agentleFS
talas9talas9/anti-hall/.claude/skills/gh-work/SKILL.md

gh-work

Skill2 starsChanged 46 days agoMITView on GitHub

Dev-only, anti-hall repo (not part of the shipped plugin). Use when starting, picking, triaging, branching, opening a PR for, or closing any piece of anti-hall work - the issue -> branch -> PR -> merge lifecycle on GitHub and the "anti-hall roadmap" board, with exact gh commands and board field ids.

  • Commits and pushes

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    All anti-hall work lives on GitHub: tasks, bugs, features, plans and the roadmap.

    line 8
  2. Very specific:

    Every agent or lane brief names its issue #n.

    line 15
  3. Very specific:

    Stop only at owner-only gates (billing, secrets, destructive actions): comment on the issue, label it status:blocked, set board Status …

    line 33
  4. Very specific:

    Every commit message references #n: fix(engine): <summary> (#n).

    line 153
  5. Distinctive:

    Do not post for each commit, test run, retry or tool call.

    line 187
  6. Distinctive:

    Prefer editing your last comment of the same kind over adding another when only a detail changed.

    line 189

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

gh-work: issue -> branch -> PR -> merge

All anti-hall work lives on GitHub: tasks, bugs, features, plans and the roadmap. The session task list is built FROM the board, never the other way round. Repo talas9/anti-hall; board "anti-hall roadmap" = user project 3, owner talas9. Text on GitHub stays agnostic (public repo): no private paths, names, emails or session ids.

HARD RULE: every agent and lane is tracked on its issue

  • Every agent or lane brief names its issue #n.
  • At start: move the item to In progress and comment the approach.
  • At milestones and at the end: comment evidence (commit SHAs, test names and results).
  • A paused or stopped lane gets a state comment immediately.
  • PRs use Closes #n.

Autonomous operating loop

  1. Session start / resume: sync the board. gh project item-list 3 --owner talas9 --format json --limit 500 (the Now view first, then Next by priority; Backlog, By milestone and Board are the other views). Rebuild the session task list from it; every task subject starts with #n.
  2. Anything new (owner request, idea, bug, plan, or something found while working): open the issue FIRST (step A below), triage it immediately (step B), then work. Plans go in the issue body or a linked comment; a local plan file carries a link to its issue.
  3. Pick the next work yourself: the highest-priority Now item with no in-flight agent, else the top of Next. Move it to In progress, comment the approach, work it, comment the evidence, close it through a PR with Closes #n. Never wait for permission on accepted items. Stop only at owner-only gates (billing, secrets, destructive actions): comment on the issue, label it status:blocked, set board Status Blocked, take the next item.
  4. Every few hours: re-sync the board, triage everything new (bot suggestions included), post a one-line progress comment on each status:in-progress issue.

A. Find or open the issue

gh issue list -R talas9/anti-hall --state open --search "<keywords>" --json number,title,labels

Forms (.github/ISSUE_TEMPLATE/, blank issues are off): Bug report (bug: title, labels bug,type:bug,status:triage), False positive (false-positive), Feature request (feature: , labels enhancement,type:feature,status:triage). The CLI cannot fill a form, so write the body in the rendered form layout (### <Field label> + blank line + value); the triage.yml workflow parses the Priority, Estimate, Area and Estimate in hours headings:

gh issue create -R talas9/anti-hall --title "feature: <summary>" \
  --label enhancement,type:feature,status:triage --milestone "v0.300.0" --body "$(cat <<'EOF'
### Problem

<what is wrong or missing>

### Proposed behaviour

<what should happen>

### Which port

Both

### Area

engine

### Priority

P2 (normal)

### Estimate

M (up to 1 day)

### Estimate in hours

6

### Acceptance criteria

- [ ] <checkable outcome>
EOF
)"

Area values: engine, devswarm, hooks, statusline, codex, ci, docs, release. Priority values: P0 (critical, blocks users or a release), P1 (high, next release), P2 (normal), P3 (low, nice to have). Estimate values: S (up to 2 hours), M (up to 1 day), L (up to 3 days), XL (more than 3 days).

B. Triage (decide in the same session)

Labels: type:feature|bug|docs|chore, priority:P0|P1|P2|P3, size:S|M|L|XL, area:*, status:triage|accepted|in-progress|blocked. Milestones: v0.300.0 (engine decides hooks, Node is the fallback), v0.301 engine-only (Node removed), v1.0 (no Node, everything through the engine, contract frozen).

Accept:

gh issue edit <n> -R talas9/anti-hall --remove-label status:triage \
  --add-label status:accepted,type:feature,priority:P2,size:M,area:engine --milestone "v1.0"
U=https://github.com/talas9/anti-hall/issues/<n>
gh project item-add 3 --owner talas9 --url "$U"          # no-op if the board already has it
gh project item-edit 3 --owner talas9 --url "$U" --field Status   --value Accepted
gh project item-edit 3 --owner talas9 --url "$U" --field Priority --value P2
gh project item-edit 3 --owner talas9 --url "$U" --field Size     --value M
gh project item-edit 3 --owner talas9 --url "$U" --field Estimate --number 6
gh project item-edit 3 --owner talas9 --url "$U" --field Target   --text "v1.0"

Defer: do not accept it (it stays out of Now/Next); comment why and what would change the decision (one comment: gh issue comment <n> --body "Deferred: <reason>"; the status itself goes on the labels and board). Never close or delete an issue automatically.

Board field and option ids

From gh project field-list 3 --owner talas9 --format json (re-run it if an edit fails; ids change only when a field is recreated). Use them with the id form, one field per call:

ITEM=$(gh project item-list 3 --owner talas9 --format json --limit 500 \
  --jq '.items[] | select(.content.number==<n>) | .id')
gh project item-edit --project-id PVT_kwHOABnoa84BmYbQ --id "$ITEM" \
  --field-id PVTSSF_lAHOABnoa84BmYbQzhlBtY0 --single-select-option-id dbf76781   # Status = In progress
FieldField idTypeOptions (name = id)
(project)PVT_kwHOABnoa84BmYbQproject node id
StatusPVTSSF_lAHOABnoa84BmYbQzhlBtY0single selectTriage = 81ce65fd, Accepted = 49e0e02f, In progress = dbf76781, Blocked = 6aabb99f, Done = 6c3416e5
PriorityPVTSSF_lAHOABnoa84BmYbQzhlBtd0single selectP0 = bcde1ac9, P1 = c806d2de, P2 = 0dfd336f, P3 = fa6760c3
SizePVTSSF_lAHOABnoa84BmYbQzhlBtewsingle selectS = f7a60b7e, M = 44ad0140, L = dae84bf6, XL = 96a5febd
EstimatePVTF_lAHOABnoa84BmYbQzhlBtfsnumber (hours)--number
TargetPVTF_lAHOABnoa84BmYbQzhlBtfwtext--text (milestone name)
MilestonePVTF_lAHOABnoa84BmYbQzhlBtZAbuilt-inset on the issue (gh issue edit --milestone)

Status mirrors the status:* label: keep both in step whenever either changes.

C. Start work

gh issue edit <n> --remove-label status:accepted --add-label status:in-progress
gh project item-edit 3 --owner talas9 --url "$U" --field Status --value "In progress"
gh issue comment <n> --body "Approach: <plan, lanes, risks>"      # milestone 1 of 3, see "Issue comments"
git fetch origin && git switch -c <type>/<n>-<slug> origin/dev

Branch <type>/<n>-<slug> (type = feat, fix, docs, chore, ci, refactor, perf, test) off dev. Every commit message references #n: fix(engine): <summary> (#n). Commit as the repo's git config identity; no AI-credit trailers (git-guard blocks them). Never force push, never rewrite a pushed branch.

D. Pull request into dev

git push -u origin <type>/<n>-<slug>
gh pr create -R talas9/anti-hall --base dev --head <type>/<n>-<slug> \
  --title "fix(engine): <summary>" --body "$(printf 'Closes #%s\n\n<what changed, how it was verified>' <n>)"
gh pr checks <pr> --watch
gh pr merge <pr> --merge          # merge commit; never --squash for dev -> main
  • Title: Conventional Commits type(scope): summary (pr-title.yml enforces it on PRs to dev and main; release is also an allowed type).
  • Body: Closes #n so the merge closes the issue and the board moves it to Done.
  • Issue comments follow the milestone rule below. Status changes (accepted, in progress, blocked, done) go on the labels and the board Status field, never in a comment.
  • Blocked: gh issue edit <n> --add-label status:blocked, board Status Blocked; one comment saying what blocks it (and what unblocks it) is allowed, because the blocker is information, not status.

Issue comments (agents)

Comment on your issue only at milestones, one concise comment each, no step-by-step chatter:

WhenComment starts withContent
StartApproach:The plan, lanes and risks, in a few lines.
Meaningful progressDone: then Left:One line each: what is finished, what remains.
EndEvidence:Commit/PR shas, test or replay output, the acceptance criteria checked.
  • The board "Progress" field is read from the newest trusted comment's Done: / Left: line, so keep those two lines short and current; do not post progress prose elsewhere.
  • Do not post for each commit, test run, retry or tool call. If nothing changed in Done or Left, do not comment.
  • Prefer editing your last comment of the same kind over adding another when only a detail changed.
  • The repo's own bots keep ONE sticky comment per purpose (marker <!-- ah-bot:<purpose> -->, edited in place with an "Updated <date>" footer) and stay silent on the owner's own issues and PRs, except the PR check, delivered-on-dev, and replies to /triage, /explain, /convert.

E. Release PRs (dev -> main)

main accepts PRs only from this repo's dev (pr-source.yml check dev-only), and needs tests-passed (test.yml). Merge with a merge commit (or fast-forward), never squash, so dev and main stay in step:

gh pr create -R talas9/anti-hall --base main --head dev --title "release: v<version>" --body "<CHANGELOG section>"
gh pr merge <pr> --merge

The full release sequence is the release skill.

F. Bot outputs and what to do

SignalSourceAction
status:triage + priority:*/size:*/area:*triage.yml (form fields)Confirm or correct, then accept or defer (B).
triage brief commentcommunity.yml (rules + model first pass)Read the comment as a suggestion; verify against code before labelling.
area:* on a PRtriage.yml + .github/labeler.yml (paths)Nothing; fix labeler.yml if wrong.
triage suggestion commentmoderation/triage workflowsApply what the evidence supports, reply on what you reject, then remove the suggestion state.
moderation:reviewmoderation workflowA human-review flag (possible spam, abuse, duplicate). Read it; act only on clear evidence; never delete content, hide only spam/abuse with a comment why; remove the label once handled.
needs-issuePR moderation (PR has no Closes #n; dev head is exempt)Open or find the issue, add Closes #n to the PR body, remove the label.
risk:workflow, risk:securityPR moderation (touches workflows/scripts or guards/hooks/security files)Extra review: read the full diff, run a deadly-loop on it, never auto-merge; note the review in the PR.
stalestale.yml (60 days idle; never closes)Comment whether it still matters; a human closes.

Bots never close, lock or delete; neither does a session without the owner's OK.

Model routing for agents and lanes: see the "Model routing" section in .claude/skills/README.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.