len(i.Assignees))
+	for _, a := range i.Assignees {
+		out = append(out, a.Login)
+	}
+	return out
+}
+
+// MilestoneTitle is the milestone's title, or "" when there is none.
+func (i *Issue) MilestoneTitle() string {
+	if i.Milestone == nil {
+		return ""
+	}
+	return i.Milestone.Title
+}
+
+// KeyIn is this issue's cross-repo handle. The payload's own repository wins
+// when it carries one — a dependency listing answers with issues from other
+// repositories — and fallback is the repository that was asked.
+func (i *Issue) KeyIn(fallback Repo) Key {
+	repo := fallback
+	if i.Repository != nil {
+		if r, err := ParseRepo(i.Repository.FullName); err == nil {
+			repo = r
+		}
+	}
+	return Key{Repo: repo, Number: i.Number}
+}
+
+// Comment is one entry in an issue's thread.
+//
+// Read only, in practice: a thread is flattened to markdown for a reader and
+// nothing writes that markdown back, which is why the rendering may be as lossy
+// as a reader needs.
+type Comment struct {
+	ID        int64  `json:"id"`
+	Body      string `json:"body"`
+	HTMLURL   string `json:"html_url"`
+	User      User   `json:"user"`
+	CreatedAt string `json:"created_at"`
+	UpdatedAt string `json:"updated_at"`
+}

cli/vendor/gopkg.in/yaml.v3/LICENSE

@@ -0,0 +1,50 @@
+
+This project is covered by two different licenses: MIT and Apache.
+
+#### MIT License ####
+
+The following files were ported to Go from C files of libyaml, and thus
+are still covered by their original MIT license, with the additional
+copyright staring in 2011 when the project was ported over:
+
+    apic.go emitterc.go parserc.go readerc.go scannerc.go
+    writerc.go yamlh.go yamlprivateh.go
+
+Copyright (c) 2006-2010 Kirill Simonov
+Copyright (c) 2006-2011 Kirill Simonov
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+### Apache License ###
+
+All the remaining project files are covered by the Apache license:
+
+Copyright (c) 2011-2019 Canonical Ltd
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.

cli/vendor/gopkg.in/yaml.v3/NOTICE

@@ -0,0 +1,13 @@
+Copyright 2011-2016 Canonical Ltd.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.

cli/vendor/gopkg.in/yaml.v3/README.md

@@ -0,0 +1,150 @@
+# YAML support for the Go language
+
+Introduction
+------------
+
+The yaml package enables Go programs to comfortably encode and decode YAML
+values. It was developed within [Canonical](https://www.canonical.com) as
+part of the [juju](https://juju.ubuntu.com) project, and is based on a
+pure Go port of the well-known [libyaml](http://pyyaml.org/wiki/LibYAML)
+C library to parse and generate YAML data quickly and reliably.
+
+Compatibility
+-------------
+
+The yaml package supports most of YAML 1.2, but preserves some behavior
+from 1.1 for backwards compatibility.
+
+Specifically, as of v3 of the yaml package:
+
+ - YAML 1.1 bools (_yes/no, on/off_) are supported as long as they are being
+   decoded into a typed bool value. Otherwise they behave as a string. Booleans
+   in YAML 1.2 are _true/false_ only.
+ - Octals encode and decode as _0777_ per YAML 1.1, rather than _0o777_
+   as specified in YAML 1.2, because most parsers still use the old format.
+   Octals in the  _0o777_ format are supported though, so new files work.
+ - Does not support base-60 floats. These are gone from YAML 1.2, and were
+   actually never supported by this package as it's clearly a poor choice.
+
+and offers backwards
+compatibility with YAML 1.1 in some cases.
+1.2, including support for
+anchors, tags, map merging, etc. Multi-document unmarshalling is not yet
+implemented, and base-60 floats from YAML 1.1 are purposefully not
+supported since they're a poor design and are gone in YAML 1.2.
+
+Installation and usage
+----------------------
+
+The import path for the package is *gopkg.in/yaml.v3*.
+
+To install it, run:
+
+    go get gopkg.in/yaml.v3
+
+API documentation
+-----------------
+
+If opened in a browser, the import path itself leads to the API documentation:
+
+  - [https://gopkg.in/yaml.v3](https://gopkg.in/yaml.v3)
+
+API stability
+-------------
+
+The package API for yaml v3 will remain stable as described in [gopkg.in](https://gopkg.in).
+
+
+License
+-------
+
+The yaml package is licensed under the MIT and Apache License 2.0 licenses.
+Please see the LICENSE file for details.
+
+
+Example
+-------
+
+```Go
+package main
+
+import (
+        "fmt"
+        "log"
+
+        "gopkg.in/yaml.v3"
+)
+
+var data = `
+a: Easy!
+b:
+  c: 2
+  d: [3, 4]
+`
+
+// Note: struct fields must be public in order for unmarshal to
+// correctly populate the data.
+type T struct {
+        A string
+        B struct {
+                RenamedC int   `yaml:"c"`
+                D        []int `yaml:",flow"`
+        }
+}
+
+func main() {
+        t := T{}
+    
+        err := yaml.Unmarshal([]byte(data), &t)
+        if err != nil {
+                log.Fatalf("error: %v", err)
+        }
+        fmt.Printf("--- t:\n%v\n\n", t)
+    
+        d, err := yaml.Marshal(&t)
+        if err != nil {
+                log.Fatalf("error: %v", err)
+        }
+        fmt.Printf("--- t dump:\n%s\n\n", string(d))
+    
+        m := make(map[interface{}]interface{})
+    
+        err = yaml.Unmarshal([]byte(data), &m)
+        if err != nil {
+                log.Fatalf("error: %v", err)
+        }
+        fmt.Printf("--- m:\n%v\n\n", m)
+    
+        d, err = yaml.Marshal(&m)
+        if err != nil {
+                log.Fatalf("error: %v", err)
+        }
+        fmt.Printf("--- m dump:\n%s\n\n", string(d))
+}
+```
+
+This example will generate the following output:
+
+```
+--- t:
+{Easy! {2 [3 4]}}
+
+--- t dump:
+a: Easy!
+b:
+  c: 2
+  d: [3, 4]
+
+
+--- m:
+map[a:Easy! b:map[c:2 d:[3 4]]]
+
+--- m dump:
+a: Easy!
+b:
+  c: 2
+  d:
+  - 3
+  - 4
+```
+

cli/vendor/gopkg.in/yaml.v3/apic.go

@@ -0,0 +1,747 @@
+// 
+// Copyright (c) 2011-2019 Canonical Ltd
+// Copyright (c) 2006-2010 Kirill Simonov
+// 
+// Permission is hereby granted, free of charge, to any person obtaining a copy of
+// this software and associated documentation files (the "Software"), to deal in
+// the Software without restriction, including without limitation the rights to
+// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+// of the Software, and to permit persons to whom the Software is furnished to do
+// so, subject to the following conditions:
+// 
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+package yaml
+
+import (
+	"io"
+)
+
+func yaml_insert_token(parser *yaml_parser_t, pos int, token *yaml_token_t) {
+	//fmt.Println("yaml_insert_token", "pos:", pos, "typ:", token.typ, "head:", parser.tokens_head, "len:", len(parser.tokens))
+
+	// Check if we can move the queue at the beginning of the buffer.
+	if parser.tokens_head > 0 && len(parser.tokens) == cap(parser.tokens) {
+		if parser.tokens_head != len(parser.tokens) {
+			copy(parser.tokens, parser.tokens[parser.tokens_head:])
+		}
+		parser.tokens = parser.tokens[:len(parser.tokens)-parser.tokens_head]
+		parser.tokens_head = 0
+	}
+	parser.tokens = append(parser.tokens, *token)
+	if pos < 0 {
+		return
+	}
+	copy(parser.tokens[parser.tokens_head+pos+1:], parser.tokens[parser.tokens_head+pos:])
+	parser.tokens[parser.tokens_head+pos] = *token
+}
+
+// Create a new parser object.
+func yaml_parser_initialize(parser *yaml_parser_t) bool {
+	*parser = yaml_parser_t{
+		raw_buffer: make([]byte, 0, input_raw_buffer_size),
+		buffer:     make([]byte, 0, input_buffer_size),
+	}
+	return true
+}
+
+// Destroy a parser object.
+func yaml_parser_delete(parser *yaml_parser_t) {
+	*parser = yaml_parser_t{}
+}
+
+// String read handler.
+func yaml_string_read_handler(parser *yaml_parser_t, buffer []byte) (n int, err error) {
+	if parser.input_pos == len(parser.input) {
+		return 0, io.EOF
+	}
+	n = copy(buffer, parser.input[parser.input_pos:])
+	parser.input_pos += n
+	return n, nil
+}
+
+// Reader read handler.
+func yaml_reader_read_handler(parser *yaml_parser_t, buffer []byte) (n int, err error) {
+	return parser.input_reader.Read(buffer)
+}
+
+// Set a string input.
+func yaml_parser_set_input_string(parser *yaml_parser_t, input []byte) {
+	if parser.read_handler != nil {
+		panic("must set the input source only once")
+	}
+	parser.read_handler = yaml_string_read_handler
+	parser.input = input
+	parser.input_pos = 0
+}
+
+// Set a file input.
+func yaml_parser_set_input_reader(parser *yaml_parser_t, r io.Reader) {
+	if parser.read_handler != nil {
+		panic("must set the input source only once")
+	}
+	parser.read_handler = yaml_reader_read_handler
+	parser.input_reader = r
+}
+
+// Set the source encoding.
+func yaml_parser_set_encoding(parser *yaml_parser_t, encoding yaml_encoding_t) {
+	if parser.encoding != yaml_ANY_ENCODING {
+		panic("must set the encoding only once")
+	}
+	parser.encoding = encoding
+}
+
+// Create a new emitter object.
+func yaml_emitter_initialize(emitter *yaml_emitter_t) {
+	*emitter = yaml_emitter_t{
+		buffer:     make([]byte, output_buffer_size),
+		raw_buffer: make([]byte, 0, output_raw_buffer_size),
+		states:     make([]yaml_emitter_state_t, 0, initial_stack_size),
+		events:     make([]yaml_event_t, 0, initial_queue_size),
+		best_width: -1,
+	}
+}
+
+// Destroy an emitter object.
+func yaml_emitter_delete(emitter *yaml_emitter_t) {
+	*emitter = yaml_emitter_t{}
+}
+
+// String write handler.
+func yaml_string_write_handler(emitter *yaml_emitter_t, buffer []byte) error {
+	*emitter.output_buffer = append(*emitter.output_buffer, buffer...)
+	return nil
+}
+
+// yaml_writer_write_handler uses emitter.output_writer to write the
+// emitted text.
+func yaml_writer_write_handler(emitter *yaml_emitter_t, buffer []byte) error {
+	_, err := emitter.output_writer.Write(buffer)
+	return err
+}
+
+// Set a string output.
+func yaml_emitter_set_output_string(emitter *yaml_emitter_t, output_buffer *[]byte) {
+	if emitter.write_handler != nil {
+		panic("must set the output target only once")
+	}
+	emitter.write_handler = yaml_string_write_handler
+	emitter.output_buffer = output_buffer
+}
+
+// Set a file output.
+func yaml_emitter_set_output_writer(emitter *yaml_emitter_t, w io.Writer) {
+	if emitter.write_handler != nil {
+		panic("must set the output target only once")
+	}
+	emitter.write_handler = yaml_writer_write_handler
+	emitter.output_writer = w
+}
+
+// Set the output encoding.
+func yaml_emitter_set_encoding(emitter *yaml_emitter_t, encoding yaml_encoding_t) {
+	if emitter.encoding != yaml_ANY_ENCODING {
+		panic("must set the output encoding only once")
+	}
+	emitter.encoding = encoding
+}
+
+// Set the canonical output style.
+func yaml_emitter_set_canonical(emitter *yaml_emitter_t, canonical bool) {
+	emitter.canonical = canonical
+}
+
+// Set the indentation increment.
+func yaml_emitter_set_indent(emitter *yaml_emitter_t, indent int) {
+	if indent < 2 || indent > 9 {
+		indent = 2
+	}
+	emitter.best_indent = indent
+}
+
+// Set the preferred line width.
+func yaml_emitter_set_width(emitter *yaml_emitter_t, width int) {
+	if width < 0 {
+		width = -1
+	}
+	emitter.best_width = width
+}
+
+// Set if unescaped non-ASCII characters are allowed.
+func yaml_emitter_set_unicode(emitter *yaml_emitter_t, unicode bool) {
+	emitter.unicode = unicode
+}
+
+// Set the preferred line break character.
+func yaml_emitter_set_break(emitter *yaml_emitter_t, line_break yaml_break_t) {
+	emitter.line_break = line_break
+}
+
+///*
+// * Destroy a token object.
+// */
+//
+//YAML_DECLARE(void)
+//yaml_token_delete(yaml_token_t *token)
+//{
+//    assert(token);  // Non-NULL token object expected.
+//
+//    switch (token.type)
+//    {
+//        case YAML_TAG_DIRECTIVE_TOKEN:
+//            yaml_free(token.data.tag_directive.handle);
+//            yaml_free(token.data.tag_directive.prefix);
+//            break;
+//
+//        case YAML_ALIAS_TOKEN:
+//            yaml_free(token.data.alias.value);
+//            break;
+//
+//        case YAML_ANCHOR_TOKEN:
+//            yaml_free(token.data.anchor.value);
+//            break;
+//
+//        case YAML_TAG_TOKEN:
+//            yaml_free(token.data.tag.handle);
+//            yaml_free(token.data.tag.suffix);
+//            break;
+//
+//        case YAML_SCALAR_TOKEN:
+//            yaml_free(token.data.scalar.value);
+//            break;
+//
+//        default:
+//            break;
+//    }
+//
+//    memset(token, 0, sizeof(yaml_token_t));
+//}
+//
+///*
+// * Check if a string is a valid UTF-8 sequence.
+// *
+// * Check 'reader.c' for more details on UTF-8 encoding.
+// */
+//
+//static int
+//yaml_check_utf8(yaml_char_t *start, size_t length)
+//{
+//    yaml_char_t *end = start+length;
+//    yaml_char_t *pointer = start;
+//
+//    while (pointer < end) {
+//        unsigned char octet;
+//        unsigned int width;
+//        unsigned int value;
+//        size_t k;
+//
+//        octet = pointer[0];
+//        width = (octet & 0x80) == 0x00 ? 1 :
+//                (octet & 0xE0) == 0xC0 ? 2 :
+//                (octet & 0xF0) == 0xE0 ? 3 :
+//                (octet & 0xF8) == 0xF0 ? 4 : 0;
+//        value = (octet & 0x80) == 0x00 ? octet & 0x7F :
+//                (octet & 0xE0) == 0xC0 ? octet & 0x1F :
+//                (octet & 0xF0) == 0xE0 ? octet & 0x0F :
+//                (octet & 0xF8) == 0xF0 ? octet & 0x07 : 0;
+//        if (!width) return 0;
+//        if (pointer+width > end) return 0;
+//        for (k = 1; k < width; k ++) {
+//            octet = pointer[k];
+//            if ((octet & 0xC0) != 0x80) return 0;
+//            value = (value << 6) + (octet & 0x3F);
+//        }
+//        if (!((width == 1) ||
+//            (width == 2 && value >= 0x80) ||
+//            (width == 3 && value >= 0x800) ||
+//            (width == 4 && value >= 0x10000))) return 0;
+//
+//        pointer += width;
+//    }
+//
+//    return 1;
+//}
+//
+
+// Create STREAM-START.
+func yaml_stream_start_event_initialize(event *yaml_event_t, encoding yaml_encoding_t) {
+	*event = yaml_event_t{
+		typ:      yaml_STREAM_START_EVENT,
+		encoding: encoding,
+	}
+}
+
+// Create STREAM-END.
+func yaml_stream_end_event_initialize(event *yaml_event_t) {
+	*event = yaml_event_t{
+		typ: yaml_STREAM_END_EVENT,
+	}
+}
+
+// Create DOCUMENT-START.
+func yaml_document_start_event_initialize(
+	event *yaml_event_t,
+	version_directive *yaml_version_directive_t,
+	tag_directives []yaml_tag_directive_t,
+	implicit bool,
+) {
+	*event = yaml_event_t{
+		typ:               yaml_DOCUMENT_START_EVENT,
+		version_directive: version_directive,
+		tag_directives:    tag_directives,
+		implicit:          implicit,
+	}
+}
+
+// Create DOCUMENT-END.
+func yaml_document_end_event_initialize(event *yaml_event_t, implicit bool) {
+	*event = yaml_event_t{
+		typ:      yaml_DOCUMENT_END_EVENT,
+		implicit: implicit,
+	}
+}
+
+// Create ALIAS.
+func yaml_alias_event_initialize(event *yaml_event_t, anchor []byte) bool {
+	*event = yaml_event_t{
+		typ:    yaml_ALIAS_EVENT,
+		anchor: anchor,
+	}
+	return true
+}
+
+// Create SCALAR.
+func yaml_scalar_event_initialize(event *yaml_event_t, anchor, tag, value []byte, plain_implicit, quoted_implicit bool, style yaml_scalar_style_t) bool {
+	*event = yaml_event_t{
+		typ:             yaml_SCALAR_EVENT,
+		anchor:          anchor,
+		tag:             tag,
+		value:           value,
+		implicit:        plain_implicit,
+		quoted_implicit: quoted_implicit,
+		style:           yaml_style_t(style),
+	}
+	return true
+}
+
+// Create SEQUENCE-START.
+func yaml_sequence_start_event_initialize(event *yaml_event_t, anchor, tag []byte, implicit bool, style yaml_sequence_style_t) bool {
+	*event = yaml_event_t{
+		typ:      yaml_SEQUENCE_START_EVENT,
+		anchor:   anchor,
+		tag:      tag,
+		implicit: implicit,
+		style:    yaml_style_t(style),
+	}
+	return true
+}
+
+// Create SEQUENCE-END.
+func yaml_sequence_end_event_initialize(event *yaml_event_t) bool {
+	*event = yaml_event_t{
+		typ: yaml_SEQUENCE_END_EVENT,
+	}
+	return true
+}
+
+// Create MAPPING-START.
+func yaml_mapping_start_event_initialize(event *yaml_event_t, anchor, tag []byte, implicit bool, style yaml_mapping_style_t) {
+	*event = yaml_event_t{
+		typ:      yaml_MAPPING_START_EVENT,
+		anchor:   anchor,
+		tag:      tag,
+		implicit: implicit,
+		style:    yaml_style_t(style),
+	}
+}
+
+// Create MAPPING-END.
+func yaml_mapping_end_event_initialize(event *yaml_event_t) {
+	*event = yaml_event_t{
+		typ: yaml_MAPPING_END_EVENT,
+	}
+}
+
+// Destroy an event object.
+func yaml_event_delete(event *yaml_event_t) {
+	*event = yaml_event_t{}
+}
+
+///*
+// * Create a document object.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_initialize(document *yaml_document_t,
+//        version_directive *yaml_version_directive_t,
+//        tag_directives_start *yaml_tag_directive_t,
+//        tag_directives_end *yaml_tag_directive_t,
+//        start_implicit int, end_implicit int)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//    struct {
+//        start *yaml_node_t
+//        end *yaml_node_t
+//        top *yaml_node_t
+//    } nodes = { NULL, NULL, NULL }
+//    version_directive_copy *yaml_version_directive_t = NULL
+//    struct {
+//        start *yaml_tag_directive_t
+//        end *yaml_tag_directive_t
+//        top *yaml_tag_directive_t
+//    } tag_directives_copy = { NULL, NULL, NULL }
+//    value yaml_tag_directive_t = { NULL, NULL }
+//    mark yaml_mark_t = { 0, 0, 0 }
+//
+//    assert(document) // Non-NULL document object is expected.
+//    assert((tag_directives_start && tag_directives_end) ||
+//            (tag_directives_start == tag_directives_end))
+//                            // Valid tag directives are expected.
+//
+//    if (!STACK_INIT(&context, nodes, INITIAL_STACK_SIZE)) goto error
+//
+//    if (version_directive) {
+//        version_directive_copy = yaml_malloc(sizeof(yaml_version_directive_t))
+//        if (!version_directive_copy) goto error
+//        version_directive_copy.major = version_directive.major
+//        version_directive_copy.minor = version_directive.minor
+//    }
+//
+//    if (tag_directives_start != tag_directives_end) {
+//        tag_directive *yaml_tag_directive_t
+//        if (!STACK_INIT(&context, tag_directives_copy, INITIAL_STACK_SIZE))
+//            goto error
+//        for (tag_directive = tag_directives_start
+//                tag_directive != tag_directives_end; tag_directive ++) {
+//            assert(tag_directive.handle)
+//            assert(tag_directive.prefix)
+//            if (!yaml_check_utf8(tag_directive.handle,
+//                        strlen((char *)tag_directive.handle)))
+//                goto error
+//            if (!yaml_check_utf8(tag_directive.prefix,
+//                        strlen((char *)tag_directive.prefix)))
+//                goto error
+//            value.handle = yaml_strdup(tag_directive.handle)
+//            value.prefix = yaml_strdup(tag_directive.prefix)
+//            if (!value.handle || !value.prefix) goto error
+//            if (!PUSH(&context, tag_directives_copy, value))
+//                goto error
+//            value.handle = NULL
+//            value.prefix = NULL
+//        }
+//    }
+//
+//    DOCUMENT_INIT(*document, nodes.start, nodes.end, version_directive_copy,
+//            tag_directives_copy.start, tag_directives_copy.top,
+//            start_implicit, end_implicit, mark, mark)
+//
+//    return 1
+//
+//error:
+//    STACK_DEL(&context, nodes)
+//    yaml_free(version_directive_copy)
+//    while (!STACK_EMPTY(&context, tag_directives_copy)) {
+//        value yaml_tag_directive_t = POP(&context, tag_directives_copy)
+//        yaml_free(value.handle)
+//        yaml_free(value.prefix)
+//    }
+//    STACK_DEL(&context, tag_directives_copy)
+//    yaml_free(value.handle)
+//    yaml_free(value.prefix)
+//
+//    return 0
+//}
+//
+///*
+// * Destroy a document object.
+// */
+//
+//YAML_DECLARE(void)
+//yaml_document_delete(document *yaml_document_t)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//    tag_directive *yaml_tag_directive_t
+//
+//    context.error = YAML_NO_ERROR // Eliminate a compiler warning.
+//
+//    assert(document) // Non-NULL document object is expected.
+//
+//    while (!STACK_EMPTY(&context, document.nodes)) {
+//        node yaml_node_t = POP(&context, document.nodes)
+//        yaml_free(node.tag)
+//        switch (node.type) {
+//            case YAML_SCALAR_NODE:
+//                yaml_free(node.data.scalar.value)
+//                break
+//            case YAML_SEQUENCE_NODE:
+//                STACK_DEL(&context, node.data.sequence.items)
+//                break
+//            case YAML_MAPPING_NODE:
+//                STACK_DEL(&context, node.data.mapping.pairs)
+//                break
+//            default:
+//                assert(0) // Should not happen.
+//        }
+//    }
+//    STACK_DEL(&context, document.nodes)
+//
+//    yaml_free(document.version_directive)
+//    for (tag_directive = document.tag_directives.start
+//            tag_directive != document.tag_directives.end
+//            tag_directive++) {
+//        yaml_free(tag_directive.handle)
+//        yaml_free(tag_directive.prefix)
+//    }
+//    yaml_free(document.tag_directives.start)
+//
+//    memset(document, 0, sizeof(yaml_document_t))
+//}
+//
+///**
+// * Get a document node.
+// */
+//
+//YAML_DECLARE(yaml_node_t *)
+//yaml_document_get_node(document *yaml_document_t, index int)
+//{
+//    assert(document) // Non-NULL document object is expected.
+//
+//    if (index > 0 && document.nodes.start + index <= document.nodes.top) {
+//        return document.nodes.start + index - 1
+//    }
+//    return NULL
+//}
+//
+///**
+// * Get the root object.
+// */
+//
+//YAML_DECLARE(yaml_node_t *)
+//yaml_document_get_root_node(document *yaml_document_t)
+//{
+//    assert(document) // Non-NULL document object is expected.
+//
+//    if (document.nodes.top != document.nodes.start) {
+//        return document.nodes.start
+//    }
+//    return NULL
+//}
+//
+///*
+// * Add a scalar node to a document.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_add_scalar(document *yaml_document_t,
+//        tag *yaml_char_t, value *yaml_char_t, length int,
+//        style yaml_scalar_style_t)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//    mark yaml_mark_t = { 0, 0, 0 }
+//    tag_copy *yaml_char_t = NULL
+//    value_copy *yaml_char_t = NULL
+//    node yaml_node_t
+//
+//    assert(document) // Non-NULL document object is expected.
+//    assert(value) // Non-NULL value is expected.
+//
+//    if (!tag) {
+//        tag = (yaml_char_t *)YAML_DEFAULT_SCALAR_TAG
+//    }
+//
+//    if (!yaml_check_utf8(tag, strlen((char *)tag))) goto error
+//    tag_copy = yaml_strdup(tag)
+//    if (!tag_copy) goto error
+//
+//    if (length < 0) {
+//        length = strlen((char *)value)
+//    }
+//
+//    if (!yaml_check_utf8(value, length)) goto error
+//    value_copy = yaml_malloc(length+1)
+//    if (!value_copy) goto error
+//    memcpy(value_copy, value, length)
+//    value_copy[length] = '\0'
+//
+//    SCALAR_NODE_INIT(node, tag_copy, value_copy, length, style, mark, mark)
+//    if (!PUSH(&context, document.nodes, node)) goto error
+//
+//    return document.nodes.top - document.nodes.start
+//
+//error:
+//    yaml_free(tag_copy)
+//    yaml_free(value_copy)
+//
+//    return 0
+//}
+//
+///*
+// * Add a sequence node to a document.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_add_sequence(document *yaml_document_t,
+//        tag *yaml_char_t, style yaml_sequence_style_t)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//    mark yaml_mark_t = { 0, 0, 0 }
+//    tag_copy *yaml_char_t = NULL
+//    struct {
+//        start *yaml_node_item_t
+//        end *yaml_node_item_t
+//        top *yaml_node_item_t
+//    } items = { NULL, NULL, NULL }
+//    node yaml_node_t
+//
+//    assert(document) // Non-NULL document object is expected.
+//
+//    if (!tag) {
+//        tag = (yaml_char_t *)YAML_DEFAULT_SEQUENCE_TAG
+//    }
+//
+//    if (!yaml_check_utf8(tag, strlen((char *)tag))) goto error
+//    tag_copy = yaml_strdup(tag)
+//    if (!tag_copy) goto error
+//
+//    if (!STACK_INIT(&context, items, INITIAL_STACK_SIZE)) goto error
+//
+//    SEQUENCE_NODE_INIT(node, tag_copy, items.start, items.end,
+//            style, mark, mark)
+//    if (!PUSH(&context, document.nodes, node)) goto error
+//
+//    return document.nodes.top - document.nodes.start
+//
+//error:
+//    STACK_DEL(&context, items)
+//    yaml_free(tag_copy)
+//
+//    return 0
+//}
+//
+///*
+// * Add a mapping node to a document.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_add_mapping(document *yaml_document_t,
+//        tag *yaml_char_t, style yaml_mapping_style_t)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//    mark yaml_mark_t = { 0, 0, 0 }
+//    tag_copy *yaml_char_t = NULL
+//    struct {
+//        start *yaml_node_pair_t
+//        end *yaml_node_pair_t
+//        top *yaml_node_pair_t
+//    } pairs = { NULL, NULL, NULL }
+//    node yaml_node_t
+//
+//    assert(document) // Non-NULL document object is expected.
+//
+//    if (!tag) {
+//        tag = (yaml_char_t *)YAML_DEFAULT_MAPPING_TAG
+//    }
+//
+//    if (!yaml_check_utf8(tag, strlen((char *)tag))) goto error
+//    tag_copy = yaml_strdup(tag)
+//    if (!tag_copy) goto error
+//
+//    if (!STACK_INIT(&context, pairs, INITIAL_STACK_SIZE)) goto error
+//
+//    MAPPING_NODE_INIT(node, tag_copy, pairs.start, pairs.end,
+//            style, mark, mark)
+//    if (!PUSH(&context, document.nodes, node)) goto error
+//
+//    return document.nodes.top - document.nodes.start
+//
+//error:
+//    STACK_DEL(&context, pairs)
+//    yaml_free(tag_copy)
+//
+//    return 0
+//}
+//
+///*
+// * Append an item to a sequence node.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_append_sequence_item(document *yaml_document_t,
+//        sequence int, item int)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//
+//    assert(document) // Non-NULL document is required.
+//    assert(sequence > 0
+//            && document.nodes.start + sequence <= document.nodes.top)
+//                            // Valid sequence id is required.
+//    assert(document.nodes.start[sequence-1].type == YAML_SEQUENCE_NODE)
+//                            // A sequence node is required.
+//    assert(item > 0 && document.nodes.start + item <= document.nodes.top)
+//                            // Valid item id is required.
+//
+//    if (!PUSH(&context,
+//                document.nodes.start[sequence-1].data.sequence.items, item))
+//        return 0
+//
+//    return 1
+//}
+//
+///*
+// * Append a pair of a key and a value to a mapping node.
+// */
+//
+//YAML_DECLARE(int)
+//yaml_document_append_mapping_pair(document *yaml_document_t,
+//        mapping int, key int, value int)
+//{
+//    struct {
+//        error yaml_error_type_t
+//    } context
+//
+//    pair yaml_node_pair_t
+//
+//    assert(document) // Non-NULL document is required.
+//    assert(mapping > 0
+//            && document.nodes.start + mapping <= document.nodes.top)
+//                            // Valid mapping id is required.
+//    assert(document.nodes.start[mapping-1].type == YAML_MAPPING_NODE)
+//                            // A mapping node is required.
+//    assert(key > 0 && document.nodes.start + key <= document.nodes.top)
+//                            // Valid key id is required.
+//    assert(value > 0 && document.nodes.start + value <= document.nodes.top)
+//                            // Valid value id is required.
+//
+//    pair.key = key
+//    pair.value = value
+//
+//    if (!PUSH(&context,
+//                document.nodes.start[mapping-1].data.mapping.pairs, pair))
+//        return 0
+//
+//    return 1
+//}
+//
+//

cli/vendor/gopkg.in/yaml.v3/encode.go

@@ -0,0 +1,577 @@
+//
+// Copyright (c) 2011-2019 Canonical Ltd
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package yaml
+
+import (
+	"encoding"
+	"fmt"
+	"io"
+	"reflect"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+	"time"
+	"unicode/utf8"
+)
+
+type encoder struct {
+	emitter  yaml_emitter_t
+	event    yaml_event_t
+	out      []byte
+	flow     bool
+	indent   int
+	doneInit bool
+}
+
+func newEncoder() *encoder {
+	e := &encoder{}
+	yaml_emitter_initialize(&e.emitter)
+	yaml_emitter_set_output_string(&e.emitter, &e.out)
+	yaml_emitter_set_unicode(&e.emitter, true)
+	return e
+}
+
+func newEncoderWithWriter(w io.Writer) *encoder {
+	e := &encoder{}
+	yaml_emitter_initialize(&e.emitter)
+	yaml_emitter_set_output_writer(&e.emitter, w)
+	yaml_emitter_set_unicode(&e.emitter, true)
+	return e
+}
+
+func (e *encoder) init() {
+	if e.doneInit {
+		return
+	}
+	if e.indent == 0 {
+		e.indent = 4
+	}
+	e.emitter.best_indent = e.indent
+	yaml_stream_start_event_initialize(&e.event, yaml_UTF8_ENCODING)
+	e.emit()
+	e.doneInit = true
+}
+
+func (e *encoder) finish() {
+	e.emitter.open_ended = false
+	yaml_stream_end_event_initialize(&e.event)
+	e.emit()
+}
+
+func (e *encoder) destroy() {
+	yaml_emitter_delete(&e.emitter)
+}
+
+func (e *encoder) emit() {
+	// This will internally delete the e.event value.
+	e.must(yaml_emitter_emit(&e.emitter, &e.event))
+}
+
+func (e *encoder) must(ok bool) {
+	if !ok {
+		msg := e.emitter.problem
+		if msg == "" {
+			msg = "unknown problem generating YAML content"
+		}
+		failf("%s", msg)
+	}
+}
+
+func (e *encoder) marshalDoc(tag string, in reflect.Value) {
+	e.init()
+	var node *Node
+	if in.IsValid() {
+		node, _ = in.Interface().(*Node)
+	}
+	if node != nil && node.Kind == DocumentNode {
+		e.nodev(in)
+	} else {
+		yaml_document_start_event_initialize(&e.event, nil, nil, true)
+		e.emit()
+		e.marshal(tag, in)
+		yaml_document_end_event_initialize(&e.event, true)
+		e.emit()
+	}
+}
+
+func (e *encoder) marshal(tag string, in reflect.Value) {
+	tag = shortTag(tag)
+	if !in.IsValid() || in.Kind() == reflect.Ptr && in.IsNil() {
+		e.nilv()
+		return
+	}
+	iface := in.Interface()
+	switch value := iface.(type) {
+	case *Node:
+		e.nodev(in)
+		return
+	case Node:
+		if !in.CanAddr() {
+			var n = reflect.New(in.Type()).Elem()
+			n.Set(in)
+			in = n
+		}
+		e.nodev(in.Addr())
+		return
+	case time.Time:
+		e.timev(tag, in)
+		return
+	case *time.Time:
+		e.timev(tag, in.Elem())
+		return
+	case time.Duration:
+		e.stringv(tag, reflect.ValueOf(value.String()))
+		return
+	case Marshaler:
+		v, err := value.MarshalYAML()
+		if err != nil {
+			fail(err)
+		}
+		if v == nil {
+			e.nilv()
+			return
+		}
+		e.marshal(tag, reflect.ValueOf(v))
+		return
+	case encoding.TextMarshaler:
+		text, err := value.MarshalText()
+		if err != nil {
+			fail(err)
+		}
+		in = reflect.ValueOf(string(text))
+	case nil:
+		e.nilv()
+		return
+	}
+	switch in.Kind() {
+	case reflect.Interface:
+		e.marshal(tag, in.Elem())
+	case reflect.Map:
+		e.mapv(tag, in)
+	case reflect.Ptr:
+		e.marshal(tag, in.Elem())
+	case reflect.Struct:
+		e.structv(tag, in)
+	case reflect.Slice, reflect.Array:
+		e.slicev(tag, in)
+	case reflect.String:
+		e.stringv(tag, in)
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		e.intv(tag, in)
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		e.uintv(tag, in)
+	case reflect.Float32, reflect.Float64:
+		e.floatv(tag, in)
+	case reflect.Bool:
+		e.boolv(tag, in)
+	default:
+		panic("cannot marshal type: " + in.Type().String())
+	}
+}
+
+func (e *encoder) mapv(tag string, in reflect.Value) {
+	e.mappingv(tag, func() {
+		keys := keyList(in.MapKeys())
+		sort.Sort(keys)
+		for _, k := range keys {
+			e.marshal("", k)
+			e.marshal("", in.MapIndex(k))
+		}
+	})
+}
+
+func (e *encoder) fieldByIndex(v reflect.Value, index []int) (field reflect.Value) {
+	for _, num := range index {
+		for {
+			if v.Kind() == reflect.Ptr {
+				if v.IsNil() {
+					return reflect.Value{}
+				}
+				v = v.Elem()
+				continue
+			}
+			break
+		}
+		v = v.Field(num)
+	}
+	return v
+}
+
+func (e *encoder) structv(tag string, in reflect.Value) {
+	sinfo, err := getStructInfo(in.Type())
+	if err != nil {
+		panic(err)
+	}
+	e.mappingv(tag, func() {
+		for _, info := range sinfo.FieldsList {
+			var value reflect.Value
+			if info.Inline == nil {
+				value = in.Field(info.Num)
+			} else {
+				value = e.fieldByIndex(in, info.Inline)
+				if !value.IsValid() {
+					continue
+				}
+			}
+			if info.OmitEmpty && isZero(value) {
+				continue
+			}
+			e.marshal("", reflect.ValueOf(info.Key))
+			e.flow = info.Flow
+			e.marshal("", value)
+		}
+		if sinfo.InlineMap >= 0 {
+			m := in.Field(sinfo.InlineMap)
+			if m.Len() > 0 {
+				e.flow = false
+				keys := keyList(m.MapKeys())
+				sort.Sort(keys)
+				for _, k := range keys {
+					if _, found := sinfo.FieldsMap[k.String()]; found {
+						panic(fmt.Sprintf("cannot have key %q in inlined map: conflicts with struct field", k.String()))
+					}
+					e.marshal("", k)
+					e.flow = false
+					e.marshal("", m.MapIndex(k))
+				}
+			}
+		}
+	})
+}
+
+func (e *encoder) mappingv(tag string, f func()) {
+	implicit := tag == ""
+	style := yaml_BLOCK_MAPPING_STYLE
+	if e.flow {
+		e.flow = false
+		style = yaml_FLOW_MAPPING_STYLE
+	}
+	yaml_mapping_start_event_initialize(&e.event, nil, []byte(tag), implicit, style)
+	e.emit()
+	f()
+	yaml_mapping_end_event_initialize(&e.event)
+	e.emit()
+}
+
+func (e *encoder) slicev(tag string, in reflect.Value) {
+	implicit := tag == ""
+	style := yaml_BLOCK_SEQUENCE_STYLE
+	if e.flow {
+		e.flow = false
+		style = yaml_FLOW_SEQUENCE_STYLE
+	}
+	e.must(yaml_sequence_start_event_initialize(&e.event, nil, []byte(tag), implicit, style))
+	e.emit()
+	n := in.Len()
+	for i := 0; i < n; i++ {
+		e.marshal("", in.Index(i))
+	}
+	e.must(yaml_sequence_end_event_initialize(&e.event))
+	e.emit()
+}
+
+// isBase60 returns whether s is in base 60 notation as defined in YAML 1.1.
+//
+// The base 60 float notation in YAML 1.1 is a terrible idea and is unsupported
+// in YAML 1.2 and by this package, but these should be marshalled quoted for
+// the time being for compatibility with other parsers.
+func isBase60Float(s string) (result bool) {
+	// Fast path.
+	if s == "" {
+		return false
+	}
+	c := s[0]
+	if !(c == '+' || c == '-' || c >= '0' && c <= '9') || strings.IndexByte(s, ':') < 0 {
+		return false
+	}
+	// Do the full match.
+	return base60float.MatchString(s)
+}
+
+// From http://yaml.org/type/float.html, except the regular expression there
+// is bogus. In practice parsers do not enforce the "\.[0-9_]*" suffix.
+var base60float = regexp.MustCompile(`^[-+]?[0-9][0-9_]*(?::[0-5]?[0-9])+(?:\.[0-9_]*)?$`)
+
+// isOldBool returns whether s is bool notation as defined in YAML 1.1.
+//
+// We continue to force strings that YAML 1.1 would interpret as booleans to be
+// rendered as quotes strings so that the marshalled output valid for YAML 1.1
+// parsing.
+func isOldBool(s string) (result bool) {
+	switch s {
+	case "y", "Y", "yes", "Yes", "YES", "on", "On", "ON",
+		"n", "N", "no", "No", "NO", "off", "Off", "OFF":
+		return true
+	default:
+		return false
+	}
+}
+
+func (e *encoder) stringv(tag string, in reflect.Value) {
+	var style yaml_scalar_style_t
+	s := in.String()
+	canUsePlain := true
+	switch {
+	case !utf8.ValidString(s):
+		if tag == binaryTag {
+			failf("explicitly tagged !!binary data must be base64-encoded")
+		}
+		if tag != "" {
+			failf("cannot marshal invalid UTF-8 data as %s", shortTag(tag))
+		}
+		// It can't be encoded directly as YAML so use a binary tag
+		// and encode it as base64.
+		tag = binaryTag
+		s = encodeBase64(s)
+	case tag == "":
+		// Check to see if it would resolve to a specific
+		// tag when encoded unquoted. If it doesn't,
+		// there's no need to quote it.
+		rtag, _ := resolve("", s)
+		canUsePlain = rtag == strTag && !(isBase60Float(s) || isOldBool(s))
+	}
+	// Note: it's possible for user code to emit invalid YAML
+	// if they explicitly specify a tag and a string containing
+	// text that's incompatible with that tag.
+	switch {
+	case strings.Contains(s, "\n"):
+		if e.flow {
+			style = yaml_DOUBLE_QUOTED_SCALAR_STYLE
+		} else {
+			style = yaml_LITERAL_SCALAR_STYLE
+		}
+	case canUsePlain:
+		style = yaml_PLAIN_SCALAR_STYLE
+	default:
+		style = yaml_DOUBLE_QUOTED_SCALAR_STYLE
+	}
+	e.emitScalar(s, "", tag, style, nil, nil, nil, nil)
+}
+
+func (e *encoder) boolv(tag string, in reflect.Value) {
+	var s string
+	if in.Bool() {
+		s = "true"
+	} else {
+		s = "false"
+	}
+	e.emitScalar(s, "", tag, yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) intv(tag string, in reflect.Value) {
+	s := strconv.FormatInt(in.Int(), 10)
+	e.emitScalar(s, "", tag, yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) uintv(tag string, in reflect.Value) {
+	s := strconv.FormatUint(in.Uint(), 10)
+	e.emitScalar(s, "", tag, yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) timev(tag string, in reflect.Value) {
+	t := in.Interface().(time.Time)
+	s := t.Format(time.RFC3339Nano)
+	e.emitScalar(s, "", tag, yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) floatv(tag string, in reflect.Value) {
+	// Issue #352: When formatting, use the precision of the underlying value
+	precision := 64
+	if in.Kind() == reflect.Float32 {
+		precision = 32
+	}
+
+	s := strconv.FormatFloat(in.Float(), 'g', -1, precision)
+	switch s {
+	case "+Inf":
+		s = ".inf"
+	case "-Inf":
+		s = "-.inf"
+	case "NaN":
+		s = ".nan"
+	}
+	e.emitScalar(s, "", tag, yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) nilv() {
+	e.emitScalar("null", "", "", yaml_PLAIN_SCALAR_STYLE, nil, nil, nil, nil)
+}
+
+func (e *encoder) emitScalar(value, anchor, tag string, style yaml_scalar_style_t, head, line, foot, tail []byte) {
+	// TODO Kill this function. Replace all initialize calls by their underlining Go literals.
+	implicit := tag == ""
+	if !implicit {
+		tag = longTag(tag)
+	}
+	e.must(yaml_scalar_event_initialize(&e.event, []byte(anchor), []byte(tag), []byte(value), implicit, implicit, style))
+	e.event.head_comment = head
+	e.event.line_comment = line
+	e.event.foot_comment = foot
+	e.event.tail_comment = tail
+	e.emit()
+}
+
+func (e *encoder) nodev(in reflect.Value) {
+	e.node(in.Interface().(*Node), "")
+}
+
+func (e *encoder) node(node *Node, tail string) {
+	// Zero nodes behave as nil.
+	if node.Kind == 0 && node.IsZero() {
+		e.nilv()
+		return
+	}
+
+	// If the tag was not explicitly requested, and dropping it won't change the
+	// implicit tag of the value, don't include it in the presentation.
+	var tag = node.Tag
+	var stag = shortTag(tag)
+	var forceQuoting bool
+	if tag != "" && node.Style&TaggedStyle == 0 {
+		if node.Kind == ScalarNode {
+			if stag == strTag && node.Style&(SingleQuotedStyle|DoubleQuotedStyle|LiteralStyle|FoldedStyle) != 0 {
+				tag = ""
+			} else {
+				rtag, _ := resolve("", node.Value)
+				if rtag == stag {
+					tag = ""
+				} else if stag == strTag {
+					tag = ""
+					forceQuoting = true
+				}
+			}
+		} else {
+			var rtag string
+			switch node.Kind {
+			case MappingNode:
+				rtag = mapTag
+			case SequenceNode:
+				rtag = seqTag
+			}
+			if rtag == stag {
+				tag = ""
+			}
+		}
+	}
+
+	switch node.Kind {
+	case DocumentNode:
+		yaml_document_start_event_initialize(&e.event, nil, nil, true)
+		e.event.head_comment = []byte(node.HeadComment)
+		e.emit()
+		for _, node := range node.Content {
+			e.node(node, "")
+		}
+		yaml_document_end_event_initialize(&e.event, true)
+		e.event.foot_comment = []byte(node.FootComment)
+		e.emit()
+
+	case SequenceNode:
+		style := yaml_BLOCK_SEQUENCE_STYLE
+		if node.Style&FlowStyle != 0 {
+			style = yaml_FLOW_SEQUENCE_STYLE
+		}
+		e.must(yaml_sequence_start_event_initialize(&e.event, []byte(node.Anchor), []byte(longTag(tag)), tag == "", style))
+		e.event.head_comment = []byte(node.HeadComment)
+		e.emit()
+		for _, node := range node.Content {
+			e.node(node, "")
+		}
+		e.must(yaml_sequence_end_event_initialize(&e.event))
+		e.event.line_comment = []byte(node.LineComment)
+		e.event.foot_comment = []byte(node.FootComment)
+		e.emit()
+
+	case MappingNode:
+		style := yaml_BLOCK_MAPPING_STYLE
+		if node.Style&FlowStyle != 0 {
+			style = yaml_FLOW_MAPPING_STYLE
+		}
+		yaml_mapping_start_event_initialize(&e.event, []byte(node.Anchor), []byte(longTag(tag)), tag == "", style)
+		e.event.tail_comment = []byte(tail)
+		e.event.head_comment = []byte(node.HeadComment)
+		e.emit()
+
+		// The tail logic below moves the foot comment of prior keys to the following key,
+		// since the value for each key may be a nested structure and the foot needs to be
+		// processed only the entirety of the value is streamed. The last tail is processed
+		// with the mapping end event.
+		var tail string
+		for i := 0; i+1 < len(node.Content); i += 2 {
+			k := node.Content[i]
+			foot := k.FootComment
+			if foot != "" {
+				kopy := *k
+				kopy.FootComment = ""
+				k = &kopy
+			}
+			e.node(k, tail)
+			tail = foot
+
+			v := node.Content[i+1]
+			e.node(v, "")
+		}
+
+		yaml_mapping_end_event_initialize(&e.event)
+		e.event.tail_comment = []byte(tail)
+		e.event.line_comment = []byte(node.LineComment)
+		e.event.foot_comment = []byte(node.FootComment)
+		e.emit()
+
+	case AliasNode:
+		yaml_alias_event_initialize(&e.event, []byte(node.Value))
+		e.event.head_comment = []byte(node.HeadComment)
+		e.event.line_comment = []byte(node.LineComment)
+		e.event.foot_comment = []byte(node.FootComment)
+		e.emit()
+
+	case ScalarNode:
+		value := node.Value
+		if !utf8.ValidString(value) {
+			if stag == binaryTag {
+				failf("explicitly tagged !!binary data must be base64-encoded")
+			}
+			if stag != "" {
+				failf("cannot marshal invalid UTF-8 data as %s", stag)
+			}
+			// It can't be encoded directly as YAML so use a binary tag
+			// and encode it as base64.
+			tag = binaryTag
+			value = encodeBase64(value)
+		}
+
+		style := yaml_PLAIN_SCALAR_STYLE
+		switch {
+		case node.Style&DoubleQuotedStyle != 0:
+			style = yaml_DOUBLE_QUOTED_SCALAR_STYLE
+		case node.Style&SingleQuotedStyle != 0:
+			style = yaml_SINGLE_QUOTED_SCALAR_STYLE
+		case node.Style&LiteralStyle != 0:
+			style = yaml_LITERAL_SCALAR_STYLE
+		case node.Style&FoldedStyle != 0:
+			style = yaml_FOLDED_SCALAR_STYLE
+		case strings.Contains(value, "\n"):
+			style = yaml_LITERAL_SCALAR_STYLE
+		case forceQuoting:
+			style = yaml_DOUBLE_QUOTED_SCALAR_STYLE
+		}
+
+		e.emitScalar(value, node.Anchor, tag, style, []byte(node.HeadComment), []byte(node.LineComment), []byte(node.FootComment), []byte(tail))
+	default:
+		failf("cannot encode node with unknown kind %d", node.Kind)
+	}
+}

cli/vendor/gopkg.in/yaml.v3/readerc.go

@@ -0,0 +1,434 @@
+// 
+// Copyright (c) 2011-2019 Canonical Ltd
+// Copyright (c) 2006-2010 Kirill Simonov
+// 
+// Permission is hereby granted, free of charge, to any person obtaining a copy of
+// this software and associated documentation files (the "Software"), to deal in
+// the Software without restriction, including without limitation the rights to
+// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+// of the Software, and to permit persons to whom the Software is furnished to do
+// so, subject to the following conditions:
+// 
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+package yaml
+
+import (
+	"io"
+)
+
+// Set the reader error and return 0.
+func yaml_parser_set_reader_error(parser *yaml_parser_t, problem string, offset int, value int) bool {
+	parser.error = yaml_READER_ERROR
+	parser.problem = problem
+	parser.problem_offset = offset
+	parser.problem_value = value
+	return false
+}
+
+// Byte order marks.
+const (
+	bom_UTF8    = "\xef\xbb\xbf"
+	bom_UTF16LE = "\xff\xfe"
+	bom_UTF16BE = "\xfe\xff"
+)
+
+// Determine the input stream encoding by checking the BOM symbol. If no BOM is
+// found, the UTF-8 encoding is assumed. Return 1 on success, 0 on failure.
+func yaml_parser_determine_encoding(parser *yaml_parser_t) bool {
+	// Ensure that we had enough bytes in the raw buffer.
+	for !parser.eof && len(parser.raw_buffer)-parser.raw_buffer_pos < 3 {
+		if !yaml_parser_update_raw_buffer(parser) {
+			return false
+		}
+	}
+
+	// Determine the encoding.
+	buf := parser.raw_buffer
+	pos := parser.raw_buffer_pos
+	avail := len(buf) - pos
+	if avail >= 2 && buf[pos] == bom_UTF16LE[0] && buf[pos+1] == bom_UTF16LE[1] {
+		parser.encoding = yaml_UTF16LE_ENCODING
+		parser.raw_buffer_pos += 2
+		parser.offset += 2
+	} else if avail >= 2 && buf[pos] == bom_UTF16BE[0] && buf[pos+1] == bom_UTF16BE[1] {
+		parser.encoding = yaml_UTF16BE_ENCODING
+		parser.raw_buffer_pos += 2
+		parser.offset += 2
+	} else if avail >= 3 && buf[pos] == bom_UTF8[0] && buf[pos+1] == bom_UTF8[1] && buf[pos+2] == bom_UTF8[2] {
+		parser.encoding = yaml_UTF8_ENCODING
+		parser.raw_buffer_pos += 3
+		parser.offset += 3
+	} else {
+		parser.encoding = yaml_UTF8_ENCODING
+	}
+	return true
+}
+
+// Update the raw buffer.
+func yaml_parser_update_raw_buffer(parser *yaml_parser_t) bool {
+	size_read := 0
+
+	// Return if the raw buffer is full.
+	if parser.raw_buffer_pos == 0 && len(parser.raw_buffer) == cap(parser.raw_buffer) {
+		return true
+	}
+
+	// Return on EOF.
+	if parser.eof {
+		return true
+	}
+
+	// Move the remaining bytes in the raw buffer to the beginning.
+	if parser.raw_buffer_pos > 0 && parser.raw_buffer_pos < len(parser.raw_buffer) {
+		copy(parser.raw_buffer, parser.raw_buffer[parser.raw_buffer_pos:])
+	}
+	parser.raw_buffer = parser.raw_buffer[:len(parser.raw_buffer)-parser.raw_buffer_pos]
+	parser.raw_buffer_pos = 0
+
+	// Call the read handler to fill the buffer.
+	size_read, err := parser.read_handler(parser, parser.raw_buffer[len(parser.raw_buffer):cap(parser.raw_buffer)])
+	parser.raw_buffer = parser.raw_buffer[:len(parser.raw_buffer)+size_read]
+	if err == io.EOF {
+		parser.eof = true
+	} else if err != nil {
+		return yaml_parser_set_reader_error(parser, "input error: "+err.Error(), parser.offset, -1)
+	}
+	return true
+}
+
+// Ensure that the buffer contains at least `length` characters.
+// Return true on success, false on failure.
+//
+// The length is supposed to be significantly less that the buffer size.
+func yaml_parser_update_buffer(parser *yaml_parser_t, length int) bool {
+	if parser.read_handler == nil {
+		panic("read handler must be set")
+	}
+
+	// [Go] This function was changed to guarantee the requested length size at EOF.
+	// The fact we need to do this is pretty awful, but the description above implies
+	// for that to be the case, and there are tests
+
+	// If the EOF flag is set and the raw buffer is empty, do nothing.
+	if parser.eof && parser.raw_buffer_pos == len(parser.raw_buffer) {
+		// [Go] ACTUALLY! Read the documentation of this function above.
+		// This is just broken. To return true, we need to have the
+		// given length in the buffer. Not doing that means every single
+		// check that calls this function to make sure the buffer has a
+		// given length is Go) panicking; or C) accessing invalid memory.
+		//return true
+	}
+
+	// Return if the buffer contains enough characters.
+	if parser.unread >= length {
+		return true
+	}
+
+	// Determine the input encoding if it is not known yet.
+	if parser.encoding == yaml_ANY_ENCODING {
+		if !yaml_parser_determine_encoding(parser) {
+			return false
+		}
+	}
+
+	// Move the unread characters to the beginning of the buffer.
+	buffer_len := len(parser.buffer)
+	if parser.buffer_pos > 0 && parser.buffer_pos < buffer_len {
+		copy(parser.buffer, parser.buffer[parser.buffer_pos:])
+		buffer_len -= parser.buffer_pos
+		parser.buffer_pos = 0
+	} else if parser.buffer_pos == buffer_len {
+		buffer_len = 0
+		parser.buffer_pos = 0
+	}
+
+	// Open the whole buffer for writing, and cut it before returning.
+	parser.buffer = parser.buffer[:cap(parser.buffer)]
+
+	// Fill the buffer until it has enough characters.
+	first := true
+	for parser.unread < length {
+
+		// Fill the raw buffer if necessary.
+		if !first || parser.raw_buffer_pos == len(parser.raw_buffer) {
+			if !yaml_parser_update_raw_buffer(parser) {
+				parser.buffer = parser.buffer[:buffer_len]
+				return false
+			}
+		}
+		first = false
+
+		// Decode the raw buffer.
+	inner:
+		for parser.raw_buffer_pos != len(parser.raw_buffer) {
+			var value rune
+			var width int
+
+			raw_unread := len(parser.raw_buffer) - parser.raw_buffer_pos
+
+			// Decode the next character.
+			switch parser.encoding {
+			case yaml_UTF8_ENCODING:
+				// Decode a UTF-8 character.  Check RFC 3629
+				// (http://www.ietf.org/rfc/rfc3629.txt) for more details.
+				//
+				// The following table (taken from the RFC) is used for
+				// decoding.
+				//
+				//    Char. number range |        UTF-8 octet sequence
+				//      (hexadecimal)    |              (binary)
+				//   --------------------+------------------------------------
+				//   0000 0000-0000 007F | 0xxxxxxx
+				//   0000 0080-0000 07FF | 110xxxxx 10xxxxxx
+				//   0000 0800-0000 FFFF | 1110xxxx 10xxxxxx 10xxxxxx
+				//   0001 0000-0010 FFFF | 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
+				//
+				// Additionally, the characters in the range 0xD800-0xDFFF
+				// are prohibited as they are reserved for use with UTF-16
+				// surrogate pairs.
+
+				// Determine the length of the UTF-8 sequence.
+				octet := parser.raw_buffer[parser.raw_buffer_pos]
+				switch {
+				case octet&0x80 == 0x00:
+					width = 1
+				case octet&0xE0 == 0xC0:
+					width = 2
+				case octet&0xF0 == 0xE0:
+					width = 3
+				case octet&0xF8 == 0xF0:
+					width = 4
+				default:
+					// The leading octet is invalid.
+					return yaml_parser_set_reader_error(parser,
+						"invalid leading UTF-8 octet",
+						parser.offset, int(octet))
+				}
+
+				// Check if the raw buffer contains an incomplete character.
+				if width > raw_unread {
+					if parser.eof {
+						return yaml_parser_set_reader_error(parser,
+							"incomplete UTF-8 octet sequence",
+							parser.offset, -1)
+					}
+					break inner
+				}
+
+				// Decode the leading octet.
+				switch {
+				case octet&0x80 == 0x00:
+					value = rune(octet & 0x7F)
+				case octet&0xE0 == 0xC0:
+					value = rune(octet & 0x1F)
+				case octet&0xF0 == 0xE0:
+					value = rune(octet & 0x0F)
+				case octet&0xF8 == 0xF0:
+					value = rune(octet & 0x07)
+				default:
+					value = 0
+				}
+
+				// Check and decode the trailing octets.
+				for k := 1; k < width; k++ {
+					octet = parser.raw_buffer[parser.raw_buffer_pos+k]
+
+					// Check if the octet is valid.
+					if (octet & 0xC0) != 0x80 {
+						return yaml_parser_set_reader_error(parser,
+							"invalid trailing UTF-8 octet",
+							parser.offset+k, int(octet))
+					}
+
+					// Decode the octet.
+					value = (value << 6) + rune(octet&0x3F)
+				}
+
+				// Check the length of the sequence against the value.
+				switch {
+				case width == 1:
+				case width == 2 && value >= 0x80:
+				case width == 3 && value >= 0x800:
+				case width == 4 && value >= 0x10000:
+				default:
+					return yaml_parser_set_reader_error(parser,
+						"invalid length of a UTF-8 sequence",
+						parser.offset, -1)
+				}
+
+				// Check the range of the value.
+				if value >= 0xD800 && value <= 0xDFFF || value > 0x10FFFF {
+					return yaml_parser_set_reader_error(parser,
+						"invalid Unicode character",
+						parser.offset, int(value))
+				}
+
+			case yaml_UTF16LE_ENCODING, yaml_UTF16BE_ENCODING:
+				var low, high int
+				if parser.encoding == yaml_UTF16LE_ENCODING {
+					low, high = 0, 1
+				} else {
+					low, high = 1, 0
+				}
+
+				// The UTF-16 encoding is not as simple as one might
+				// naively think.  Check RFC 2781
+				// (http://www.ietf.org/rfc/rfc2781.txt).
+				//
+				// Normally, two subsequent bytes describe a Unicode
+				// character.  However a special technique (called a
+				// surrogate pair) is used for specifying character
+				// values larger than 0xFFFF.
+				//
+				// A surrogate pair consists of two pseudo-characters:
+				//      high surrogate area (0xD800-0xDBFF)
+				//      low surrogate area (0xDC00-0xDFFF)
+				//
+				// The following formulas are used for decoding
+				// and encoding characters using surrogate pairs:
+				//
+				//  U  = U' + 0x10000   (0x01 00 00 <= U <= 0x10 FF FF)
+				//  U' = yyyyyyyyyyxxxxxxxxxx   (0 <= U' <= 0x0F FF FF)
+				//  W1 = 110110yyyyyyyyyy
+				//  W2 = 110111xxxxxxxxxx
+				//
+				// where U is the character value, W1 is the high surrogate
+				// area, W2 is the low surrogate area.
+
+				// Check for incomplete UTF-16 character.
+				if raw_unread < 2 {
+					if parser.eof {
+						return yaml_parser_set_reader_error(parser,
+							"incomplete UTF-16 character",
+							parser.offset, -1)
+					}
+					break inner
+				}
+
+				// Get the character.
+				value = rune(parser.raw_buffer[parser.raw_buffer_pos+low]) +
+					(rune(parser.raw_buffer[parser.raw_buffer_pos+high]) << 8)
+
+				// Check for unexpected low surrogate area.
+				if value&0xFC00 == 0xDC00 {
+					return yaml_parser_set_reader_error(parser,
+						"unexpected low surrogate area",
+						parser.offset, int(value))
+				}
+
+				// Check for a high surrogate area.
+				if value&0xFC00 == 0xD800 {
+					width = 4
+
+					// Check for incomplete surrogate pair.
+					if raw_unread < 4 {
+						if parser.eof {
+							return yaml_parser_set_reader_error(parser,
+								"incomplete UTF-16 surrogate pair",
+								parser.offset, -1)
+						}
+						break inner
+					}
+
+					// Get the next character.
+					value2 := rune(parser.raw_buffer[parser.raw_buffer_pos+low+2]) +
+						(rune(parser.raw_buffer[parser.raw_buffer_pos+high+2]) << 8)
+
+					// Check for a low surrogate area.
+					if value2&0xFC00 != 0xDC00 {
+						return yaml_parser_set_reader_error(parser,
+							"expected low surrogate area",
+							parser.offset+2, int(value2))
+					}
+
+					// Generate the value of the surrogate pair.
+					value = 0x10000 + ((value & 0x3FF) << 10) + (value2 & 0x3FF)
+				} else {
+					width = 2
+				}
+
+			default:
+				panic("impossible")
+			}
+
+			// Check if the character is in the allowed range:
+			//      #x9 | #xA | #xD | [#x20-#x7E]               (8 bit)
+			//      | #x85 | [#xA0-#xD7FF] | [#xE000-#xFFFD]    (16 bit)
+			//      | [#x10000-#x10FFFF]                        (32 bit)
+			switch {
+			case value == 0x09:
+			case value == 0x0A:
+			case value == 0x0D:
+			case value >= 0x20 && value <= 0x7E:
+			case value == 0x85:
+			case value >= 0xA0 && value <= 0xD7FF:
+			case value >= 0xE000 && value <= 0xFFFD:
+			case value >= 0x10000 && value <= 0x10FFFF:
+			default:
+				return yaml_parser_set_reader_error(parser,
+					"control characters are not allowed",
+					parser.offset, int(value))
+			}
+
+			// Move the raw pointers.
+			parser.raw_buffer_pos += width
+			parser.offset += width
+
+			// Finally put the character into the buffer.
+			if value <= 0x7F {
+				// 0000 0000-0000 007F . 0xxxxxxx
+				parser.buffer[buffer_len+0] = byte(value)
+				buffer_len += 1
+			} else if value <= 0x7FF {
+				// 0000 0080-0000 07FF . 110xxxxx 10xxxxxx
+				parser.buffer[buffer_len+0] = byte(0xC0 + (value >> 6))
+				parser.buffer[buffer_len+1] = byte(0x80 + (value & 0x3F))
+				buffer_len += 2
+			} else if value <= 0xFFFF {
+				// 0000 0800-0000 FFFF . 1110xxxx 10xxxxxx 10xxxxxx
+				parser.buffer[buffer_len+0] = byte(0xE0 + (value >> 12))
+				parser.buffer[buffer_len+1] = byte(0x80 + ((value >> 6) & 0x3F))
+				parser.buffer[buffer_len+2] = byte(0x80 + (value & 0x3F))
+				buffer_len += 3
+			} else {
+				// 0001 0000-0010 FFFF . 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
+				parser.buffer[buffer_len+0] = byte(0xF0 + (value >> 18))
+				parser.buffer[buffer_len+1] = byte(0x80 + ((value >> 12) & 0x3F))
+				parser.buffer[buffer_len+2] = byte(0x80 + ((value >> 6) & 0x3F))
+				parser.buffer[buffer_len+3] = byte(0x80 + (value & 0x3F))
+				buffer_len += 4
+			}
+
+			parser.unread++
+		}
+
+		// On EOF, put NUL into the buffer and return.
+		if parser.eof {
+			parser.buffer[buffer_len] = 0
+			buffer_len++
+			parser.unread++
+			break
+		}
+	}
+	// [Go] Read the documentation of this function above. To return true,
+	// we need to have the given length in the buffer. Not doing that means
+	// every single check that calls this function to make sure the buffer
+	// has a given length is Go) panicking; or C) accessing invalid memory.
+	// This happens here due to the EOF above breaking early.
+	for buffer_len < length {
+		parser.buffer[buffer_len] = 0
+		buffer_len++
+	}
+	parser.buffer = parser.buffer[:buffer_len]
+	return true
+}

cli/vendor/gopkg.in/yaml.v3/resolve.go

@@ -0,0 +1,326 @@
+//
+// Copyright (c) 2011-2019 Canonical Ltd
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package yaml
+
+import (
+	"encoding/base64"
+	"math"
+	"regexp"
+	"strconv"
+	"strings"
+	"time"
+)
+
+type resolveMapItem struct {
+	value interface{}
+	tag   string
+}
+
+var resolveTable = make([]byte, 256)
+var resolveMap = make(map[string]resolveMapItem)
+
+func init() {
+	t := resolveTable
+	t[int('+')] = 'S' // Sign
+	t[int('-')] = 'S'
+	for _, c := range "0123456789" {
+		t[int(c)] = 'D' // Digit
+	}
+	for _, c := range "yYnNtTfFoO~" {
+		t[int(c)] = 'M' // In map
+	}
+	t[int('.')] = '.' // Float (potentially in map)
+
+	var resolveMapList = []struct {
+		v   interface{}
+		tag string
+		l   []string
+	}{
+		{true, boolTag, []string{"true", "True", "TRUE"}},
+		{false, boolTag, []string{"false", "False", "FALSE"}},
+		{nil, nullTag, []string{"", "~", "null", "Null", "NULL"}},
+		{math.NaN(), floatTag, []string{".nan", ".NaN", ".NAN"}},
+		{math.Inf(+1), floatTag, []string{".inf", ".Inf", ".INF"}},
+		{math.Inf(+1), floatTag, []string{"+.inf", "+.Inf", "+.INF"}},
+		{math.Inf(-1), floatTag, []string{"-.inf", "-.Inf", "-.INF"}},
+		{"<<", mergeTag, []string{"<<"}},
+	}
+
+	m := resolveMap
+	for _, item := range resolveMapList {
+		for _, s := range item.l {
+			m[s] = resolveMapItem{item.v, item.tag}
+		}
+	}
+}
+
+const (
+	nullTag      = "!!null"
+	boolTag      = "!!bool"
+	strTag       = "!!str"
+	intTag       = "!!int"
+	floatTag     = "!!float"
+	timestampTag = "!!timestamp"
+	seqTag       = "!!seq"
+	mapTag       = "!!map"
+	binaryTag    = "!!binary"
+	mergeTag     = "!!merge"
+)
+
+var longTags = make(map[string]string)
+var shortTags = make(map[string]string)
+
+func init() {
+	for _, stag := range []string{nullTag, boolTag, strTag, intTag, floatTag, timestampTag, seqTag, mapTag, binaryTag, mergeTag} {
+		ltag := longTag(stag)
+		longTags[stag] = ltag
+		shortTags[ltag] = stag
+	}
+}
+
+const longTagPrefix = "tag:yaml.org,2002:"
+
+func shortTag(tag string) string {
+	if strings.HasPrefix(tag, longTagPrefix) {
+		if stag, ok := shortTags[tag]; ok {
+			return stag
+		}
+		return "!!" + tag[len(longTagPrefix):]
+	}
+	return tag
+}
+
+func longTag(tag string) string {
+	if strings.HasPrefix(tag, "!!") {
+		if ltag, ok := longTags[tag]; ok {
+			return ltag
+		}
+		return longTagPrefix + tag[2:]
+	}
+	return tag
+}
+
+func resolvableTag(tag string) bool {
+	switch tag {
+	case "", strTag, boolTag, intTag, floatTag, nullTag, timestampTag:
+		return true
+	}
+	return false
+}
+
+var yamlStyleFloat = regexp.MustCompile(`^[-+]?(\.[0-9]+|[0-9]+(\.[0-9]*)?)([eE][-+]?[0-9]+)?$`)
+
+func resolve(tag string, in string) (rtag string, out interface{}) {
+	tag = shortTag(tag)
+	if !resolvableTag(tag) {
+		return tag, in
+	}
+
+	defer func() {
+		switch tag {
+		case "", rtag, strTag, binaryTag:
+			return
+		case floatTag:
+			if rtag == intTag {
+				switch v := out.(type) {
+				case int64:
+					rtag = floatTag
+					out = float64(v)
+					return
+				case int:
+					rtag = floatTag
+					out = float64(v)
+					return
+				}
+			}
+		}
+		failf("cannot decode %s `%s` as a %s", shortTag(rtag), in, shortTag(tag))
+	}()
+
+	// Any data is accepted as a !!str or !!binary.
+	// Otherwise, the prefix is enough of a hint about what it might be.
+	hint := byte('N')
+	if in != "" {
+		hint = resolveTable[in[0]]
+	}
+	if hint != 0 && tag != strTag && tag != binaryTag {
+		// Handle things we can lookup in a map.
+		if item, ok := resolveMap[in]; ok {
+			return item.tag, item.value
+		}
+
+		// Base 60 floats are a bad idea, were dropped in YAML 1.2, and
+		// are purposefully unsupported here. They're still quoted on
+		// the way out for compatibility with other parser, though.
+
+		switch hint {
+		case 'M':
+			// We've already checked the map above.
+
+		case '.':
+			// Not in the map, so maybe a normal float.
+			floatv, err := strconv.ParseFloat(in, 64)
+			if err == nil {
+				return floatTag, floatv
+			}
+
+		case 'D', 'S':
+			// Int, float, or timestamp.
+			// Only try values as a timestamp if the value is unquoted or there's an explicit
+			// !!timestamp tag.
+			if tag == "" || tag == timestampTag {
+				t, ok := parseTimestamp(in)
+				if ok {
+					return timestampTag, t
+				}
+			}
+
+			plain := strings.Replace(in, "_", "", -1)
+			intv, err := strconv.ParseInt(plain, 0, 64)
+			if err == nil {
+				if intv == int64(int(intv)) {
+					return intTag, int(intv)
+				} else {
+					return intTag, intv
+				}
+			}
+			uintv, err := strconv.ParseUint(plain, 0, 64)
+			if err == nil {
+				return intTag, uintv
+			}
+			if yamlStyleFloat.MatchString(plain) {
+				floatv, err := strconv.ParseFloat(plain, 64)
+				if err == nil {
+					return floatTag, floatv
+				}
+			}
+			if strings.HasPrefix(plain, "0b") {
+				intv, err := strconv.ParseInt(plain[2:], 2, 64)
+				if err == nil {
+					if intv == int64(int(intv)) {
+						return intTag, int(intv)
+					} else {
+						return intTag, intv
+					}
+				}
+				uintv, err := strconv.ParseUint(plain[2:], 2, 64)
+				if err == nil {
+					return intTag, uintv
+				}
+			} else if strings.HasPrefix(plain, "-0b") {
+				intv, err := strconv.ParseInt("-"+plain[3:], 2, 64)
+				if err == nil {
+					if true || intv == int64(int(intv)) {
+						return intTag, int(intv)
+					} else {
+						return intTag, intv
+					}
+				}
+			}
+			// Octals as introduced in version 1.2 of the spec.
+			// Octals from the 1.1 spec, spelled as 0777, are still
+			// decoded by default in v3 as well for compatibility.
+			// May be dropped in v4 depending on how usage evolves.
+			if strings.HasPrefix(plain, "0o") {
+				intv, err := strconv.ParseInt(plain[2:], 8, 64)
+				if err == nil {
+					if intv == int64(int(intv)) {
+						return intTag, int(intv)
+					} else {
+						return intTag, intv
+					}
+				}
+				uintv, err := strconv.ParseUint(plain[2:], 8, 64)
+				if err == nil {
+					return intTag, uintv
+				}
+			} else if strings.HasPrefix(plain, "-0o") {
+				intv, err := strconv.ParseInt("-"+plain[3:], 8, 64)
+				if err == nil {
+					if true || intv == int64(int(intv)) {
+						return intTag, int(intv)
+					} else {
+						return intTag, intv
+					}
+				}
+			}
+		default:
+			panic("internal error: missing handler for resolver table: " + string(rune(hint)) + " (with " + in + ")")
+		}
+	}
+	return strTag, in
+}
+
+// encodeBase64 encodes s as base64 that is broken up into multiple lines
+// as appropriate for the resulting length.
+func encodeBase64(s string) string {
+	const lineLen = 70
+	encLen := base64.StdEncoding.EncodedLen(len(s))
+	lines := encLen/lineLen + 1
+	buf := make([]byte, encLen*2+lines)
+	in := buf[0:encLen]
+	out := buf[encLen:]
+	base64.StdEncoding.Encode(in, []byte(s))
+	k := 0
+	for i := 0; i < len(in); i += lineLen {
+		j := i + lineLen
+		if j > len(in) {
+			j = len(in)
+		}
+		k += copy(out[k:], in[i:j])
+		if lines > 1 {
+			out[k] = '\n'
+			k++
+		}
+	}
+	return string(out[:k])
+}
+
+// This is a subset of the formats allowed by the regular expression
+// defined at http://yaml.org/type/timestamp.html.
+var allowedTimestampFormats = []string{
+	"2006-1-2T15:4:5.999999999Z07:00", // RCF3339Nano with short date fields.
+	"2006-1-2t15:4:5.999999999Z07:00", // RFC3339Nano with short date fields and lower-case "t".
+	"2006-1-2 15:4:5.999999999",       // space separated with no time zone
+	"2006-1-2",                        // date only
+	// Notable exception: time.Parse cannot handle: "2001-12-14 21:59:43.10 -5"
+	// from the set of examples.
+}
+
+// parseTimestamp parses s as a timestamp string and
+// returns the timestamp and reports whether it succeeded.
+// Timestamp formats are defined at http://yaml.org/type/timestamp.html
+func parseTimestamp(s string) (time.Time, bool) {
+	// TODO write code to check all the formats supported by
+	// http://yaml.org/type/timestamp.html instead of using time.Parse.
+
+	// Quick check: all date formats start with YYYY-.
+	i := 0
+	for ; i < len(s); i++ {
+		if c := s[i]; c < '0' || c > '9' {
+			break
+		}
+	}
+	if i != 4 || i == len(s) || s[i] != '-' {
+		return time.Time{}, false
+	}
+	for _, format := range allowedTimestampFormats {
+		if t, err := time.Parse(format, s); err == nil {
+			return t, true
+		}
+	}
+	return time.Time{}, false
+}

cli/vendor/gopkg.in/yaml.v3/sorter.go

@@ -0,0 +1,134 @@
+//
+// Copyright (c) 2011-2019 Canonical Ltd
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package yaml
+
+import (
+	"reflect"
+	"unicode"
+)
+
+type keyList []reflect.Value
+
+func (l keyList) Len() int      { return len(l) }
+func (l keyList) Swap(i, j int) { l[i], l[j] = l[j], l[i] }
+func (l keyList) Less(i, j int) bool {
+	a := l[i]
+	b := l[j]
+	ak := a.Kind()
+	bk := b.Kind()
+	for (ak == reflect.Interface || ak == reflect.Ptr) && !a.IsNil() {
+		a = a.Elem()
+		ak = a.Kind()
+	}
+	for (bk == reflect.Interface || bk == reflect.Ptr) && !b.IsNil() {
+		b = b.Elem()
+		bk = b.Kind()
+	}
+	af, aok := keyFloat(a)
+	bf, bok := keyFloat(b)
+	if aok && bok {
+		if af != bf {
+			return af < bf
+		}
+		if ak != bk {
+			return ak < bk
+		}
+		return numLess(a, b)
+	}
+	if ak != reflect.String || bk != reflect.String {
+		return ak < bk
+	}
+	ar, br := []rune(a.String()), []rune(b.String())
+	digits := false
+	for i := 0; i < len(ar) && i < len(br); i++ {
+		if ar[i] == br[i] {
+			digits = unicode.IsDigit(ar[i])
+			continue
+		}
+		al := unicode.IsLetter(ar[i])
+		bl := unicode.IsLetter(br[i])
+		if al && bl {
+			return ar[i] < br[i]
+		}
+		if al || bl {
+			if digits {
+				return al
+			} else {
+				return bl
+			}
+		}
+		var ai, bi int
+		var an, bn int64
+		if ar[i] == '0' || br[i] == '0' {
+			for j := i - 1; j >= 0 && unicode.IsDigit(ar[j]); j-- {
+				if ar[j] != '0' {
+					an = 1
+					bn = 1
+					break
+				}
+			}
+		}
+		for ai = i; ai < len(ar) && unicode.IsDigit(ar[ai]); ai++ {
+			an = an*10 + int64(ar[ai]-'0')
+		}
+		for bi = i; bi < len(br) && unicode.IsDigit(br[bi]); bi++ {
+			bn = bn*10 + int64(br[bi]-'0')
+		}
+		if an != bn {
+			return an < bn
+		}
+		if ai != bi {
+			return ai < bi
+		}
+		return ar[i] < br[i]
+	}
+	return len(ar) < len(br)
+}
+
+// keyFloat returns a float value for v if it is a number/bool
+// and whether it is a number/bool or not.
+func keyFloat(v reflect.Value) (f float64, ok bool) {
+	switch v.Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return float64(v.Int()), true
+	case reflect.Float32, reflect.Float64:
+		return v.Float(), true
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return float64(v.Uint()), true
+	case reflect.Bool:
+		if v.Bool() {
+			return 1, true
+		}
+		return 0, true
+	}
+	return 0, false
+}
+
+// numLess returns whether a < b.
+// a and b must necessarily have the same kind.
+func numLess(a, b reflect.Value) bool {
+	switch a.Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return a.Int() < b.Int()
+	case reflect.Float32, reflect.Float64:
+		return a.Float() < b.Float()
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return a.Uint() < b.Uint()
+	case reflect.Bool:
+		return !a.Bool() && b.Bool()
+	}
+	panic("not a number")
+}

cli/vendor/gopkg.in/yaml.v3/writerc.go

@@ -0,0 +1,48 @@
+// 
+// Copyright (c) 2011-2019 Canonical Ltd
+// Copyright (c) 2006-2010 Kirill Simonov
+// 
+// Permission is hereby granted, free of charge, to any person obtaining a copy of
+// this software and associated documentation files (the "Software"), to deal in
+// the Software without restriction, including without limitation the rights to
+// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+// of the Software, and to permit persons to whom the Software is furnished to do
+// so, subject to the following conditions:
+// 
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+package yaml
+
+// Set the writer error and return false.
+func yaml_emitter_set_writer_error(emitter *yaml_emitter_t, problem string) bool {
+	emitter.error = yaml_WRITER_ERROR
+	emitter.problem = problem
+	return false
+}
+
+// Flush the output buffer.
+func yaml_emitter_flush(emitter *yaml_emitter_t) bool {
+	if emitter.write_handler == nil {
+		panic("write handler not set")
+	}
+
+	// Check if the buffer is empty.
+	if emitter.buffer_pos == 0 {
+		return true
+	}
+
+	if err := emitter.write_handler(emitter, emitter.buffer[:emitter.buffer_pos]); err != nil {
+		return yaml_emitter_set_writer_error(emitter, "write error: "+err.Error())
+	}
+	emitter.buffer_pos = 0
+	return true
+}

cli/vendor/gopkg.in/yaml.v3/yaml.go

@@ -0,0 +1,698 @@
+//
+// Copyright (c) 2011-2019 Canonical Ltd
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package yaml implements YAML support for the Go language.
+//
+// Source code and other details for the project are available at GitHub:
+//
+//   https://github.com/go-yaml/yaml
+//
+package yaml
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"reflect"
+	"strings"
+	"sync"
+	"unicode/utf8"
+)
+
+// The Unmarshaler interface may be implemented by types to customize their
+// behavior when being unmarshaled from a YAML document.
+type Unmarshaler interface {
+	UnmarshalYAML(value *Node) error
+}
+
+type obsoleteUnmarshaler interface {
+	UnmarshalYAML(unmarshal func(interface{}) error) error
+}
+
+// The Marshaler interface may be implemented by types to customize their
+// behavior when being marshaled into a YAML document. The returned value
+// is marshaled in place of the original value implementing Marshaler.
+//
+// If an error is returned by MarshalYAML, the marshaling procedure stops
+// and returns with the provided error.
+type Marshaler interface {
+	MarshalYAML() (interface{}, error)
+}
+
+// Unmarshal decodes the first document found within the in byte slice
+// and assigns decoded values into the out value.
+//
+// Maps and pointers (to a struct, string, int, etc) are accepted as out
+// values. If an internal pointer within a struct is not initialized,
+// the yaml package will initialize it if necessary for unmarshalling
+// the provided data. The out parameter must not be nil.
+//
+// The type of the decoded values should be compatible with the respective
+// values in out. If one or more values cannot be decoded due to a type
+// mismatches, decoding continues partially until the end of the YAML
+// content, and a *yaml.TypeError is returned with details for all
+// missed values.
+//
+// Struct fields are only unmarshalled if they are exported (have an
+// upper case first letter), and are unmarshalled using the field name
+// lowercased as the default key. Custom keys may be defined via the
+// "yaml" name in the field tag: the content preceding the first comma
+// is used as the key, and the following comma-separated options are
+// used to tweak the marshalling process (see Marshal).
+// Conflicting names result in a runtime error.
+//
+// For example:
+//
+//     type T struct {
+//         F int `yaml:"a,omitempty"`
+//         B int
+//     }
+//     var t T
+//     yaml.Unmarshal([]byte("a: 1\nb: 2"), &t)
+//
+// See the documentation of Marshal for the format of tags and a list of
+// supported tag options.
+//
+func Unmarshal(in []byte, out interface{}) (err error) {
+	return unmarshal(in, out, false)
+}
+
+// A Decoder reads and decodes YAML values from an input stream.
+type Decoder struct {
+	parser      *parser
+	knownFields bool
+}
+
+// NewDecoder returns a new decoder that reads from r.
+//
+// The decoder introduces its own buffering and may read
+// data from r beyond the YAML values requested.
+func NewDecoder(r io.Reader) *Decoder {
+	return &Decoder{
+		parser: newParserFromReader(r),
+	}
+}
+
+// KnownFields ensures that the keys in decoded mappings to
+// exist as fields in the struct being decoded into.
+func (dec *Decoder) KnownFields(enable bool) {
+	dec.knownFields = enable
+}
+
+// Decode reads the next YAML-encoded value from its input
+// and stores it in the value pointed to by v.
+//
+// See the documentation for Unmarshal for details about the
+// conversion of YAML into a Go value.
+func (dec *Decoder) Decode(v interface{}) (err error) {
+	d := newDecoder()
+	d.knownFields = dec.knownFields
+	defer handleErr(&err)
+	node := dec.parser.parse()
+	if node == nil {
+		return io.EOF
+	}
+	out := reflect.ValueOf(v)
+	if out.Kind() == reflect.Ptr && !out.IsNil() {
+		out = out.Elem()
+	}
+	d.unmarshal(node, out)
+	if len(d.terrors) > 0 {
+		return &TypeError{d.terrors}
+	}
+	return nil
+}
+
+// Decode decodes the node and stores its data into the value pointed to by v.
+//
+// See the documentation for Unmarshal for details about the
+// conversion of YAML into a Go value.
+func (n *Node) Decode(v interface{}) (err error) {
+	d := newDecoder()
+	defer handleErr(&err)
+	out := reflect.ValueOf(v)
+	if out.Kind() == reflect.Ptr && !out.IsNil() {
+		out = out.Elem()
+	}
+	d.unmarshal(n, out)
+	if len(d.terrors) > 0 {
+		return &TypeError{d.terrors}
+	}
+	return nil
+}
+
+func unmarshal(in []byte, out interface{}, strict bool) (err error) {
+	defer handleErr(&err)
+	d := newDecoder()
+	p := newParser(in)
+	defer p.destroy()
+	node := p.parse()
+	if node != nil {
+		v := reflect.ValueOf(out)
+		if v.Kind() == reflect.Ptr && !v.IsNil() {
+			v = v.Elem()
+		}
+		d.unmarshal(node, v)
+	}
+	if len(d.terrors) > 0 {
+		return &TypeError{d.terrors}
+	}
+	return nil
+}
+
+// Marshal serializes the value provided into a YAML document. The structure
+// of the generated document will reflect the structure of the value itself.
+// Maps and pointers (to struct, string, int, etc) are accepted as the in value.
+//
+// Struct fields are only marshalled if they are exported (have an upper case
+// first letter), and are marshalled using the field name lowercased as the
+// default key. Custom keys may be defined via the "yaml" name in the field
+// tag: the content preceding the first comma is used as the key, and the
+// following comma-separated options are used to tweak the marshalling process.
+// Conflicting names result in a runtime error.
+//
+// The field tag format accepted is:
+//
+//     `(...) yaml:"[<key>][,<flag1>[,<flag2>]]" (...)`
+//
+// The following flags are currently supported:
+//
+//     omitempty    Only include the field if it's not set to the zero
+//                  value for the type or to empty slices or maps.
+//                  Zero valued structs will be omitted if all their public
+//                  fields are zero, unless they implement an IsZero
+//                  method (see the IsZeroer interface type), in which
+//                  case the field will be excluded if IsZero returns true.
+//
+//     flow         Marshal using a flow style (useful for structs,
+//                  sequences and maps).
+//
+//     inline       Inline the field, which must be a struct or a map,
+//                  causing all of its fields or keys to be processed as if
+//                  they were part of the outer struct. For maps, keys must
+//                  not conflict with the yaml keys of other struct fields.
+//
+// In addition, if the key is "-", the field is ignored.
+//
+// For example:
+//
+//     type T struct {
+//         F int `yaml:"a,omitempty"`
+//         B int
+//     }
+//     yaml.Marshal(&T{B: 2}) // Returns "b: 2\n"
+//     yaml.Marshal(&T{F: 1}} // Returns "a: 1\nb: 0\n"
+//
+func Marshal(in interface{}) (out []byte, err error) {
+	defer handleErr(&err)
+	e := newEncoder()
+	defer e.destroy()
+	e.marshalDoc("", reflect.ValueOf(in))
+	e.finish()
+	out = e.out
+	return
+}
+
+// An Encoder writes YAML values to an output stream.
+type Encoder struct {
+	encoder *encoder
+}
+
+// NewEncoder returns a new encoder that writes to w.
+// The Encoder should be closed after use to flush all data
+// to w.
+func NewEncoder(w io.Writer) *Encoder {
+	return &Encoder{
+		encoder: newEncoderWithWriter(w),
+	}
+}
+
+// Encode writes the YAML encoding of v to the stream.
+// If multiple items are encoded to the stream, the
+// second and subsequent document will be preceded
+// with a "---" document separator, but the first will not.
+//
+// See the documentation for Marshal for details about the conversion of Go
+// values to YAML.
+func (e *Encoder) Encode(v interface{}) (err error) {
+	defer handleErr(&err)
+	e.encoder.marshalDoc("", reflect.ValueOf(v))
+	return nil
+}
+
+// Encode encodes value v and stores its representation in n.
+//
+// See the documentation for Marshal for details about the
+// conversion of Go values into YAML.
+func (n *Node) Encode(v interface{}) (err error) {
+	defer handleErr(&err)
+	e := newEncoder()
+	defer e.destroy()
+	e.marshalDoc("", reflect.ValueOf(v))
+	e.finish()
+	p := newParser(e.out)
+	p.textless = true
+	defer p.destroy()
+	doc := p.parse()
+	*n = *doc.Content[0]
+	return nil
+}
+
+// SetIndent changes the used indentation used when encoding.
+func (e *Encoder) SetIndent(spaces int) {
+	if spaces < 0 {
+		panic("yaml: cannot indent to a negative number of spaces")
+	}
+	e.encoder.indent = spaces
+}
+
+// Close closes the encoder by writing any remaining data.
+// It does not write a stream terminating string "...".
+func (e *Encoder) Close() (err error) {
+	defer handleErr(&err)
+	e.encoder.finish()
+	return nil
+}
+
+func handleErr(err *error) {
+	if v := recover(); v != nil {
+		if e, ok := v.(yamlError); ok {
+			*err = e.err
+		} else {
+			panic(v)
+		}
+	}
+}
+
+type yamlError struct {
+	err error
+}
+
+func fail(err error) {
+	panic(yamlError{err})
+}
+
+func failf(format string, args ...interface{}) {
+	panic(yamlError{fmt.Errorf("yaml: "+format, args...)})
+}
+
+// A TypeError is returned by Unmarshal when one or more fields in
+// the YAML document cannot be properly decoded into the requested
+// types. When this error is returned, the value is still
+// unmarshaled partially.
+type TypeError struct {
+	Errors []string
+}
+
+func (e *TypeError) Error() string {
+	return fmt.Sprintf("yaml: unmarshal errors:\n  %s", strings.Join(e.Errors, "\n  "))
+}
+
+type Kind uint32
+
+const (
+	DocumentNode Kind = 1 << iota
+	SequenceNode
+	MappingNode
+	ScalarNode
+	AliasNode
+)
+
+type Style uint32
+
+const (
+	TaggedStyle Style = 1 << iota
+	DoubleQuotedStyle
+	SingleQuotedStyle
+	LiteralStyle
+	FoldedStyle
+	FlowStyle
+)
+
+// Node represents an element in the YAML document hierarchy. While documents
+// are typically encoded and decoded into higher level types, such as structs
+// and maps, Node is an intermediate representation that allows detailed
+// control over the content being decoded or encoded.
+//
+// It's worth noting that although Node offers access into details such as
+// line numbers, colums, and comments, the content when re-encoded will not
+// have its original textual representation preserved. An effort is made to
+// render the data plesantly, and to preserve comments near the data they
+// describe, though.
+//
+// Values that make use of the Node type interact with the yaml package in the
+// same way any other type would do, by encoding and decoding yaml data
+// directly or indirectly into them.
+//
+// For example:
+//
+//     var person struct {
+//             Name    string
+//             Address yaml.Node
+//     }
+//     err := yaml.Unmarshal(data, &person)
+// 
+// Or by itself:
+//
+//     var person Node
+//     err := yaml.Unmarshal(data, &person)
+//
+type Node struct {
+	// Kind defines whether the node is a document, a mapping, a sequence,
+	// a scalar value, or an alias to another node. The specific data type of
+	// scalar nodes may be obtained via the ShortTag and LongTag methods.
+	Kind  Kind
+
+	// Style allows customizing the apperance of the node in the tree.
+	Style Style
+
+	// Tag holds the YAML tag defining the data type for the value.
+	// When decoding, this field will always be set to the resolved tag,
+	// even when it wasn't explicitly provided in the YAML content.
+	// When encoding, if this field is unset the value type will be
+	// implied from the node properties, and if it is set, it will only
+	// be serialized into the representation if TaggedStyle is used or
+	// the implicit tag diverges from the provided one.
+	Tag string
+
+	// Value holds the unescaped and unquoted represenation of the value.
+	Value string
+
+	// Anchor holds the anchor name for this node, which allows aliases to point to it.
+	Anchor string
+
+	// Alias holds the node that this alias points to. Only valid when Kind is AliasNode.
+	Alias *Node
+
+	// Content holds contained nodes for documents, mappings, and sequences.
+	Content []*Node
+
+	// HeadComment holds any comments in the lines preceding the node and
+	// not separated by an empty line.
+	HeadComment string
+
+	// LineComment holds any comments at the end of the line where the node is in.
+	LineComment string
+
+	// FootComment holds any comments following the node and before empty lines.
+	FootComment string
+
+	// Line and Column hold the node position in the decoded YAML text.
+	// These fields are not respected when encoding the node.
+	Line   int
+	Column int
+}
+
+// IsZero returns whether the node has all of its fields unset.
+func (n *Node) IsZero() bool {
+	return n.Kind == 0 && n.Style == 0 && n.Tag == "" && n.Value == "" && n.Anchor == "" && n.Alias == nil && n.Content == nil &&
+		n.HeadComment == "" && n.LineComment == "" && n.FootComment == "" && n.Line == 0 && n.Column == 0
+}
+
+
+// LongTag returns the long form of the tag that indicates the data type for
+// the node. If the Tag field isn't explicitly defined, one will be computed
+// based on the node properties.
+func (n *Node) LongTag() string {
+	return longTag(n.ShortTag())
+}
+
+// ShortTag returns the short form of the YAML tag that indicates data type for
+// the node. If the Tag field isn't explicitly defined, one will be computed
+// based on the node properties.
+func (n *Node) ShortTag() string {
+	if n.indicatedString() {
+		return strTag
+	}
+	if n.Tag == "" || n.Tag == "!" {
+		switch n.Kind {
+		case MappingNode:
+			return mapTag
+		case SequenceNode:
+			return seqTag
+		case AliasNode:
+			if n.Alias != nil {
+				return n.Alias.ShortTag()
+			}
+		case ScalarNode:
+			tag, _ := resolve("", n.Value)
+			return tag
+		case 0:
+			// Special case to make the zero value convenient.
+			if n.IsZero() {
+				return nullTag
+			}
+		}
+		return ""
+	}
+	return shortTag(n.Tag)
+}
+
+func (n *Node) indicatedString() bool {
+	return n.Kind == ScalarNode &&
+		(shortTag(n.Tag) == strTag ||
+			(n.Tag == "" || n.Tag == "!") && n.Style&(SingleQuotedStyle|DoubleQuotedStyle|LiteralStyle|FoldedStyle) != 0)
+}
+
+// SetString is a convenience function that sets the node to a string value
+// and defines its style in a pleasant way depending on its content.
+func (n *Node) SetString(s string) {
+	n.Kind = ScalarNode
+	if utf8.ValidString(s) {
+		n.Value = s
+		n.Tag = strTag
+	} else {
+		n.Value = encodeBase64(s)
+		n.Tag = binaryTag
+	}
+	if strings.Contains(n.Value, "\n") {
+		n.Style = LiteralStyle
+	}
+}
+
+// --------------------------------------------------------------------------
+// Maintain a mapping of keys to structure field indexes
+
+// The code in this section was copied from mgo/bson.
+
+// structInfo holds details for the serialization of fields of
+// a given struct.
+type structInfo struct {
+	FieldsMap  map[string]fieldInfo
+	FieldsList []fieldInfo
+
+	// InlineMap is the number of the field in the struct that
+	// contains an ,inline map, or -1 if there's none.
+	InlineMap int
+
+	// InlineUnmarshalers holds indexes to inlined fields that
+	// contain unmarshaler values.
+	InlineUnmarshalers [][]int
+}
+
+type fieldInfo struct {
+	Key       string
+	Num       int
+	OmitEmpty bool
+	Flow      bool
+	// Id holds the unique field identifier, so we can cheaply
+	// check for field duplicates without maintaining an extra map.
+	Id int
+
+	// Inline holds the field index if the field is part of an inlined struct.
+	Inline []int
+}
+
+var structMap = make(map[reflect.Type]*structInfo)
+var fieldMapMutex sync.RWMutex
+var unmarshalerType reflect.Type
+
+func init() {
+	var v Unmarshaler
+	unmarshalerType = reflect.ValueOf(&v).Elem().Type()
+}
+
+func getStructInfo(st reflect.Type) (*structInfo, error) {
+	fieldMapMutex.RLock()
+	sinfo, found := structMap[st]
+	fieldMapMutex.RUnlock()
+	if found {
+		return sinfo, nil
+	}
+
+	n := st.NumField()
+	fieldsMap := make(map[string]fieldInfo)
+	fieldsList := make([]fieldInfo, 0, n)
+	inlineMap := -1
+	inlineUnmarshalers := [][]int(nil)
+	for i := 0; i != n; i++ {
+		field := st.Field(i)
+		if field.PkgPath != "" && !field.Anonymous {
+			continue // Private field
+		}
+
+		info := fieldInfo{Num: i}
+
+		tag := field.Tag.Get("yaml")
+		if tag == "" && strings.Index(string(field.Tag), ":") < 0 {
+			tag = string(field.Tag)
+		}
+		if tag == "-" {
+			continue
+		}
+
+		inline := false
+		fields := strings.Split(tag, ",")
+		if len(fields) > 1 {
+			for _, flag := range fields[1:] {
+				switch flag {
+				case "omitempty":
+					info.OmitEmpty = true
+				case "flow":
+					info.Flow = true
+				case "inline":
+					inline = true
+				default:
+					return nil, errors.New(fmt.Sprintf("unsupported flag %q in tag %q of type %s", flag, tag, st))
+				}
+			}
+			tag = fields[0]
+		}
+
+		if inline {
+			switch field.Type.Kind() {
+			case reflect.Map:
+				if inlineMap >= 0 {
+					return nil, errors.New("multiple ,inline maps in struct " + st.String())
+				}
+				if field.Type.Key() != reflect.TypeOf("") {
+					return nil, errors.New("option ,inline needs a map with string keys in struct " + st.String())
+				}
+				inlineMap = info.Num
+			case reflect.Struct, reflect.Ptr:
+				ftype := field.Type
+				for ftype.Kind() == reflect.Ptr {
+					ftype = ftype.Elem()
+				}
+				if ftype.Kind() != reflect.Struct {
+					return nil, errors.New("option ,inline may only be used on a struct or map field")
+				}
+				if reflect.PtrTo(ftype).Implements(unmarshalerType) {
+					inlineUnmarshalers = append(inlineUnmarshalers, []int{i})
+				} else {
+					sinfo, err := getStructInfo(ftype)
+					if err != nil {
+						return nil, err
+					}
+					for _, index := range sinfo.InlineUnmarshalers {
+						inlineUnmarshalers = append(inlineUnmarshalers, append([]int{i}, index...))
+					}
+					for _, finfo := range sinfo.FieldsList {
+						if _, found := fieldsMap[finfo.Key]; found {
+							msg := "duplicated key '" + finfo.Key + "' in struct " + st.String()
+							return nil, errors.New(msg)
+						}
+						if finfo.Inline == nil {
+							finfo.Inline = []int{i, finfo.Num}
+						} else {
+							finfo.Inline = append([]int{i}, finfo.Inline...)
+						}
+						finfo.Id = len(fieldsList)
+						fieldsMap[finfo.Key] = finfo
+						fieldsList = append(fieldsList, finfo)
+					}
+				}
+			default:
+				return nil, errors.New("option ,inline may only be used on a struct or map field")
+			}
+			continue
+		}
+
+		if tag != "" {
+			info.Key = tag
+		} else {
+			info.Key = strings.ToLower(field.Name)
+		}
+
+		if _, found = fieldsMap[info.Key]; found {
+			msg := "duplicated key '" + info.Key + "' in struct " + st.String()
+			return nil, errors.New(msg)
+		}
+
+		info.Id = len(fieldsList)
+		fieldsList = append(fieldsList, info)
+		fieldsMap[info.Key] = info
+	}
+
+	sinfo = &structInfo{
+		FieldsMap:          fieldsMap,
+		FieldsList:         fieldsList,
+		InlineMap:          inlineMap,
+		InlineUnmarshalers: inlineUnmarshalers,
+	}
+
+	fieldMapMutex.Lock()
+	structMap[st] = sinfo
+	fieldMapMutex.Unlock()
+	return sinfo, nil
+}
+
+// IsZeroer is used to check whether an object is zero to
+// determine whether it should be omitted when marshaling
+// with the omitempty flag. One notable implementation
+// is time.Time.
+type IsZeroer interface {
+	IsZero() bool
+}
+
+func isZero(v reflect.Value) bool {
+	kind := v.Kind()
+	if z, ok := v.Interface().(IsZeroer); ok {
+		if (kind == reflect.Ptr || kind == reflect.Interface) && v.IsNil() {
+			return true
+		}
+		return z.IsZero()
+	}
+	switch kind {
+	case reflect.String:
+		return len(v.String()) == 0
+	case reflect.Interface, reflect.Ptr:
+		return v.IsNil()
+	case reflect.Slice:
+		return v.Len() == 0
+	case reflect.Map:
+		return v.Len() == 0
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		return v.Int() == 0
+	case reflect.Float32, reflect.Float64:
+		return v.Float() == 0
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		return v.Uint() == 0
+	case reflect.Bool:
+		return !v.Bool()
+	case reflect.Struct:
+		vt := v.Type()
+		for i := v.NumField() - 1; i >= 0; i-- {
+			if vt.Field(i).PkgPath != "" {
+				continue // Private field
+			}
+			if !isZero(v.Field(i)) {
+				return false
+			}
+		}
+		return true
+	}
+	return false
+}

cli/vendor/gopkg.in/yaml.v3/yamlh.go

@@ -0,0 +1,807 @@
+//
+// Copyright (c) 2011-2019 Canonical Ltd
+// Copyright (c) 2006-2010 Kirill Simonov
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy of
+// this software and associated documentation files (the "Software"), to deal in
+// the Software without restriction, including without limitation the rights to
+// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+// of the Software, and to permit persons to whom the Software is furnished to do
+// so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+package yaml
+
+import (
+	"fmt"
+	"io"
+)
+
+// The version directive data.
+type yaml_version_directive_t struct {
+	major int8 // The major version number.
+	minor int8 // The minor version number.
+}
+
+// The tag directive data.
+type yaml_tag_directive_t struct {
+	handle []byte // The tag handle.
+	prefix []byte // The tag prefix.
+}
+
+type yaml_encoding_t int
+
+// The stream encoding.
+const (
+	// Let the parser choose the encoding.
+	yaml_ANY_ENCODING yaml_encoding_t = iota
+
+	yaml_UTF8_ENCODING    // The default UTF-8 encoding.
+	yaml_UTF16LE_ENCODING // The UTF-16-LE encoding with BOM.
+	yaml_UTF16BE_ENCODING // The UTF-16-BE encoding with BOM.
+)
+
+type yaml_break_t int
+
+// Line break types.
+const (
+	// Let the parser choose the break type.
+	yaml_ANY_BREAK yaml_break_t = iota
+
+	yaml_CR_BREAK   // Use CR for line breaks (Mac style).
+	yaml_LN_BREAK   // Use LN for line breaks (Unix style).
+	yaml_CRLN_BREAK // Use CR LN for line breaks (DOS style).
+)
+
+type yaml_error_type_t int
+
+// Many bad things could happen with the parser and emitter.
+const (
+	// No error is produced.
+	yaml_NO_ERROR yaml_error_type_t = iota
+
+	yaml_MEMORY_ERROR   // Cannot allocate or reallocate a block of memory.
+	yaml_READER_ERROR   // Cannot read or decode the input stream.
+	yaml_SCANNER_ERROR  // Cannot scan the input stream.
+	yaml_PARSER_ERROR   // Cannot parse the input stream.
+	yaml_COMPOSER_ERROR // Cannot compose a YAML document.
+	yaml_WRITER_ERROR   // Cannot write to the output stream.
+	yaml_EMITTER_ERROR  // Cannot emit a YAML stream.
+)
+
+// The pointer position.
+type yaml_mark_t struct {
+	index  int // The position index.
+	line   int // The position line.
+	column int // The position column.
+}
+
+// Node Styles
+
+type yaml_style_t int8
+
+type yaml_scalar_style_t yaml_style_t
+
+// Scalar styles.
+const (
+	// Let the emitter choose the style.
+	yaml_ANY_SCALAR_STYLE yaml_scalar_style_t = 0
+
+	yaml_PLAIN_SCALAR_STYLE         yaml_scalar_style_t = 1 << iota // The plain scalar style.
+	yaml_SINGLE_QUOTED_SCALAR_STYLE                                 // The single-quoted scalar style.
+	yaml_DOUBLE_QUOTED_SCALAR_STYLE                                 // The double-quoted scalar style.
+	yaml_LITERAL_SCALAR_STYLE                                       // The literal scalar style.
+	yaml_FOLDED_SCALAR_STYLE                                        // The folded scalar style.
+)
+
+type yaml_sequence_style_t yaml_style_t
+
+// Sequence styles.
+const (
+	// Let the emitter choose the style.
+	yaml_ANY_SEQUENCE_STYLE yaml_sequence_style_t = iota
+
+	yaml_BLOCK_SEQUENCE_STYLE // The block sequence style.
+	yaml_FLOW_SEQUENCE_STYLE  // The flow sequence style.
+)
+
+type yaml_mapping_style_t yaml_style_t
+
+// Mapping styles.
+const (
+	// Let the emitter choose the style.
+	yaml_ANY_MAPPING_STYLE yaml_mapping_style_t = iota
+
+	yaml_BLOCK_MAPPING_STYLE // The block mapping style.
+	yaml_FLOW_MAPPING_STYLE  // The flow mapping style.
+)
+
+// Tokens
+
+type yaml_token_type_t int
+
+// Token types.
+const (
+	// An empty token.
+	yaml_NO_TOKEN yaml_token_type_t = iota
+
+	yaml_STREAM_START_TOKEN // A STREAM-START token.
+	yaml_STREAM_END_TOKEN   // A STREAM-END token.
+
+	yaml_VERSION_DIRECTIVE_TOKEN // A VERSION-DIRECTIVE token.
+	yaml_TAG_DIRECTIVE_TOKEN     // A TAG-DIRECTIVE token.
+	yaml_DOCUMENT_START_TOKEN    // A DOCUMENT-START token.
+	yaml_DOCUMENT_END_TOKEN      // A DOCUMENT-END token.
+
+	yaml_BLOCK_SEQUENCE_START_TOKEN // A BLOCK-SEQUENCE-START token.
+	yaml_BLOCK_MAPPING_START_TOKEN  // A BLOCK-SEQUENCE-END token.
+	yaml_BLOCK_END_TOKEN            // A BLOCK-END token.
+
+	yaml_FLOW_SEQUENCE_START_TOKEN // A FLOW-SEQUENCE-START token.
+	yaml_FLOW_SEQUENCE_END_TOKEN   // A FLOW-SEQUENCE-END token.
+	yaml_FLOW_MAPPING_START_TOKEN  // A FLOW-MAPPING-START token.
+	yaml_FLOW_MAPPING_END_TOKEN    // A FLOW-MAPPING-END token.
+
+	yaml_BLOCK_ENTRY_TOKEN // A BLOCK-ENTRY token.
+	yaml_FLOW_ENTRY_TOKEN  // A FLOW-ENTRY token.
+	yaml_KEY_TOKEN         // A KEY token.
+	yaml_VALUE_TOKEN       // A VALUE token.
+
+	yaml_ALIAS_TOKEN  // An ALIAS token.
+	yaml_ANCHOR_TOKEN // An ANCHOR token.
+	yaml_TAG_TOKEN    // A TAG token.
+	yaml_SCALAR_TOKEN // A SCALAR token.
+)
+
+func (tt yaml_token_type_t) String() string {
+	switch tt {
+	case yaml_NO_TOKEN:
+		return "yaml_NO_TOKEN"
+	case yaml_STREAM_START_TOKEN:
+		return "yaml_STREAM_START_TOKEN"
+	case yaml_STREAM_END_TOKEN:
+		return "yaml_STREAM_END_TOKEN"
+	case yaml_VERSION_DIRECTIVE_TOKEN:
+		return "yaml_VERSION_DIRECTIVE_TOKEN"
+	case yaml_TAG_DIRECTIVE_TOKEN:
+		return "yaml_TAG_DIRECTIVE_TOKEN"
+	case yaml_DOCUMENT_START_TOKEN:
+		return "yaml_DOCUMENT_START_TOKEN"
+	case yaml_DOCUMENT_END_TOKEN:
+		return "yaml_DOCUMENT_END_TOKEN"
+	case yaml_BLOCK_SEQUENCE_START_TOKEN:
+		return "yaml_BLOCK_SEQUENCE_START_TOKEN"
+	case yaml_BLOCK_MAPPING_START_TOKEN:
+		return "yaml_BLOCK_MAPPING_START_TOKEN"
+	case yaml_BLOCK_END_TOKEN:
+		return "yaml_BLOCK_END_TOKEN"
+	case yaml_FLOW_SEQUENCE_START_TOKEN:
+		return "yaml_FLOW_SEQUENCE_START_TOKEN"
+	case yaml_FLOW_SEQUENCE_END_TOKEN:
+		return "yaml_FLOW_SEQUENCE_END_TOKEN"
+	case yaml_FLOW_MAPPING_START_TOKEN:
+		return "yaml_FLOW_MAPPING_START_TOKEN"
+	case yaml_FLOW_MAPPING_END_TOKEN:
+		return "yaml_FLOW_MAPPING_END_TOKEN"
+	case yaml_BLOCK_ENTRY_TOKEN:
+		return "yaml_BLOCK_ENTRY_TOKEN"
+	case yaml_FLOW_ENTRY_TOKEN:
+		return "yaml_FLOW_ENTRY_TOKEN"
+	case yaml_KEY_TOKEN:
+		return "yaml_KEY_TOKEN"
+	case yaml_VALUE_TOKEN:
+		return "yaml_VALUE_TOKEN"
+	case yaml_ALIAS_TOKEN:
+		return "yaml_ALIAS_TOKEN"
+	case yaml_ANCHOR_TOKEN:
+		return "yaml_ANCHOR_TOKEN"
+	case yaml_TAG_TOKEN:
+		return "yaml_TAG_TOKEN"
+	case yaml_SCALAR_TOKEN:
+		return "yaml_SCALAR_TOKEN"
+	}
+	return "<unknown token>"
+}
+
+// The token structure.
+type yaml_token_t struct {
+	// The token type.
+	typ yaml_token_type_t
+
+	// The start/end of the token.
+	start_mark, end_mark yaml_mark_t
+
+	// The stream encoding (for yaml_STREAM_START_TOKEN).
+	encoding yaml_encoding_t
+
+	// The alias/anchor/scalar value or tag/tag directive handle
+	// (for yaml_ALIAS_TOKEN, yaml_ANCHOR_TOKEN, yaml_SCALAR_TOKEN, yaml_TAG_TOKEN, yaml_TAG_DIRECTIVE_TOKEN).
+	value []byte
+
+	// The tag suffix (for yaml_TAG_TOKEN).
+	suffix []byte
+
+	// The tag directive prefix (for yaml_TAG_DIRECTIVE_TOKEN).
+	prefix []byte
+
+	// The scalar style (for yaml_SCALAR_TOKEN).
+	style yaml_scalar_style_t
+
+	// The version directive major/minor (for yaml_VERSION_DIRECTIVE_TOKEN).
+	major, minor int8
+}
+
+// Events
+
+type yaml_event_type_t int8
+
+// Event types.
+const (
+	// An empty event.
+	yaml_NO_EVENT yaml_event_type_t = iota
+
+	yaml_STREAM_START_EVENT   // A STREAM-START event.
+	yaml_STREAM_END_EVENT     // A STREAM-END event.
+	yaml_DOCUMENT_START_EVENT // A DOCUMENT-START event.
+	yaml_DOCUMENT_END_EVENT   // A DOCUMENT-END event.
+	yaml_ALIAS_EVENT          // An ALIAS event.
+	yaml_SCALAR_EVENT         // A SCALAR event.
+	yaml_SEQUENCE_START_EVENT // A SEQUENCE-START event.
+	yaml_SEQUENCE_END_EVENT   // A SEQUENCE-END event.
+	yaml_MAPPING_START_EVENT  // A MAPPING-START event.
+	yaml_MAPPING_END_EVENT    // A MAPPING-END event.
+	yaml_TAIL_COMMENT_EVENT
+)
+
+var eventStrings = []string{
+	yaml_NO_EVENT:             "none",
+	yaml_STREAM_START_EVENT:   "stream start",
+	yaml_STREAM_END_EVENT:     "stream end",
+	yaml_DOCUMENT_START_EVENT: "document start",
+	yaml_DOCUMENT_END_EVENT:   "document end",
+	yaml_ALIAS_EVENT:          "alias",
+	yaml_SCALAR_EVENT:         "scalar",
+	yaml_SEQUENCE_START_EVENT: "sequence start",
+	yaml_SEQUENCE_END_EVENT:   "sequence end",
+	yaml_MAPPING_START_EVENT:  "mapping start",
+	yaml_MAPPING_END_EVENT:    "mapping end",
+	yaml_TAIL_COMMENT_EVENT:   "tail comment",
+}
+
+func (e yaml_event_type_t) String() string {
+	if e < 0 || int(e) >= len(eventStrings) {
+		return fmt.Sprintf("unknown event %d", e)
+	}
+	return eventStrings[e]
+}
+
+// The event structure.
+type yaml_event_t struct {
+
+	// The event type.
+	typ yaml_event_type_t
+
+	// The start and end of the event.
+	start_mark, end_mark yaml_mark_t
+
+	// The document encoding (for yaml_STREAM_START_EVENT).
+	encoding yaml_encoding_t
+
+	// The version directive (for yaml_DOCUMENT_START_EVENT).
+	version_directive *yaml_version_directive_t
+
+	// The list of tag directives (for yaml_DOCUMENT_START_EVENT).
+	tag_directives []yaml_tag_directive_t
+
+	// The comments
+	head_comment []byte
+	line_comment []byte
+	foot_comment []byte
+	tail_comment []byte
+
+	// The anchor (for yaml_SCALAR_EVENT, yaml_SEQUENCE_START_EVENT, yaml_MAPPING_START_EVENT, yaml_ALIAS_EVENT).
+	anchor []byte
+
+	// The tag (for yaml_SCALAR_EVENT, yaml_SEQUENCE_START_EVENT, yaml_MAPPING_START_EVENT).
+	tag []byte
+
+	// The scalar value (for yaml_SCALAR_EVENT).
+	value []byte
+
+	// Is the document start/end indicator implicit, or the tag optional?
+	// (for yaml_DOCUMENT_START_EVENT, yaml_DOCUMENT_END_EVENT, yaml_SEQUENCE_START_EVENT, yaml_MAPPING_START_EVENT, yaml_SCALAR_EVENT).
+	implicit bool
+
+	// Is the tag optional for any non-plain style? (for yaml_SCALAR_EVENT).
+	quoted_implicit bool
+
+	// The style (for yaml_SCALAR_EVENT, yaml_SEQUENCE_START_EVENT, yaml_MAPPING_START_EVENT).
+	style yaml_style_t
+}
+
+func (e *yaml_event_t) scalar_style() yaml_scalar_style_t     { return yaml_scalar_style_t(e.style) }
+func (e *yaml_event_t) sequence_style() yaml_sequence_style_t { return yaml_sequence_style_t(e.style) }
+func (e *yaml_event_t) mapping_style() yaml_mapping_style_t   { return yaml_mapping_style_t(e.style) }
+
+// Nodes
+
+const (
+	yaml_NULL_TAG      = "tag:yaml.org,2002:null"      // The tag !!null with the only possible value: null.
+	yaml_BOOL_TAG      = "tag:yaml.org,2002:bool"      // The tag !!bool with the values: true and false.
+	yaml_STR_TAG       = "tag:yaml.org,2002:str"       // The tag !!str for string values.
+	yaml_INT_TAG       = "tag:yaml.org,2002:int"       // The tag !!int for integer values.
+	yaml_FLOAT_TAG     = "tag:yaml.org,2002:float"     // The tag !!float for float values.
+	yaml_TIMESTAMP_TAG = "tag:yaml.org,2002:timestamp" // The tag !!timestamp for date and time values.
+
+	yaml_SEQ_TAG = "tag:yaml.org,2002:seq" // The tag !!seq is used to denote sequences.
+	yaml_MAP_TAG = "tag:yaml.org,2002:map" // The tag !!map is used to denote mapping.
+
+	// Not in original libyaml.
+	yaml_BINARY_TAG = "tag:yaml.org,2002:binary"
+	yaml_MERGE_TAG  = "tag:yaml.org,2002:merge"
+
+	yaml_DEFAULT_SCALAR_TAG   = yaml_STR_TAG // The default scalar tag is !!str.
+	yaml_DEFAULT_SEQUENCE_TAG = yaml_SEQ_TAG // The default sequence tag is !!seq.
+	yaml_DEFAULT_MAPPING_TAG  = yaml_MAP_TAG // The default mapping tag is !!map.
+)
+
+type yaml_node_type_t int
+
+// Node types.
+const (
+	// An empty node.
+	yaml_NO_NODE yaml_node_type_t = iota
+
+	yaml_SCALAR_NODE   // A scalar node.
+	yaml_SEQUENCE_NODE // A sequence node.
+	yaml_MAPPING_NODE  // A mapping node.
+)
+
+// An element of a sequence node.
+type yaml_node_item_t int
+
+// An element of a mapping node.
+type yaml_node_pair_t struct {
+	key   int // The key of the element.
+	value int // The value of the element.
+}
+
+// The node structure.
+type yaml_node_t struct {
+	typ yaml_node_type_t // The node type.
+	tag []byte           // The node tag.
+
+	// The node data.
+
+	// The scalar parameters (for yaml_SCALAR_NODE).
+	scalar struct {
+		value  []byte              // The scalar value.
+		length int                 // The length of the scalar value.
+		style  yaml_scalar_style_t // The scalar style.
+	}
+
+	// The sequence parameters (for YAML_SEQUENCE_NODE).
+	sequence struct {
+		items_data []yaml_node_item_t    // The stack of sequence items.
+		style      yaml_sequence_style_t // The sequence style.
+	}
+
+	// The mapping parameters (for yaml_MAPPING_NODE).
+	mapping struct {
+		pairs_data  []yaml_node_pair_t   // The stack of mapping pairs (key, value).
+		pairs_start *yaml_node_pair_t    // The beginning of the stack.
+		pairs_end   *yaml_node_pair_t    // The end of the stack.
+		pairs_top   *yaml_node_pair_t    // The top of the stack.
+		style       yaml_mapping_style_t // The mapping style.
+	}
+
+	start_mark yaml_mark_t // The beginning of the node.
+	end_mark   yaml_mark_t // The end of the node.
+
+}
+
+// The document structure.
+type yaml_document_t struct {
+
+	// The document nodes.
+	nodes []yaml_node_t
+
+	// The version directive.
+	version_directive *yaml_version_directive_t
+
+	// The list of tag directives.
+	tag_directives_data  []yaml_tag_directive_t
+	tag_directives_start int // The beginning of the tag directives list.
+	tag_directives_end   int // The end of the tag directives list.
+
+	start_implicit int // Is the document start indicator implicit?
+	end_implicit   int // Is the document end indicator implicit?
+
+	// The start/end of the document.
+	start_mark, end_mark yaml_mark_t
+}
+
+// The prototype of a read handler.
+//
+// The read handler is called when the parser needs to read more bytes from the
+// source. The handler should write not more than size bytes to the buffer.
+// The number of written bytes should be set to the size_read variable.
+//
+// [in,out]   data        A pointer to an application data specified by
+//                        yaml_parser_set_input().
+// [out]      buffer      The buffer to write the data from the source.
+// [in]       size        The size of the buffer.
+// [out]      size_read   The actual number of bytes read from the source.
+//
+// On success, the handler should return 1.  If the handler failed,
+// the returned value should be 0. On EOF, the handler should set the
+// size_read to 0 and return 1.
+type yaml_read_handler_t func(parser *yaml_parser_t, buffer []byte) (n int, err error)
+
+// This structure holds information about a potential simple key.
+type yaml_simple_key_t struct {
+	possible     bool        // Is a simple key possible?
+	required     bool        // Is a simple key required?
+	token_number int         // The number of the token.
+	mark         yaml_mark_t // The position mark.
+}
+
+// The states of the parser.
+type yaml_parser_state_t int
+
+const (
+	yaml_PARSE_STREAM_START_STATE yaml_parser_state_t = iota
+
+	yaml_PARSE_IMPLICIT_DOCUMENT_START_STATE           // Expect the beginning of an implicit document.
+	yaml_PARSE_DOCUMENT_START_STATE                    // Expect DOCUMENT-START.
+	yaml_PARSE_DOCUMENT_CONTENT_STATE                  // Expect the content of a document.
+	yaml_PARSE_DOCUMENT_END_STATE                      // Expect DOCUMENT-END.
+	yaml_PARSE_BLOCK_NODE_STATE                        // Expect a block node.
+	yaml_PARSE_BLOCK_NODE_OR_INDENTLESS_SEQUENCE_STATE // Expect a block node or indentless sequence.
+	yaml_PARSE_FLOW_NODE_STATE                         // Expect a flow node.
+	yaml_PARSE_BLOCK_SEQUENCE_FIRST_ENTRY_STATE        // Expect the first entry of a block sequence.
+	yaml_PARSE_BLOCK_SEQUENCE_ENTRY_STATE              // Expect an entry of a block sequence.
+	yaml_PARSE_INDENTLESS_SEQUENCE_ENTRY_STATE         // Expect an entry of an indentless sequence.
+	yaml_PARSE_BLOCK_MAPPING_FIRST_KEY_STATE           // Expect the first key of a block mapping.
+	yaml_PARSE_BLOCK_MAPPING_KEY_STATE                 // Expect a block mapping key.
+	yaml_PARSE_BLOCK_MAPPING_VALUE_STATE               // Expect a block mapping value.
+	yaml_PARSE_FLOW_SEQUENCE_FIRST_ENTRY_STATE         // Expect the first entry of a flow sequence.
+	yaml_PARSE_FLOW_SEQUENCE_ENTRY_STATE               // Expect an entry of a flow sequence.
+	yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_KEY_STATE   // Expect a key of an ordered mapping.
+	yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_VALUE_STATE // Expect a value of an ordered mapping.
+	yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_END_STATE   // Expect the and of an ordered mapping entry.
+	yaml_PARSE_FLOW_MAPPING_FIRST_KEY_STATE            // Expect the first key of a flow mapping.
+	yaml_PARSE_FLOW_MAPPING_KEY_STATE                  // Expect a key of a flow mapping.
+	yaml_PARSE_FLOW_MAPPING_VALUE_STATE                // Expect a value of a flow mapping.
+	yaml_PARSE_FLOW_MAPPING_EMPTY_VALUE_STATE          // Expect an empty value of a flow mapping.
+	yaml_PARSE_END_STATE                               // Expect nothing.
+)
+
+func (ps yaml_parser_state_t) String() string {
+	switch ps {
+	case yaml_PARSE_STREAM_START_STATE:
+		return "yaml_PARSE_STREAM_START_STATE"
+	case yaml_PARSE_IMPLICIT_DOCUMENT_START_STATE:
+		return "yaml_PARSE_IMPLICIT_DOCUMENT_START_STATE"
+	case yaml_PARSE_DOCUMENT_START_STATE:
+		return "yaml_PARSE_DOCUMENT_START_STATE"
+	case yaml_PARSE_DOCUMENT_CONTENT_STATE:
+		return "yaml_PARSE_DOCUMENT_CONTENT_STATE"
+	case yaml_PARSE_DOCUMENT_END_STATE:
+		return "yaml_PARSE_DOCUMENT_END_STATE"
+	case yaml_PARSE_BLOCK_NODE_STATE:
+		return "yaml_PARSE_BLOCK_NODE_STATE"
+	case yaml_PARSE_BLOCK_NODE_OR_INDENTLESS_SEQUENCE_STATE:
+		return "yaml_PARSE_BLOCK_NODE_OR_INDENTLESS_SEQUENCE_STATE"
+	case yaml_PARSE_FLOW_NODE_STATE:
+		return "yaml_PARSE_FLOW_NODE_STATE"
+	case yaml_PARSE_BLOCK_SEQUENCE_FIRST_ENTRY_STATE:
+		return "yaml_PARSE_BLOCK_SEQUENCE_FIRST_ENTRY_STATE"
+	case yaml_PARSE_BLOCK_SEQUENCE_ENTRY_STATE:
+		return "yaml_PARSE_BLOCK_SEQUENCE_ENTRY_STATE"
+	case yaml_PARSE_INDENTLESS_SEQUENCE_ENTRY_STATE:
+		return "yaml_PARSE_INDENTLESS_SEQUENCE_ENTRY_STATE"
+	case yaml_PARSE_BLOCK_MAPPING_FIRST_KEY_STATE:
+		return "yaml_PARSE_BLOCK_MAPPING_FIRST_KEY_STATE"
+	case yaml_PARSE_BLOCK_MAPPING_KEY_STATE:
+		return "yaml_PARSE_BLOCK_MAPPING_KEY_STATE"
+	case yaml_PARSE_BLOCK_MAPPING_VALUE_STATE:
+		return "yaml_PARSE_BLOCK_MAPPING_VALUE_STATE"
+	case yaml_PARSE_FLOW_SEQUENCE_FIRST_ENTRY_STATE:
+		return "yaml_PARSE_FLOW_SEQUENCE_FIRST_ENTRY_STATE"
+	case yaml_PARSE_FLOW_SEQUENCE_ENTRY_STATE:
+		return "yaml_PARSE_FLOW_SEQUENCE_ENTRY_STATE"
+	case yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_KEY_STATE:
+		return "yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_KEY_STATE"
+	case yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_VALUE_STATE:
+		return "yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_VALUE_STATE"
+	case yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_END_STATE:
+		return "yaml_PARSE_FLOW_SEQUENCE_ENTRY_MAPPING_END_STATE"
+	case yaml_PARSE_FLOW_MAPPING_FIRST_KEY_STATE:
+		return "yaml_PARSE_FLOW_MAPPING_FIRST_KEY_STATE"
+	case yaml_PARSE_FLOW_MAPPING_KEY_STATE:
+		return "yaml_PARSE_FLOW_MAPPING_KEY_STATE"
+	case yaml_PARSE_FLOW_MAPPING_VALUE_STATE:
+		return "yaml_PARSE_FLOW_MAPPING_VALUE_STATE"
+	case yaml_PARSE_FLOW_MAPPING_EMPTY_VALUE_STATE:
+		return "yaml_PARSE_FLOW_MAPPING_EMPTY_VALUE_STATE"
+	case yaml_PARSE_END_STATE:
+		return "yaml_PARSE_END_STATE"
+	}
+	return "<unknown parser state>"
+}
+
+// This structure holds aliases data.
+type yaml_alias_data_t struct {
+	anchor []byte      // The anchor.
+	index  int         // The node id.
+	mark   yaml_mark_t // The anchor mark.
+}
+
+// The parser structure.
+//
+// All members are internal. Manage the structure using the
+// yaml_parser_ family of functions.
+type yaml_parser_t struct {
+
+	// Error handling
+
+	error yaml_error_type_t // Error type.
+
+	problem string // Error description.
+
+	// The byte about which the problem occurred.
+	problem_offset int
+	problem_value  int
+	problem_mark   yaml_mark_t
+
+	// The error context.
+	context      string
+	context_mark yaml_mark_t
+
+	// Reader stuff
+
+	read_handler yaml_read_handler_t // Read handler.
+
+	input_reader io.Reader // File input data.
+	input        []byte    // String input data.
+	input_pos    int
+
+	eof bool // EOF flag
+
+	buffer     []byte // The working buffer.
+	buffer_pos int    // The current position of the buffer.
+
+	unread int // The number of unread characters in the buffer.
+
+	newlines int // The number of line breaks since last non-break/non-blank character
+
+	raw_buffer     []byte // The raw buffer.
+	raw_buffer_pos int    // The current position of the buffer.
+
+	encoding yaml_encoding_t // The input encoding.
+
+	offset int         // The offset of the current position (in bytes).
+	mark   yaml_mark_t // The mark of the current position.
+
+	// Comments
+
+	head_comment []byte // The current head comments
+	line_comment []byte // The current line comments
+	foot_comment []byte // The current foot comments
+	tail_comment []byte // Foot comment that happens at the end of a block.
+	stem_comment []byte // Comment in item preceding a nested structure (list inside list item, etc)
+
+	comments      []yaml_comment_t // The folded comments for all parsed tokens
+	comments_head int
+
+	// Scanner stuff
+
+	stream_start_produced bool // Have we started to scan the input stream?
+	stream_end_produced   bool // Have we reached the end of the input stream?
+
+	flow_level int // The number of unclosed '[' and '{' indicators.
+
+	tokens          []yaml_token_t // The tokens queue.
+	tokens_head     int            // The head of the tokens queue.
+	tokens_parsed   int            // The number of tokens fetched from the queue.
+	token_available bool           // Does the tokens queue contain a token ready for dequeueing.
+
+	indent  int   // The current indentation level.
+	indents []int // The indentation levels stack.
+
+	simple_key_allowed bool                // May a simple key occur at the current position?
+	simple_keys        []yaml_simple_key_t // The stack of simple keys.
+	simple_keys_by_tok map[int]int         // possible simple_key indexes indexed by token_number
+
+	// Parser stuff
+
+	state          yaml_parser_state_t    // The current parser state.
+	states         []yaml_parser_state_t  // The parser states stack.
+	marks          []yaml_mark_t          // The stack of marks.
+	tag_directives []yaml_tag_directive_t // The list of TAG directives.
+
+	// Dumper stuff
+
+	aliases []yaml_alias_data_t // The alias data.
+
+	document *yaml_document_t // The currently parsed document.
+}
+
+type yaml_comment_t struct {
+
+	scan_mark  yaml_mark_t // Position where scanning for comments started
+	token_mark yaml_mark_t // Position after which tokens will be associated with this comment
+	start_mark yaml_mark_t // Position of '#' comment mark
+	end_mark   yaml_mark_t // Position where comment terminated
+
+	head []byte
+	line []byte
+	foot []byte
+}
+
+// Emitter Definitions
+
+// The prototype of a write handler.
+//
+// The write handler is called when the emitter needs to flush the accumulated
+// characters to the output.  The handler should write @a size bytes of the
+// @a buffer to the output.
+//
+// @param[in,out]   data        A pointer to an application data specified by
+//                              yaml_emitter_set_output().
+// @param[in]       buffer      The buffer with bytes to be written.
+// @param[in]       size        The size of the buffer.
+//
+// @returns On success, the handler should return @c 1.  If the handler failed,
+// the returned value should be @c 0.
+//
+type yaml_write_handler_t func(emitter *yaml_emitter_t, buffer []byte) error
+
+type yaml_emitter_state_t int
+
+// The emitter states.
+const (
+	// Expect STREAM-START.
+	yaml_EMIT_STREAM_START_STATE yaml_emitter_state_t = iota
+
+	yaml_EMIT_FIRST_DOCUMENT_START_STATE       // Expect the first DOCUMENT-START or STREAM-END.
+	yaml_EMIT_DOCUMENT_START_STATE             // Expect DOCUMENT-START or STREAM-END.
+	yaml_EMIT_DOCUMENT_CONTENT_STATE           // Expect the content of a document.
+	yaml_EMIT_DOCUMENT_END_STATE               // Expect DOCUMENT-END.
+	yaml_EMIT_FLOW_SEQUENCE_FIRST_ITEM_STATE   // Expect the first item of a flow sequence.
+	yaml_EMIT_FLOW_SEQUENCE_TRAIL_ITEM_STATE   // Expect the next item of a flow sequence, with the comma already written out
+	yaml_EMIT_FLOW_SEQUENCE_ITEM_STATE         // Expect an item of a flow sequence.
+	yaml_EMIT_FLOW_MAPPING_FIRST_KEY_STATE     // Expect the first key of a flow mapping.
+	yaml_EMIT_FLOW_MAPPING_TRAIL_KEY_STATE     // Expect the next key of a flow mapping, with the comma already written out
+	yaml_EMIT_FLOW_MAPPING_KEY_STATE           // Expect a key of a flow mapping.
+	yaml_EMIT_FLOW_MAPPING_SIMPLE_VALUE_STATE  // Expect a value for a simple key of a flow mapping.
+	yaml_EMIT_FLOW_MAPPING_VALUE_STATE         // Expect a value of a flow mapping.
+	yaml_EMIT_BLOCK_SEQUENCE_FIRST_ITEM_STATE  // Expect the first item of a block sequence.
+	yaml_EMIT_BLOCK_SEQUENCE_ITEM_STATE        // Expect an item of a block sequence.
+	yaml_EMIT_BLOCK_MAPPING_FIRST_KEY_STATE    // Expect the first key of a block mapping.
+	yaml_EMIT_BLOCK_MAPPING_KEY_STATE          // Expect the key of a block mapping.
+	yaml_EMIT_BLOCK_MAPPING_SIMPLE_VALUE_STATE // Expect a value for a simple key of a block mapping.
+	yaml_EMIT_BLOCK_MAPPING_VALUE_STATE        // Expect a value of a block mapping.
+	yaml_EMIT_END_STATE                        // Expect nothing.
+)
+
+// The emitter structure.
+//
+// All members are internal.  Manage the structure using the @c yaml_emitter_
+// family of functions.
+type yaml_emitter_t struct {
+
+	// Error handling
+
+	error   yaml_error_type_t // Error type.
+	problem string            // Error description.
+
+	// Writer stuff
+
+	write_handler yaml_write_handler_t // Write handler.
+
+	output_buffer *[]byte   // String output data.
+	output_writer io.Writer // File output data.
+
+	buffer     []byte // The working buffer.
+	buffer_pos int    // The current position of the buffer.
+
+	raw_buffer     []byte // The raw buffer.
+	raw_buffer_pos int    // The current position of the buffer.
+
+	encoding yaml_encoding_t // The stream encoding.
+
+	// Emitter stuff
+
+	canonical   bool         // If the output is in the canonical style?
+	best_indent int          // The number of indentation spaces.
+	best_width  int          // The preferred width of the output lines.
+	unicode     bool         // Allow unescaped non-ASCII characters?
+	line_break  yaml_break_t // The preferred line break.
+
+	state  yaml_emitter_state_t   // The current emitter state.
+	states []yaml_emitter_state_t // The stack of states.
+
+	events      []yaml_event_t // The event queue.
+	events_head int            // The head of the event queue.
+
+	indents []int // The stack of indentation levels.
+
+	tag_directives []yaml_tag_directive_t // The list of tag directives.
+
+	indent int // The current indentation level.
+
+	flow_level int // The current flow level.
+
+	root_context       bool // Is it the document root context?
+	sequence_context   bool // Is it a sequence context?
+	mapping_context    bool // Is it a mapping context?
+	simple_key_context bool // Is it a simple mapping key context?
+
+	line       int  // The current line.
+	column     int  // The current column.
+	whitespace bool // If the last character was a whitespace?
+	indention  bool // If the last character was an indentation character (' ', '-', '?', ':')?
+	open_ended bool // If an explicit document end is required?
+
+	space_above bool // Is there's an empty line above?
+	foot_indent int  // The indent used to write the foot comment above, or -1 if none.
+
+	// Anchor analysis.
+	anchor_data struct {
+		anchor []byte // The anchor value.
+		alias  bool   // Is it an alias?
+	}
+
+	// Tag analysis.
+	tag_data struct {
+		handle []byte // The tag handle.
+		suffix []byte // The tag suffix.
+	}
+
+	// Scalar analysis.
+	scalar_data struct {
+		value                 []byte              // The scalar value.
+		multiline             bool                // Does the scalar contain line breaks?
+		flow_plain_allowed    bool                // Can the scalar be expessed in the flow plain style?
+		block_plain_allowed   bool                // Can the scalar be expressed in the block plain style?
+		single_quoted_allowed bool                // Can the scalar be expressed in the single quoted style?
+		block_allowed         bool                // Can the scalar be expressed in the literal or folded styles?
+		style                 yaml_scalar_style_t // The output style.
+	}
+
+	// Comments
+	head_comment []byte
+	line_comment []byte
+	foot_comment []byte
+	tail_comment []byte
+
+	key_line_comment []byte
+
+	// Dumper stuff
+
+	opened bool // If the stream was already opened?
+	closed bool // If the stream was already closed?
+
+	// The information associated with the document nodes.
+	anchors *struct {
+		references int  // The number of references.
+		anchor     int  // The anchor id.
+		serialized bool // If the node has been emitted?
+	}
+
+	last_anchor_id int // The last assigned anchor id.
+
+	document *yaml_document_t // The currently emitted document.
+}

cli/vendor/gopkg.in/yaml.v3/yamlprivateh.go

@@ -0,0 +1,198 @@
+// 
+// Copyright (c) 2011-2019 Canonical Ltd
+// Copyright (c) 2006-2010 Kirill Simonov
+// 
+// Permission is hereby granted, free of charge, to any person obtaining a copy of
+// this software and associated documentation files (the "Software"), to deal in
+// the Software without restriction, including without limitation the rights to
+// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+// of the Software, and to permit persons to whom the Software is furnished to do
+// so, subject to the following conditions:
+// 
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+package yaml
+
+const (
+	// The size of the input raw buffer.
+	input_raw_buffer_size = 512
+
+	// The size of the input buffer.
+	// It should be possible to decode the whole raw buffer.
+	input_buffer_size = input_raw_buffer_size * 3
+
+	// The size of the output buffer.
+	output_buffer_size = 128
+
+	// The size of the output raw buffer.
+	// It should be possible to encode the whole output buffer.
+	output_raw_buffer_size = (output_buffer_size*2 + 2)
+
+	// The size of other stacks and queues.
+	initial_stack_size  = 16
+	initial_queue_size  = 16
+	initial_string_size = 16
+)
+
+// Check if the character at the specified position is an alphabetical
+// character, a digit, '_', or '-'.
+func is_alpha(b []byte, i int) bool {
+	return b[i] >= '0' && b[i] <= '9' || b[i] >= 'A' && b[i] <= 'Z' || b[i] >= 'a' && b[i] <= 'z' || b[i] == '_' || b[i] == '-'
+}
+
+// Check if the character at the specified position is a digit.
+func is_digit(b []byte, i int) bool {
+	return b[i] >= '0' && b[i] <= '9'
+}
+
+// Get the value of a digit.
+func as_digit(b []byte, i int) int {
+	return int(b[i]) - '0'
+}
+
+// Check if the character at the specified position is a hex-digit.
+func is_hex(b []byte, i int) bool {
+	return b[i] >= '0' && b[i] <= '9' || b[i] >= 'A' && b[i] <= 'F' || b[i] >= 'a' && b[i] <= 'f'
+}
+
+// Get the value of a hex-digit.
+func as_hex(b []byte, i int) int {
+	bi := b[i]
+	if bi >= 'A' && bi <= 'F' {
+		return int(bi) - 'A' + 10
+	}
+	if bi >= 'a' && bi <= 'f' {
+		return int(bi) - 'a' + 10
+	}
+	return int(bi) - '0'
+}
+
+// Check if the character is ASCII.
+func is_ascii(b []byte, i int) bool {
+	return b[i] <= 0x7F
+}
+
+// Check if the character at the start of the buffer can be printed unescaped.
+func is_printable(b []byte, i int) bool {
+	return ((b[i] == 0x0A) || // . == #x0A
+		(b[i] >= 0x20 && b[i] <= 0x7E) || // #x20 <= . <= #x7E
+		(b[i] == 0xC2 && b[i+1] >= 0xA0) || // #0xA0 <= . <= #xD7FF
+		(b[i] > 0xC2 && b[i] < 0xED) ||
+		(b[i] == 0xED && b[i+1] < 0xA0) ||
+		(b[i] == 0xEE) ||
+		(b[i] == 0xEF && // #xE000 <= . <= #xFFFD
+			!(b[i+1] == 0xBB && b[i+2] == 0xBF) && // && . != #xFEFF
+			!(b[i+1] == 0xBF && (b[i+2] == 0xBE || b[i+2] == 0xBF))))
+}
+
+// Check if the character at the specified position is NUL.
+func is_z(b []byte, i int) bool {
+	return b[i] == 0x00
+}
+
+// Check if the beginning of the buffer is a BOM.
+func is_bom(b []byte, i int) bool {
+	return b[0] == 0xEF && b[1] == 0xBB && b[2] == 0xBF
+}
+
+// Check if the character at the specified position is space.
+func is_space(b []byte, i int) bool {
+	return b[i] == ' '
+}
+
+// Check if the character at the specified position is tab.
+func is_tab(b []byte, i int) bool {
+	return b[i] == '\t'
+}
+
+// Check if the character at the specified position is blank (space or tab).
+func is_blank(b []byte, i int) bool {
+	//return is_space(b, i) || is_tab(b, i)
+	return b[i] == ' ' || b[i] == '\t'
+}
+
+// Check if the character at the specified position is a line break.
+func is_break(b []byte, i int) bool {
+	return (b[i] == '\r' || // CR (#xD)
+		b[i] == '\n' || // LF (#xA)
+		b[i] == 0xC2 && b[i+1] == 0x85 || // NEL (#x85)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA8 || // LS (#x2028)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA9) // PS (#x2029)
+}
+
+func is_crlf(b []byte, i int) bool {
+	return b[i] == '\r' && b[i+1] == '\n'
+}
+
+// Check if the character is a line break or NUL.
+func is_breakz(b []byte, i int) bool {
+	//return is_break(b, i) || is_z(b, i)
+	return (
+		// is_break:
+		b[i] == '\r' || // CR (#xD)
+		b[i] == '\n' || // LF (#xA)
+		b[i] == 0xC2 && b[i+1] == 0x85 || // NEL (#x85)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA8 || // LS (#x2028)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA9 || // PS (#x2029)
+		// is_z:
+		b[i] == 0)
+}
+
+// Check if the character is a line break, space, or NUL.
+func is_spacez(b []byte, i int) bool {
+	//return is_space(b, i) || is_breakz(b, i)
+	return (
+		// is_space:
+		b[i] == ' ' ||
+		// is_breakz:
+		b[i] == '\r' || // CR (#xD)
+		b[i] == '\n' || // LF (#xA)
+		b[i] == 0xC2 && b[i+1] == 0x85 || // NEL (#x85)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA8 || // LS (#x2028)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA9 || // PS (#x2029)
+		b[i] == 0)
+}
+
+// Check if the character is a line break, space, tab, or NUL.
+func is_blankz(b []byte, i int) bool {
+	//return is_blank(b, i) || is_breakz(b, i)
+	return (
+		// is_blank:
+		b[i] == ' ' || b[i] == '\t' ||
+		// is_breakz:
+		b[i] == '\r' || // CR (#xD)
+		b[i] == '\n' || // LF (#xA)
+		b[i] == 0xC2 && b[i+1] == 0x85 || // NEL (#x85)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA8 || // LS (#x2028)
+		b[i] == 0xE2 && b[i+1] == 0x80 && b[i+2] == 0xA9 || // PS (#x2029)
+		b[i] == 0)
+}
+
+// Determine the width of the character.
+func width(b byte) int {
+	// Don't replace these by a switch without first
+	// confirming that it is being inlined.
+	if b&0x80 == 0x00 {
+		return 1
+	}
+	if b&0xE0 == 0xC0 {
+		return 2
+	}
+	if b&0xF0 == 0xE0 {
+		return 3
+	}
+	if b&0xF8 == 0xF0 {
+		return 4
+	}
+	return 0
+
+}

cli/vendor/modules.txt

@@ -0,0 +1,3 @@
+# gopkg.in/yaml.v3 v3.0.1
+## explicit
+gopkg.in/yaml.v3