AgentSkillsCN

write-failing-test

为测试计划中所有预定的验收标准编写失败测试。在“/plan-tests”之后、实施之前使用此功能。可通过“/write-failing-test path/to/spec.md”或“编写失败测试”“红色阶段”“开始TDD”等指令触发。

SKILL.md
--- frontmatter
name: write-failing-test
description: "Write failing tests for all planned acceptance criteria from the test plan. Use after /plan-tests, before implementation. Invoke with '/write-failing-test path/to/spec.md' or 'write failing test', 'red phase', 'start TDD'."
contract:
  tags: [tdd, testing, red-phase]
  state_source: spec
  inputs:
    params:
      - name: spec_path
        required: true
    gates:
      - field: "Test Plan.status"
        value: "Planned"
  outputs:
    mutates:
      - field: "Test Plan.status"
        sets_to: "Tests Written"
    side_effects: []
  next: [implement-to-pass]
  human_gate: false

Write Failing Test (Red Phase)

Writes failing tests according to the test plan in the spec. Works autonomously through each planned test until all have properly failing tests.

Why This Skill Exists

TDD requires seeing tests fail before implementing. This skill executes the test plan created by /plan-tests, writing all failing tests that define the feature scope.

When to Use

  • •After /plan-tests has added a test plan to the spec
  • •Before any implementation begins

Prerequisites

  1. •Spec document exists and is approved
  2. •Test Plan section exists in the spec (created by /plan-tests)
  3. •Path to spec provided as argument
  4. •Test infrastructure is running (see Step 0)

Instructions

Step 0: Pre-Flight Infrastructure Check

Before writing any tests, verify ALL infrastructure is ready.

See .claude/primitives/testing-conventions.md for project-specific infrastructure requirements.

Run these checks:

bash
# 1. Docker running
docker info

# 2. Supabase running
supabase status

# 3. Test command works
npm test  # Should run even if 0 tests

Required state:

  • •✅ Docker info returns successfully (not "Cannot connect to Docker daemon")
  • •✅ Supabase status shows all services running (DB URL: 127.0.0.1:54322)
  • •✅ npm test runs without errors (may show 0 tests, that's fine)

If ANY check fails:

  1. •Report the specific failure to user
  2. •Provide fix command (e.g., "Run supabase start")
  3. •STOP - do not proceed until user confirms fix

Only proceed when ALL checks pass.

CRITICAL: Skipped tests are NOT acceptable. If tests skip due to missing infrastructure, this is a BLOCKING failure. Never claim "Ready for implementation" with skipped tests.

Step 1: Read the Spec and Test Plan

Read the spec document. Locate the ## Test Plan section.

See shared/spec-io.md for Test Plan section structure and how to read the status field.

Gate check: Verify Test Plan **Status:** is Planned. If already Tests Written or Passing, tests have already been written — stop.

If no Test Plan section exists:

code
This spec doesn't have a test plan yet. Run `/plan-tests path/to/spec.md` first.

Extract from the Test Plan:

  • •Which test files to expand vs create
  • •Each planned test with its criteria and expected failure
  • •Manual checks (skip these - they're for QA)

Step 2: Loop Through All Test Files

For EACH test file in the plan, perform steps 3-5. Work autonomously — do not pause between files.

The manager will review all tests together when you report back.

Step 3: Write the Tests

Write tests according to the plan. The plan tells you:

  • •File to create or expand
  • •Test descriptions (it('should...'))
  • •Which criteria each test covers
  • •Test type (Unit/Integration/E2E)

When expanding existing tests:

  • •Look for existing describe blocks that match
  • •Add new it blocks within existing structure
  • •Share setup code (beforeAll, test fixtures)

Test Type-Specific Guidance:

For Unit/Integration tests (TypeScript/JavaScript):

See .claude/primitives/testing-conventions.md for project-specific patterns and examples.

For E2E tests (Playwright/Python):

See shared/e2e-patterns.md for reconnaissance-then-action pattern and server management.

Use skills/webapp-testing/scripts/with_server.py for server lifecycle management.

Meaningful Test Checklist (MUST pass all for EACH test):

  1. •

    Does the test CALL the code under test?

    • •❌ Bad: Test only sets up data and asserts setup worked
    • •✅ Good: Test calls the function/endpoint/UI being tested
  2. •

    Does the assertion verify the BEHAVIOR in the criterion?

    • •❌ Bad: expect(response.status).not.toBe(400) (proves nothing)
    • •❌ Bad: expect(data).toBeTruthy() (just checks something exists)
    • •✅ Good: expect(response.error).toMatch(/replyTo.*required/) (verifies specific behavior)
  3. •

    Will the test PASS when the feature is correctly implemented?

    • •If you can't answer "yes" clearly, the test doesn't test the criterion
  4. •

    Is this test deferring to a non-existent test?

    • •❌ Never write: "actual verification is in integration test"
    • •If you can't test it here, flag it and ask

See .claude/primitives/testing-conventions.md for good vs bad test examples specific to this project.

Step 4: Run the Test

Run the test using the project's test command (usually npm test).

Step 5: Verify Correct Failure

The test MUST fail for the RIGHT reason:

Good failures (feature doesn't exist):

  • •"Cannot find module '../components/MyComponent'"
  • •"Expected true but received undefined"
  • •"TypeError: myFunction is not a function"

Bad failures (test infrastructure problems):

  • •Syntax errors in the test file
  • •Missing test dependencies
  • •Configuration issues
  • •Typos in import paths
  • •SKIPPED TESTS (Supabase not running, Docker not running, etc.)

Skipped tests are NOT acceptable. A skipped test proves nothing. If tests skip:

  1. •Stop immediately
  2. •Fix the infrastructure (see Step 0)
  3. •Re-run tests
  4. •Only proceed when tests actually FAIL (not skip)

If "bad failure": fix and retry (up to 3 attempts per test).

Step 6: Update Test Plan Status

After all tests are written, update the spec's Test Plan section:

  • •From: **Status:** Planned
  • •To: **Status:** Tests Written

See shared/spec-io.md for how to update the Test Plan status field.

Step 7: Complete and Hand Off

After ALL planned tests have failing tests:

code
## Red Phase Complete

**Spec:** Documents/specs/42-feature-spec.md
**Tests written:** X tests across Y files

### Summary

**Expanded:** src/test/edge-functions/send-email.test.ts
- ✅ `it('accepts valid replyTo email')` - fails: "replyTo not implemented"

**Created:** src/test/edge-functions/send-email-validation.test.ts
- ✅ `it('rejects invalid replyTo format')` - fails: "validateEmail is not a function"

Ready for implementation. Run `/implement-to-pass` to continue.

THIS SKILL NOW ENDS. Do not proceed to implementation.

What NOT to Do

  • •❌ Write implementation code
  • •❌ Stop after one test (complete ALL tests in the plan)
  • •❌ Wait for user input between tests
  • •❌ Skip running tests
  • •❌ Accept tests that pass (means nothing to implement)
  • •❌ Accept tests that fail for wrong reasons
  • •❌ Write placeholder tests that don't call the code under test
  • •❌ Write tests that "defer to integration tests"
  • •❌ Use weak assertions like expect(x).toBeTruthy()
  • •❌ Claim "Ready for implementation" when tests are skipped
  • •❌ Deviate from the test plan (if plan seems wrong, ask first)

When to Pause

Only pause and ask user for help when:

  • •Stuck on same error 3+ times for one test
  • •Test plan seems wrong or incomplete
  • •Missing information needed to write test
  • •Blocking infrastructure issue (test framework broken)

Do NOT pause for:

  • •Normal test failures (that's expected!)
  • •Moving to next test
  • •Minor decisions within test implementation

Handoff

This skill hands off to:

  • •/implement-to-pass - Implements code to make all tests pass