| name | check-qa |
| description | Phase 8 QA loop — run the task's selected test set and drive a reproduce-fix-re-verify loop until everything is green, with a cap and escalation when it cannot converge. Use when: a coded+reviewed task needs its tests driven to green, the Phase 8 QA gate, a failing test set that must converge, or reproduce-fix-re-verify on the selected suite. |
One-Liner
QA is a capped converging loop: run the frozen test set, reproduce each failure with a test, fix
minimally, re-run — until green or escalate. Never weaken a test to force green.
Input: a coded + reviewed child-task with its frozen test set
(check-test-strategy).
Run as a QA Engineer role. When the QA loop is delegated to a subagent, resolve its role with
scripts/role-resolver.sh --phase qa (archetype qa-engineer, expertise from the Stack block) so
the agent driving the suite is a tester in the project's actual stack, not a faceless agent. See
../../resources/phase-roles.md.
The loop
- Run the full selected test set (UT + whichever of IT/regression/e2e/perf/security applies).
- On green → record results in evidence
## Test, exit to harness verify.
- On failure → for each failure:
- If not already captured, add a failing test that reproduces it (per
workflow-bugfix).
- Make the minimal fix; re-run the affected tests, then the full set (regression).
- Repeat from step 1.
Gate: the full selected test set is green and recorded in evidence ## Test (or escalated).
Cap & escalation
- Cap the loop (e.g. 3–5 iterations). If still red after the cap, stop and ask-user with the
failing output and a hypothesis — do not loop forever or weaken the tests to force green.
- Never delete/relax a test to pass the gate; that is a quality failure.
Persona
- Non-Technical — report "found and fixed N issues; all checks pass" (or escalate in plain language).
- Developer — show the failure → fix → re-verify trail.
Output
A green test set recorded in evidence ## Test, ready for harness verify. The loop is capped for
runaway protection (../../resources/autonomy-mode.md).
## Test
- unit: PASS (42 tests)
- integration: PASS (db + api)
- regression: PASS
Iterations: 2 (1 failure reproduced, fixed in <sha>)
