QA

/qa

The active-QA skill. Runs the repo's test suite, parses failures, applies targeted fixes for common patterns (snapshot drift, lint nits, type errors, missing imports), re-runs, and reports a clean pass or a structured stuck-state. Distinct from /qa-only, which never edits.

When to use

• You just finished implementation and want a clean test pass before /review or /ship
• CI is red and you want to reproduce + fix locally before pushing
• After a refactor where snapshots and fixtures need to catch up

When NOT to use

• Tests have never passed on this branch — first establish a green baseline manually
• Failures are clearly product-logic bugs — use /investigate instead (hypothesis-driven, no auto-fix)
• Production deploy gate — use /qa-only (read-only, no edits)

Inputs

• Optional --scope <path> — run only tests matching the path (default: full suite)
• Optional --no-fix — report failures, do not attempt fixes (same shape as /qa-only)
• Optional --max-iterations <N> — stop after N fix+rerun cycles (default: 3)

Workflow

1. Detect test runner — read package.json / pyproject.toml / Cargo.toml / equivalent. Pick the canonical script (npm test, pytest, cargo test). If multiple: ask via AskUserQuestion.
2. Baseline run — execute, capture stdout/stderr + exit code. If clean: report + exit.
3. Failure classification — group failures into buckets:
   • Snapshot drift (stable diff, no logic change) → auto-fix candidate
   • Lint / format nit → auto-fix candidate
   • Type error from new code → auto-fix candidate (add missing import, fix obvious typo)
   • Assertion failure (real product behavior) → DO NOT auto-fix, escalate
   • Flaky / timeout → retry once, then escalate
4. Auto-fix cycle — for each auto-fix candidate, apply the targeted Edit. Group fixes into one commit-able batch. NEVER touch product code without explicit operator confirmation.
5. Re-run — repeat steps 2-4 up to --max-iterations.
6. Final state — either CLEAN PASS or STUCK with structured report.

Report format

QA Status: <branch>

Runner: npm test
Iterations: 2/3
Initial failures: 7
Auto-fixed: 5
  - 3× snapshot drift (src/components/case/__snapshots__/)
  - 1× lint (src/lib/dlxClient.ts: unused import)
  - 1× missing import (src/hooks/useCaseBrain.ts: useEffect)
Remaining: 2

## Stuck — escalating
[FAIL] tests/billing.test.ts:42 — expected 100, got 200
  Likely cause: real product logic — refund doubles amount
  Recommendation: /investigate

[FAIL] tests/e2e/auth.spec.ts:18 — timeout waiting for #login-btn
  Likely cause: flaky — but failed twice in a row, suspect real
  Recommendation: read playwright trace at .playwright/trace.zip

Compliance integration

• Sanity-scan on every Edit before applying (no secrets/customer-data in fix payload).
• If auto-fix would touch a frozen-zone path (per project CLAUDE.md): block + escalate.
• Audit-log every auto-fix mechanically (Layer 2 traceability): source "${LINTEL_SOURCE_ROOT:-$(git rev-parse --show-toplevel)}/bin/_audit.sh"; audit_log qa-fixes auto_fix file=<path> fix_kind=<lint|snapshot|import|assertion> → .claude/runtime/audit/qa-fixes.jsonl.

Failure modes

• No test runner detected: report + ask operator to declare via package.json scripts or ~/.lintel/qa.yaml.
• Runner crashes (not test failure, runner itself): report + exit. Do not retry blindly.
• Max iterations hit with failures remaining: STUCK state — surface full failure list + recommendations. Operator chooses next move.
• Auto-fix introduces a NEW failure: revert the fix, mark that failure non-auto-fixable, continue with remaining.

Examples

Clean pass:

> /qa
✓ All tests pass (npm test, 1 iteration, 142 tests / 0 failures)
Ready for /review.

Auto-fix success:

> /qa
Iteration 1: 7 failures
Iteration 2: 0 failures (auto-fixed 7)
✓ Clean after 2 iterations. Fixes batched.

Stuck on real bug:

> /qa
Iteration 3: 2 failures remaining (non-auto-fixable)
✗ STUCK — see report. Recommendation: /investigate

See also

• /qa-only — read-only variant for ship-gate verification
• /investigate — when QA finds a real product-logic bug
• /review — runs after QA clean for diff-scoped review
• /ship — reads QA status as a pre-flight gate