qa

QA Skill — Full Quality Assurance Pass

Run a complete QA pass on all current changes.

Step 1: Scope the Changes

git diff --name-only
git diff --cached --name-only

Categorize every changed file:

• backend → Python tests
• frontend → JS/TS tests + type checking
• ML → ML invariant tests + auditor check
• config → validate no secrets, correct formats
• tests → run them to make sure they pass

Step 2: Detect Test Framework

Before running anything, detect the project's test setup:

• Python: Check pyproject.toml, setup.cfg, pytest.ini, Makefile for pytest/unittest/tox config
• Node/JS/TS: Check package.json scripts (test, typecheck), look for jest/vitest/mocha config
• Type checking: Check for tsconfig.json (tsc), mypy.ini/pyproject.toml (mypy)
• Linting: Check for eslint, ruff, flake8 configs
• E2E: Check for playwright, cypress configs
• Monorepo: Check for workspace config (pnpm-workspace.yaml, nx.json, turbo.json) — run tests per package

Use the detected commands for all subsequent steps.

Step 3: Run Tests (ALL applicable)

Run the test commands discovered in Step 2. Examples:

• pytest -v --tb=short (or whatever the project uses)
• npm test / pnpm test / yarn test
• npx tsc --noEmit / mypy .

If new code has no tests → WRITE TESTS FIRST.

Step 4: Write Missing Tests

For every new function, endpoint, or component without test coverage:

1. Create the test file in the appropriate tests/ directory
2. Write tests covering: happy path, edge cases, error cases
3. Run them and confirm they pass

Step 5: Regression Check

Run the FULL suite — not just tests for changed files. Use the same commands from Step 2, but ensure full scope (no path filters).

Step 6: Report

QA PASS REPORT
==============
Scope:       [backend/frontend/ML/config]
Tests:       X passed, Y failed, Z new tests written
TypeScript:  clean/X errors
E2E:         ran/skipped (reason)
Regressions: none found / [list]

VERDICT: PASS / FAIL

Step 7: Fix Issues

If FAIL: fix every issue found, then re-run from Step 3. Loop until PASS.

Rules

• This skill produces EVIDENCE, not opinions
• "It should work" is NOT acceptable — run it
• If there's no test infrastructure: build it
• Every QA run must end with a clear PASS or FAIL verdict