# QA Safe Healing

> Diagnose Playwright failures using repository evidence and apply only policy-permitted, guarded repairs.

- Skill: `mallikarjun-roddannavar/qa-safe-healing` (Agent Skill)
- Install (CLI): `npx skillmds@latest add mallikarjun-roddannavar/qa-safe-healing`
- Raw SKILL.md: https://api.skillmd.com/api/skills/mallikarjun-roddannavar/qa-safe-healing/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Mallikarjun-Roddannavar (https://skillmd.com/u/mallikarjun-roddannavar)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/mallikarjun-roddannavar/qa-safe-healing

---


# QA Safe Healing

Use this skill for a failed Playwright test, a request to heal a test, or a diagnosis-only QA report. It complements `full-stack-incident-analyst`; use that skill when source-to-runtime tracing needs a full cross-layer RCA.

1. Read `qa/failure-taxonomy.json` and `qa/evidence-schema.json`.
2. Preserve test output, trace/screenshot when available, API/network evidence, source, and relevant product knowledge.
3. Explain test intent and supported behavior separately; emit classification, confidence, evidence, root-cause hypothesis, and `testModificationAllowed`.
4. In diagnosis-only mode, change nothing. Missing evidence means `UNKNOWN`, not a guess.
5. Modify only if policy permits it. Never weaken assertions, skip/fixme/delete tests, swallow errors, or force actions.
6. Make the smallest repair, run `npm run qa:guardrails`, rerun the affected test, and report the exact change and evidence location.

`APPLICATION_DEFECT`, `API_CONTRACT`, `ENVIRONMENT`, and `UNKNOWN` preserve the failure. `LOCATOR_DRIFT` permits only a high-confidence locator-only repair. `TIMING`, `TEST_DATA`, and `ASSERTION_ERROR` require review.