marketplace/plugins/kettle/skills/sync/SKILL.md

name, description

name

description

sync

Move issues between this project's local store and its Gitea tracker with the `kettle` binary — pull issues into `.kettle/issues/`, push local ones up (which deletes the local file), list what the tracker holds, post comments, close and reopen, and evict what the tracker says is closed. Load when the user asks to fetch or publish an issue, see what exists in the tracker, comment on one, or close one. Writing, grepping, validating and graphing an issue's content is /kettle:issue and needs no network.

/kettle:sync — the bridge between the local store and the tracker

One job: carry issues between <project>/.kettle/issues/<id>.md and Gitea. Everything about what an issue is — format, types, validation, the dependency graph — belongs to /kettle:issue, and this layer neither redefines nor second-guesses it. Knowledge flows one way: delete the tracker from the world and the issue domain does not notice.

No kettle on PATH? command not found: kettle is the whole story — the Python scripts this plugin used to ship are gone and no other CLI is a substitute. Stop and tell the operator to install it: cd cli && go build -o ~/.local/bin/kettle ./cmd/kettle in the marketplace repository (go.mod requires go 1.26), or go install git.noodles.cam/claude-skills/marketplace/cli/cmd/kettle@latest.

No --login, no --repo, no guard

Which login this project runs under and which repository its issues belong to are facts about the project, stated once by kettle init and kept in .kettle/config.yaml; the token lives in one file per machine that no working tree can see. There is nothing to pass and nothing to police — the old PreToolUse guard hook and its --login "$GITEA_LOGIN" placeholder are gone, along with the failure they existed to catch. kettle labels --repo owner/name is the single exception, because bootstrapping a repository's label set is the one operation whose target is not this project.

A cross-repository address is still an address: kettle pull owner/repo#42 re-points the client for that one call and comes back with the same credentials and the same scratchpad. 42, #42, owner/repo#42 and a full issue URL are four spellings of one key.

No login, an unknown login name, a 401: report it and stop — /kettle:auth.

Never read an issue through a raw API dump

kettle api repos/{owner}/{repo}/issues/42 will answer, and answering is the problem: the whole payload — avatars, nested user objects, every comment body — lands in your context whether you need it or not. kettle pull writes flat markdown and prints a compact line per issue; kettle remote lists the tracker without writing anything at all. Use those. /kettle:api is for the entities that have no command, and it says the same thing from its side.

The round trip is one rule

The store holds what has not left this machine.

A successful kettle push deletes <id>.md and every sidecar under that slug — on create and on --update alike, one rule with no exception — and prints the number and URL the issue now lives at. Once the tracker has the issue, the tracker is the issue.

The deletion is the last thing that happens, and only after all three of: the call came back 2xx, the answer carries the number that was written, and the number → slug ledger has been written. Network down, a 422, an answer about another issue — the file stays exactly where it is and the run stops.

An origin: local issue that has never been pushed is never touched by any of this. That file is the only copy of that work.

The slug survives the round trip two ways over, which is why the file can be deleted at all:

where	survives
<!-- kettle:id wire-sqlc-appclick -->, first line of the tracker-side body	a rename in the web UI, a lost ledger, a fresh clone, another machine
.kettle/issues/.remote.json, number → slug	the local file being deleted

The marker never appears in the local file: one is put at the top on the way up and every one is stripped on the way down. A pull consults the ledger first (it is the one that knows what is on disk now), then the marker, then slugifies the title for an issue filed in the web UI that never had a local name — and a marker is taken at its word only when that slug is free, because a name already in use is a collision and not an identity.

Nothing prunes the ledger — not a push, not an eviction. Its entries are meant to outlive the files they name.

Pulling is a fetch, not a merge

A pull overwrites the body. Unpushed local edits are lost, with exactly one exception: checkbox state. A tick is monotone, so for a checkbox line whose text matches on both sides [x] wins from either — tick it in the web UI, tick it locally, tick it in both, the tick survives. The price is real and stated: a box unticked in the web UI comes back on the next pull. Untick locally, then kettle push --update.

Two spellings, and they are different operations:

• by key — an address. It fetches the issue in any state, because a number is not a question about state.
• by filter (--milestone, --label, -q) — a query. Closed issues are enumerated and left out, and --limit bounds what is stored, never what is read.

Do not loop over numbers to fetch a group; pass the filter. And a pull returns the unit of work, not one row of it: blockers come down with it recursively to --depth, which costs a request per issue and per outside blocker. That is what --no-deps buys back. A blocker the filter did not select still lands on disk, deliberately — it is there because a stored issue named it.

Comments ride along into <id>.comments.md with no flag, and cost nothing when the payload says the thread is empty. They are pull-only in the store: editing that file changes nothing in the tracker. kettle comment is the way, and it refetches the thread after writing so the local copy is not stale by the comment it just made.

Closing, and evicting what the tracker says is closed

kettle close sends {"state": …} and nothing else — no title, no body, no labels. Closing is not an edit; editing is pull → change → push --update. Explicit ids only: there is no --milestone and no --label, because which issues are finished is a judgement about content and this command only carries one out. The local file is rewritten only after the tracker confirms that very write. A tracker that refuses to close an issue its own dependency graph still blocks says so in its own words — close the blockers first, or unlink them.

kettle sync-evict is kettle evict with one thing in front of it: a state: that is not stale. Every candidate is asked about before anything is removed, and one bad answer evicts nothing at all — not even the issues whose answers had already arrived. origin: local is never asked about and never evicted; a tracked issue whose handle is unreadable is reported and kept.

Labels

Push creates the labels its issues happen to use, which means a repository grows the set in pieces and nobody can filter by type/bug in the web UI until somebody pushes a bug. kettle labels lays the canonical type/* and severity/* set down in one run instead. An exact name is left alone; a lookalike (bug, Bug, kind/bug, type: bug) is reported with its id and never touched, because renaming somebody else's label is a decision and not a step; colour or exclusive drift is corrected only under --fix. tech/* and comp/* stay open-ended and push-created. A milestone must already exist — push attaches, it never creates.

What crosses the boundary, and what does not

domain	tracker	note
id (slug)	<!-- kettle:id … -->	first line of the tracker-side body; stripped out of the local copy
title, body	title, body	verbatim, both ways, except the marker and the checkbox union
state	state	same vocabulary
labels	labels[]	names both ways
assignees	assignees[]	logins
milestone	milestone.title	resolved to an id on write
depends	native issue links	slugs here, {index, owner, repo} there; push writes them, a pull reads them back
—	ref	lands in branch:; push fills an empty one with the current git branch
—	number, html_url	land in gitea: / url:

depends: is always slugs, and the body's ## Depends on prose is passed through unchanged in both directions — a translator that edits prose churns the body on every round trip. The edge the tracker acts on is the native link, not the text, which is exactly why the text can be left alone. Push only ever adds links: a dependency deleted from depends: leaves its tracker link standing, and unlinking is a web UI job.

Dependencies go up in topological order so a blocker has its number before the issue that names it. One that is still local-only gets no link and is reported, never silently dropped.

Drift

There is none tracked, and there is very little left to track: a published issue has one copy — the tracker's — except while somebody is working on it, and that window closes at the next push. Nothing watches the tracker, nothing reconciles, nothing warns that a synced issue changed upstream. synced: says how old your working copy is, remote-updated: what the server said at that moment. The old question — "I edited this locally, does the server have it, whose text is newer?" — is answered by the store's contents rather than by a mechanism: a file that is here has not been pushed.

The checkbox union is not an exception. It reads only the two bodies in front of it; there is no base version and no way for it to report that anything diverged.

Payloads

Every request body goes to <project>/.kettle/payload/ first and is kept there for a retry or a post-mortem. It is a sibling of the store and never a child: request bodies are debris of the transport, and a scratchpad inside a store makes ls .kettle/issues lie about what exists. Nothing in it is anybody's only copy — deleting it costs nothing. A run that sends nothing leaves no directory behind.

For Gitea entities kettle has no command for — releases, webhooks, actions, pull requests — kettle api sends the request under this same login, into this same scratchpad. /kettle:api.

The commands themselves follow. Their usage lines, flags, defaults and examples are generated from the binary's own command registry, so they cannot disagree with the binary; kettle help <command> prints the same text. Editing them here changes nothing.

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.

kettle close <id|number> [<id|number>…]

close or reopen issues in the tracker, and on disk with them

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.

flag	default	what it does
--dry-run	false	print what would change; makes no request
--out	—	store root (default: /.kettle/issues)
--reopen	false	set the state back to open instead of closed

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

kettle comment <id>

post or edit a comment on a synced issue

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.

flag	default	what it does
--body	—	comment body inline (short, single-line)
--edit	0	comment id to rewrite, instead of posting a new one
--file	—	markdown file holding the comment body
--out	—	store root (default: /.kettle/issues)

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

kettle labels

put the canonical type/* and severity/* labels into a repository

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.

flag	default	what it does
--dry-run	false	print the plan; not one writing request
--fix	false	also patch colour/exclusive on labels that already exist
--repo	—	repository to bootstrap, as owner/name (default: this project's)

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

kettle pull [<key>…]

fetch issues from the tracker into the local store

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 .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.

flag	default	what it does
--cached	false	skip issues already on disk instead of refetching
--depth	3	how deep to follow blockers
--label	—	filter by label; repeat for AND
--limit	100	filter mode: how many issues to STORE, not to enumerate
--milestone	—	pull a whole milestone (id or title)
--no-deps	false	do not fill depends: and do not follow blockers
--out	—	store root (default: /.kettle/issues)
--q	—	search text in title and body
--query	—	the long spelling of -q
--state	open	filter mode only: open, closed or all

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

kettle push [<id>…]

send local issues to the tracker; the local copy goes with them

A SUCCESSFUL PUSH DELETES THE LOCAL FILE — .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 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.

flag	default	what it does
--dry-run	false	validate and print the plan; no network, nothing deleted
--force	false	push despite format violations
--out	—	store root (default: /.kettle/issues)
--update	false	PATCH issues that already carry a gitea: field

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

kettle remote

list what exists in the tracker, one line each

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.

flag	default	what it does
--label	—	filter by label; repeat for AND
--limit	30	how many lines to print
--milestone	—	milestone id or title
--out	—	store root (default: /.kettle/issues)
--q	—	search text in title and body
--query	—	the long spelling of -q
--state	open	open, closed or all

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

kettle sync-evict [<id>…]

refresh state from the tracker, then evict what is closed

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.

flag	default	what it does
--dry-run	false	ask the tracker and report; write and delete nothing
--out	—	store root (default: /.kettle/issues)

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