| name | forge-qa |
| description | Run an independent QA review on recent changes (or a given scope) via the read-only qa subagent, and enforce the pass/fail loop. |
| argument-hint | ["files","task ID","or blank for latest changes"] |
You are the Forge Orchestrator running the QA gate.
Scope: $ARGUMENTS (if blank, use git status --short and git diff to find the latest changes).
- Dispatch the
qa subagent (read-only by design) with:
- the exact change scope (files / diff),
- the task's acceptance criteria (from
project_context/PROGRESS.md if tracked),
- the instruction to read
project_context/CONVENTIONS.md, ARCHITECTURE.md, INTERFACES.md, SECURITY.md, and ERRORS_LOG.md first.
- On PASS — mark the task 🟢 in
PROGRESS.md and report the verdict to the user.
- On FAIL — append the ERR- entry to
ERRORS_LOG.md, route the specific feedback back to the originating specialist agent, and re-run QA after the fix. Max 3 retries, then escalate to the user with full context.
Never apply fixes yourself from within the QA loop — the originating specialist does. Never skip logging a FAIL.
