ISSUE_ROOT was the relative `tmp/issues`, so "the store" was whatever
directory the shell happened to be standing in. It is the --out default
in all eight scripts of both layers, which made one `cd` — and a `cd`
outlives the command that ran it — enough for readers to report an empty
store on a full one and for writers to quietly build a second store
beside the first. `issue_index.py` run from inside tmp/issues left
tmp/issues/tmp/issues/ behind and exited 0.
The anchor is issue.py's own __file__, not cwd. A script's location is a
fact about the installation; cwd is a fact about the last `cd`, and the
scripts are invoked by path from wherever the agent happens to be. From
there `store_root()` walks up to the nearest repo marker — `.git`
(exists(), not isdir(): a worktree's .git is a file) or AGENTS.md for a
copy taken out of git — and joins tmp/issues. Markers rather than a
fixed number of `..` hops, because the layout is not a promise. cwd is
tried only if the scripts are not inside a repository at all.
The function lives in the domain layer and skills/sync imports it, so
both layers agree by construction — the direction the layering rule
allows. skills/issue stays stdlib-only.
An explicit --out still wins and is used exactly as typed: a relative
--out stays relative to cwd, because that is what the operator asked
for. No new environment surface.
Two consequences the issue also asked for:
- Missing is no longer reported as empty. `store_error()` returns one
message for a path that is not there and another for a store with no
issues in it.
- Nothing conjures a store as a side effect of a write. save() and
issue_index.build() require it instead of os.makedirs'ing it; only
issue_new.py and pull.py create one, and both say so on stderr.
Establishes tests/ — plain stdlib unittest, no pytest, no dependencies.
The store tests build a throwaway repo in a TemporaryDirectory (a .git
marker, a copy of both script layers, fixture issues) and run the real
scripts inside it as subprocesses from five different working
directories; tmp/issues/ is never touched. Against the pre-fix scripts
15 of the 21 fail, reproducing the report exactly — five stray stores,
including tmp/issues/tmp/issues.
python3 -m unittest discover -s tests -v
Closesclaude-skills/tea#15
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The skills carry meaning, the scripts carry work. Splitting the second
half onto a cheap model keeps the main session's context for the part
that needs judgement.
tea-runner is a Haiku subagent with Bash/Read/Grep/Glob/Skill and
nothing else. It loads /tea:sync or /tea:issue for the command table
rather than carrying its own copy, so the skills stay the single source
of truth for the script surface.
It executes and reports; it decides nothing. No Edit and no Write, so an
issue body is out of reach. No raw tea, no --force, no closing or
retitling, no pushing past the set it was handed, one retry maximum. A
failed validation, a missing type, an unpushed dependency come back as a
question in a `blocked:` line. The reply is a fixed receipt — commands
with ok/FAIL, touched paths, stderr verbatim — never a payload dump.
Knowledge still flows one way: nothing under skills/ knows the agent
exists, and deleting agents/ changes nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Five tracker issues, all in the bridge layer except the last.
pull.py fetches comments by default (#6). The thread was reachable only
through --comments, and only for a single issue, so a bulk pull left every
local copy silently incomplete: a missing <id>.comments.md could mean "no
comments" or "never asked". Now every written issue gets its thread, in key
and filter mode alike; an empty one costs no request (the count rides in the
list payload) and writes no file, and a file left over from an earlier pull
is deleted. --cached skips the thread along with the body. The --comments
flag is gone.
labels.py bootstraps the canonical label set (#7). Labels used to appear as a
side effect of the first push that happened to use them, so a repo could not
be filtered by type/bug until somebody pushed a bug. The set is finite and
already described by the domain taxonomy — 6 type/* and 5 severity/* — which
makes it a run, not a decision. Names and exclusivity come from issue.TYPES /
SEVERITIES / EXCLUSIVE_NS, colors from map.label_specs; no list is duplicated.
An exact name is never re-created or patched. Lookalikes (bug, Bug, "type:
bug", kind/bug) are reported with their id and left alone — renaming somebody
else's label is a decision, not a migration. Color or exclusive drift is
printed, and changed only under --fix.
branch: carries Gitea's ref (#8). map.to_payload sends ref only when the field
is non-empty, since ref="" would clear whatever the server has; from_api reads
it back; push fills an empty one from `git rev-parse --abbrev-ref HEAD` and
writes it into the issue file. A hand-written value is never overwritten, on
create or on --update. Detached HEAD and running outside a repo warn and send
no ref. Reading the branch is the only thing these scripts ask of git. The
domain needs no change: unknown keys already ride in Issue.extra and render
after the domain fields.
Bulk pulls no longer store closed issues (#10). Filter mode wrote every
payload the server returned, so --state all dragged the closed backlog into a
store that gets read whole — INDEX.md, grep over tmp/issues/*.md. They are
still enumerated, the number left out goes to stderr, and an issue already on
disk is refreshed either way so the local copy learns it was closed instead of
staying open forever. --state closed stores them, and key mode is exempt: an
address is not a bulk read.
/tea:issue gains a "Writing a proper description" procedure (#9). Six steps
from reading an issue to issue_check.py, the rule that a missing fact is found
in the repository or asked about rather than invented, and the note that the
procedure is identical for origin: local and origin: gitea while delivery to
the tracker belongs to /tea:sync. No new script.
Verified: labels.py run for real against claude-skills/tea (9 created, 2
already present) and idempotent on a second run; pull.py exercised live for
the closed-skip, --state closed, key-mode and comment paths; the push write
path covered offline with the transport stubbed. skills/issue/scripts/ still
imports stdlib only, with no subprocess.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
An issue was a Gitea row that happened to be cached locally: its identity
was the tracker's number (42.md), its dependencies were tracker numbers
(depends: [#12]), and a local issue existed only as a draft that push
deleted on success. Nothing could be planned or tracked without a tracker.
Split into layers, with knowledge flowing one way:
skills/issue DOMAIN what an issue is: format, validation, dep graph
^ offline; stdlib imports only, no subprocess
| imports
skills/sync BRIDGE map.py md <-> Gitea JSON, pure, no I/O
_gitea.py login pin, api, pagination, filters
skills/use REFERENCE tea CLI docs for non-issue entities
skills/issue never imports skills/sync. Delete the sync layer and the
domain keeps working.
Identity is now a slug derived from the title (wire-sqlc-appclick.md) and
is stable across retitles and pushes. Tracker numbers live in a `gitea:`
field, never in a file name and never in `depends:`; the pair is indexed
in .remote.json, which is a cache over the files, not a second source of
truth.
Behavior changes:
- Pushing is additive. The file is never deleted; it gains gitea:/url:/
synced: and origin: flips from local to gitea. `origin: local` is a
durable state, not a pending one.
- Pushes go in topological order so dependencies get numbers first.
- The dependency graph is computed offline from `depends:` metadata; body
prose is passed through unchanged in both directions rather than being
rewritten between slugs and #N.
- `origin` is domain-owned (whether work exists elsewhere is a fact about
the work); the handle and how to reach it stay with sync.
Script moves:
issue_get.py -> sync/pull.py
issue_push.py -> sync/push.py
issue_list.py -> sync/remote.py
issue_index.py -> issue/issue_index.py
_tea.py -> split into issue/issue.py, sync/map.py, sync/_gitea.py
New: issue/issue_new.py, issue/issue_check.py, issue/issue_tree.py, and
sync/comment.py — comment posting was the last issue operation still
hand-rolled through raw `tea api`.
references/issue-format.md moves to skills/issue/references/format.md;
label hex colors move out of it into map.py, since a color is how a
tracker paints a chip, not what an issue is.
Verified: offline path end to end (new, check, tree, index, push
--dry-run) and read-only against Gitea (remote listing, pull with
mapping, comment guard). Write paths of push.py and comment.py are not
exercised here.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Replace fetch_issue.py with four scripts around a flat, greppable cache in
tmp/issues/. Planning stays offline and issues reach Gitea in one push:
- issue_get.py: fetch by key or by filter (--milestone/--label/-q). The list
endpoint carries issue bodies, so a whole milestone costs one request per 50
issues. Gitea silently ignores an unresolvable milestones= filter and returns
the entire backlog, so the milestone is resolved up front and every returned
issue is re-checked locally. --deps walks the dependency graph downwards via
the structured sections plus native dependencies and writes tree-<slug>.md.
- issue_push.py: validate a local draft against the canonical format, create
missing labels with the right colors and exclusivity, POST, delete the draft.
- issue_list.py: discovery to stdout, writes nothing.
- issue_index.py: rebuild INDEX.md from what is on disk.
Files use one metadata field per line with inline lists so plain grep works
without a parser. This is a cache and a drafting area, not a mirror: no drift
tracking, no sync back.
Projects are not fetchable — the projects API is 404 on Gitea 1.26; documented
alongside the milestone caveat.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
agents-sync recreates CLAUDE.md -> AGENTS.md in every directory on each
Bash call, so tracking them adds churn and duplicates content that
already lives in AGENTS.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Match the Issues-Workflow wiki (claude-skills/tea#1): types expand to
bug|task|refactor|test|feature|draft (feature becomes a container,
task takes over new functionality), add severity/tech/comp label
namespaces, an optional "Depends on" section, and templates for
test and feature. Milestone/Project containers documented.
Refs claude-skills/tea#1
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Raw `tea issues -o json` / `tea api` dumps the full payload (avatars,
nested users, every comment) into the model context. The script writes
trimmed markdown to tmp/issue/<n>/ and prints only a compact index;
login comes from the operator pin, never an argument.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Keep AGENTS.md as the single source of agent docs: agents-sync runs
before every Bash command and makes each directory canonical —
AGENTS.md real file, CLAUDE.md a symlink to it. Renames a lone
CLAUDE.md, creates or re-points symlinks, replaces identical
duplicates; differing files are only reported, never merged. Fails
open so it can never block a command.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
`/tea:login` could not be invoked as `/login` because Claude Code
reserves that name as a built-in command. Renaming the skill to `auth`
frees the short form: `/tea:auth` and `/auth` both resolve.
Updates all references in the guard hook, README, plugin manifests, and
the use skill.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The install steps used a non-existent "plugins" settings key, so a fresh
clone loaded nothing. Add .claude-plugin/marketplace.json and document the
supported /plugin marketplace add + /plugin install tea@tea flow.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The previous guard required $GITEA_LOGIN to be set in the Bash environment,
which (a) only happens after a session restart and (b) let Claude name any
login it liked as long as one was set. Two failures: pinning needed a restart
to take effect, and Claude could pick the wrong identity from memory.
Rewrite the guard (now python3 for JSON in/out) to RESOLVE the login itself:
- Claude must write the literal placeholder --login "$GITEA_LOGIN".
- The hook reads the operator's pin from .claude/settings.local.json
(env.GITEA_LOGIN) at call time — from the FILE via CLAUDE_PROJECT_DIR/cwd
walk-up — and rewrites the command to that literal via updatedInput.
- A literal login, another variable, an empty value, or a missing --login are
all blocked: Claude may not choose the identity, only the operator may.
- No pin -> block with a pointer to /tea:login.
Effect: pinning works in the same session (no restart), and Claude can no
longer act under a login it picked. /tea:login now mandates an explicit
operator choice (AskUserQuestion), never inferring from memory. /tea:use
documents the placeholder-only contract.
Guard unit-tested across 13 rewrite/block/passthrough cases incl. -l,
--login=, ${...}, compound+walkup+pipe. claude plugin validate passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Convert the standalone `tea` skill into a skills-dir plugin so commands are
namespaced and an enforcement hook can ship with it:
- /tea:login — pin the project Gitea login into .claude/settings.local.json
- /tea:use — tea CLI reference (was the old root SKILL.md), with the
login rule slimmed since the hook now enforces it
- hooks/tea-guard.sh — PreToolUse(Bash) guard: blocks any `tea` command that
touches Gitea unless it carries --login and $GITEA_LOGIN is set. Exempts
`tea logins list` and `tea --version/--help` so /tea:login can bootstrap.
References moved under skills/use/references/. `claude plugin validate` passes;
guard unit-tested across allow/block cases.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
naudachu
naudachu