om-auto-manage-issues

money-tracker.online extension of the installed om-auto-manage-issues skill. Human-hold labels, category mapping, the readiness verifier and the hand-on to afk-triage.

SKILL.md

GitHub

Extends the installed skill: read .claude/skills/om-auto-manage-issues/SKILL.md and follow it. On top:

  • Human-hold labels (skip the issue entirely): blocked, do-not-close, do-not-merge.
  • Category comes from the issue form (.github/ISSUE_TEMPLATE): a bug report is bug, an enhancement is feature, a maintenance item is refactor; Dependabot is dependencies.
  • The Definition of Ready is in SDLC.md; the not-ready comment names the missing ticket-level items and points at the issue form section that holds each.
  • Run the readiness verifier before recording not-ready. When step 6 of the installed skill (references/enrich-existing-issue.md) finds a ticket-level item missing, follow .ai/skills/om-auto-fix-issue/SKILL.md § Readiness verifier: READINESS: VERIFIED records READY_STATUS=ready and posts the readiness verified comment instead of the not-ready one; a NEEDS_HUMAN row that is a decision is settled through .ai/skills/afk-decide/SKILL.md (see "Settle, do not ask" below); only a row nobody but the reporter can answer records not-ready, and the not-ready comment lists only those rows. A verified issue is ready for every rule below.
  • Hand on by doing nothing. In the issue-groom workflow afk-triage runs right after this skill, in the same run (ADR 0029), and turns a ready groom into an Agent Brief and a not-ready one into a brief or a blocked hold. Add no label to hand the issue on: the Open Mercato labels are the whole vocabulary (.ai/trackers/github.md). On READY_STATUS=not-ready the not-ready comment is the whole output of this step.
  • Never groom a Sentry issue. A body with a <!-- sentry-issue: marker was filed from production error data, and the maintainer's login on it does not make it trusted, so the author gate does not stop it. Stop with NO_ACTION_NEEDED before any mutation: no automation grooms, briefs or implements it (ADR 0029, .ai/scripts/backlog-status.mjs).
  • A spec parent needs no brief. An issue whose body starts with Spec: .ai/specs/ is a parent: afk-to-tickets slices it into tickets that carry their own Agent Brief, and afk-triage stops on it. Groom it for labels and readiness as usual, and never leave that decision to a human in the report or a comment.
  • Settle, do not ask. Grooming runs unattended, so the installed skill's "the not-ready comment names what a human must add" is narrowed here to what only the reporter knows (a repro, a version, what they expected). Everything else is settled in this run through .ai/skills/afk-decide/SKILL.md and never makes the issue not-ready:
    • A missing Out of scope section is written by this run: ask the product owner specialist for it, append it to the issue body with update-issue under an ## Out of scope heading marked as agent-written, and record the verdict as a Resolved assumptions row in the groom comment.
    • An open question (product, design, scope, naming) is put to the fitting specialist and recorded as a non-blocking Resolved assumptions row (# | Question | Applied default | Why) in the groom comment. It is never addressed to the issue author and never makes the issue not-ready.
    • "A human's call", "decided by a maintainer" or similar wording copied into a ticket from a spec or an earlier comment is an open question like any other, never a blocker: settle it the same way.
  • Implementation notes must name the owner module for each concept (docs/SPEC.md §8) and warn when the change would touch a protected surface in BACKWARD_COMPATIBILITY.md. An issue whose ask is listed in docs/agents/out-of-scope.md is reported as out of scope with the row's ADR, never groomed into a ticket.