Test Skill

Run tests and report results with actionable failure analysis.

Protocol

1. Read the test command

Read claude/build.md for the test command and any special flags.

2. Run tests

# Use the command from claude/build.md
# Example: npm test 2>&1 | tail -50

For targeted tests (faster feedback):

# Example: npm test -- --testPathPattern=<filename> 2>&1 | tail -30

3. Analyze results

• All pass: report summary and exit
• Failures: analyze each failure

4. For each failing test

A. Read the exact assertion failure:

Expected: <value>
Received: <value>

B. Find the source of the failure:

• What code is the test exercising?
• What did the recent change touch?
• Is this a pre-existing failure or newly introduced?

C. Fix or flag:

• If newly introduced by current change: fix it
• If pre-existing: document it, do NOT fix unrelated failures
• If test itself is wrong: flag for user review (never silently delete a failing test)

5. Re-run after fixes

Keep running until all tests pass (or all failures are documented as pre-existing).

6. Report

Tests: PASS — 142 passed, 0 failed, 0 skipped

or

Tests: FAIL — 140 passed, 2 failed, 0 skipped
Newly introduced failures:
  - test/auth.test.ts:45 — [failure reason + fix applied]
Pre-existing failures (not caused by current change):
  - test/legacy.test.ts:12 — [noted, not fixed]

Rules

• NEVER delete a failing test to make the suite pass
• NEVER mock the thing being tested
• NEVER mark a task complete with failing tests
• Pre-existing failures must be distinguished from introduced failures