qa-triage

fatmuh/agentic-workflow-skills@main

SKILL.md

/qa-triage

A queue nobody sorts decays into a wall of maybe. This skill walks the tracker's open queue and moves every incoming item to exactly one actionable state, leaving a trail on each item so the next session resumes instead of re-asking. It is a QA run: it starts from nothing, re-derives its context from the repo, the tracker config, and the docs, and trusts no claim just because a previous session wrote it down.

Two commitments hold for the whole run:

  • Every processed item ends in exactly one state of the five-label vocabulary, and the state change and its reason land on the item together. A label without its recorded reason is half an action.
  • Nothing is ever closed silently. The one state that removes an item from the open queue (wontfix) always carries its reason in a note the reporter will read.

Triage observes and routes; it never fixes. Verification may run the repo's tests or a reproduction, but no source is edited during the run.

1. Read the tracker config

Everything downstream is configured, never assumed. Read docs/agentic-skills/tracker.md and take three facts from it:

  • The tracker kind — one of github, gitlab, local-markdown, freeform. It decides every read and write operation (section 6).
  • The label mapping — the five canonical roles (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix) against the label strings this tracker actually carries. Speak the canonical roles in every note and report; apply the mapped strings when you touch the tracker. If the config is missing, or names no kind, stop and point at the setup skill (/agentic-skills-init): triage without config would be guessing where issues even live.
  • The PR-surface flag — whether pull requests count as incoming requests alongside issues.