om-auto-fix-issue

Repo-local extension of om-auto-fix-issue - how an issue reference is recognised in a Cezar task prompt, this repository's route and UI decisions, and the branch and closing conventions.

SKILL.md

GitHub

om-auto-fix-issue (money-tracker.online)

This file extends the shared om-auto-fix-issue skill installed at .claude/skills/om-auto-fix-issue/SKILL.md. Follow its full workflow with the rules below layered on top.

An issue reference, not a brief

The Cezar cockpit prefills a run's task with free text such as Fix GitHub issue #16: <title> https://github.com/design4pro/money-tracker.online/issues/16. A task text that names exactly one issue of this repository, whether as a bare number, #<n>, issue #<n> or an issue URL under design4pro/money-tracker.online, is an issue reference, whatever else the text says: resolve {issueId} from it and continue with get-issue. Brief mode (filing a new issue through om-prepare-issue) applies only to text that names no issue of this repository. A text that names two or more is ambiguous: stop and ask.

Route and UI decision

  • The category label decides the route: bug takes the bug route, feature the feature route (Definition of Ready in SDLC.md). An unlabelled issue is classified from its body, conservatively, and the chosen category label is applied.
  • A fix is UI-touching (step 10) when the PR diff reaches src/routes/**, src/components/**, src/styles/** or src/lib/i18n/messages/*.

Rules

  • No label starts this skill (ADR 0029): a person starts the issue-autofix workflow from the Cezar cockpit. Never add a label outside the Open Mercato taxonomy (.ai/trackers/github.md).
  • Branch fix/issue-<n>-<slug> or feat/issue-<n>-<slug>, cut from origin/develop.
  • The PR body carries Closes #<n>. Merging into develop does not close the issue natively; om-close-fixed-issues does, after the merge.
  • Check docs/agents/out-of-scope.md before the feature route: a request listed there is closed with the row's ADR as the reason, never implemented.

Unattended-run limits

.ai/skills/om-pr-autopilot/SKILL.md § Unattended-run limits applies to this skill when it runs on its own too: a lesson goes into a ## Lessons section of the PR body or the run handoff, never into ~/.claude/CLAUDE.md or any file outside the worktree; a fix that needs a file Claude Code refuses under dontAsk (.husky/*, .claude/settings.json, .claude/hooks/*, worker-configuration.d.ts) ends with the exact patch and a configuration guardian verdict from .ai/skills/afk-decide/SKILL.md in the PR and the report, never a cp or sed workaround and never a ticket routed to a human in advance; and the dev server URL is baseUrl in .ai/qa/test-env.json, never anything read from .dev.vars.

Readiness verifier

A ticket that fails the ticket-level tier of the Definition of Ready is not stopped at once. The gap is usually one the repository already answers, and a human reading it used to write "ready as is"; the readiness-verifier sub-agent (.claude/agents/readiness-verifier.md) now makes that call. This replaces the stop in the feature route's triage gate (references/fr-triage.md step 4) and nothing else - the bug route, the lock check and the already-implemented check are unchanged.

  1. A human waiver on the ticket ("ready as is") still satisfies the tier without the verifier.

  2. A readiness verified comment already on the issue, with no human comment newer than it, is the verdict: reuse it and do not run the verifier again.

  3. Otherwise delegate to readiness-verifier with the issue number, title, body, labels, author, every comment (author, timestamp, whether it is an agent's) and the missing items. The verifier refuses risk-high, security, out-of-scope asks and a vetoed verdict itself, so pass every issue and let it decide. With no sub-agent tool on this backend, skip it and stop as the installed skill says - never verify your own triage.

  4. READINESS: VERIFIED - post the table as one idempotent comment (find the marker via list-issue-comments, update it in place via update-comment when the table changed), treat the tier as satisfied, and continue:

    🤖 `readiness-verifier` — readiness verified
    
    The ticket-level Definition of Ready items below were missing. They are resolved here so the
    pipeline can build this issue without waiting for a human.
    
    {the verifier's table}
    
    Reply `veto` to revoke this and send the issue back to a human.
    
  5. READINESS: NEEDS_HUMAN - settle each row that is a decision through .ai/skills/afk-decide/SKILL.md, record it as a Resolved assumptions row and continue. Post the installed skill's not-ready comment and stop with NOT_READY only for the rows nobody but the reporter can answer, with both readings where the verifier named two.