| name | test-fix |
| description | Run unit tests and fix failures. Use when the user asks to run tests, fix failing tests, make tests pass, tests are broken, fix test errors, or get tests green. |
| compatibility | Requires pnpm (or npm/yarn); project must define a test script. |
Test Fix
Trigger scenarios
Activate this skill when the user says or implies:
- Run tests, fix failing tests, make tests pass
- Tests failed, tests are broken, fix test errors, get tests green
Purpose
Run the test suite; if it fails, read the output, apply minimal fixes, and re-run until tests pass or an iteration limit is reached.
Commands for this repo
Run from the repository root.
- Test:
pnpm test (runs pnpm --recursive test; the package uses vitest run).
- Test with coverage:
pnpm test:coverage (runs Vitest with coverage across workspace).
- Coverage report:
pnpm coverage:report — runs tests with coverage, writes coverage-report.json, exits 1 if below thresholds. Use when verifying test coverage before completion.
Fixer loop
- Run: Execute
pnpm test from the repository root.
- Identify: Read the test output for failing file, test name, and assertion or error message.
- Fix: Apply the minimum necessary change (assertion, fixture, code under test, or mock) to resolve the failure. Prefer one logical fix per iteration.
- Verify: Re-run
pnpm test.
- Repeat until all tests pass or up to 5 iterations to avoid unbounded loops.
Common error types
- Assertion mismatch: Adjust the expectation or fix the implementation to match the intended behavior.
- Type / snapshot / import errors: Fix types, snapshot, or imports in the reported file.
- Missing or stale fixtures: Update test data or paths.
Example
First run (tests fail):
pnpm test
After editing files to fix reported failures:
pnpm test
Verifier integration
Verifier step 3 is the coverage report gate—use the Coverage report command in Commands for this repo above (pnpm coverage:report).
When used by the verifier agent, the verifier also runs pnpm coverage:report after tests pass. If coverage:report fails (belowThreshold or test failures), add or improve tests per this skill's fixer loop and re-run pnpm coverage:report until it exits 0.
Other projects
If the project uses a different package manager or test command, run the equivalent from the repo root (e.g. npm test, yarn test, pnpm test, cargo test). Use the same fixer loop: run tests → read failures → fix → re-run.
