golden-qa-regressionlisted
Install: claude install-skill Shadysmetools/claude-skills
# Golden-Reference QA Regression
You execute a prebuilt QA kit against a running app. **The kit is the oracle**: never invent
expected behavior, read it.
## The kit (configure paths once, e.g. in the repo's CLAUDE.md)
| Item | Default | Contents |
|---|---|---|
| `MATRIX` | `qa/TEST-MATRIX.xlsx` (or `.csv`) | sheet `Test Cases`; columns `ID, Feature Area, Entry Route, Precondition, Test Data, Steps, Expected Result, Persona, Priority, Status, Notes / Evidence, Linked Bug` |
| `GOLDEN` | `qa/happy-paths/<area>.md` | what "correct" looks like per area |
| `PLAYBOOK` | `qa/PLAYBOOK.md` | click-by-click action scripts per case (optional) |
| `RUNBOOK` | `qa/QA-RUNBOOK.md` | safety rules for this app |
| `FIXTURES` | `qa/REAL-DATA-SCENARIOS.md` | named test data |
| `EVIDENCE` | `qa/_evidence/` | screenshots per case |
| `APP_URL` | `https://app.example.com` | the app under test |
No matrix yet? Offer to draft one from the golden reference first, and stop until the user approves it.
## Step 0: Authenticate
Prefer importing the user's already-logged-in browser session (cookie import) into the QA browser,
or let the user log in once and take over. Verify: load an authenticated page and confirm it renders
(not the login page). If it bounces to login, STOP and ask.
## Step 0.5: Production safety rails (non-negotiable)
- Reads anywhere. **Writes and destructive actions only inside a sandbox** (e.g. a project named
`__QA_SANDBOX__` created by the QA account).
- **Never complete