Skills

Install

$ npx ai-agents-skills add --skill verification-protocol
Behavioral v1.0

Verification Protocol

Iron law: “Evidence before claims.” Gate function ensures verification happens before marking tasks complete.

When to Use

  • Verifying task completion
  • Validating implementation against requirements
  • Confirming tests pass before claiming success
  • Ensuring evidence exists before assertions

Don’t use for:

  • Code review (use code-review skill)
  • Debugging (use systematic-debugging skill)

Critical Patterns

✅ REQUIRED: Iron Law (Evidence Before Claims)

NEVER claim completion without verification evidence.

# ❌ WRONG: Claim without evidence
✅ Task 1: User registration implemented

# ✅ CORRECT: Evidence before claim
## Task 1: User registration

**Verification**:
- Ran: `npm test -- auth.test.ts`
- Result: 5/5 tests passed
- Output: "Tests: 5 passed, 5 total"
- Tested scenarios:
  - Valid email/password → 201 + user object
  - Duplicate email → 409 error
  - Invalid email → 400 error
  - Weak password → 400 error
  - Missing fields → 400 error

**Evidence**: All tests green ✅

**Claim**: ✅ User registration implemented and verified

Why evidence first?

  • Prevents false positives (claiming done when it’s not)
  • Creates audit trail
  • Builds confidence in completion
  • Enables independent verification

✅ REQUIRED: Gate Function (5 Steps)

Every verification follows this sequence. NEVER skip steps.

1. IDENTIFY what to verify 2. RUN the verification (test, build, manual check) 3. READ the output 4. VERIFY result matches expectation 5. CLAIM completion (only if step 4 passed)

## Verification: Build succeeds

1. **IDENTIFY**: TypeScript compilation must succeed with 0 errors
2. **RUN**: `npm run build`
3. **READ**:

tsc —project tsconfig.json

Successfully compiled TypeScript files Time: 2.3s

4. **VERIFY**: ✅ Build succeeded
- 0 errors
- 0 warnings
- dist/ folder created with .js files
5. **CLAIM**: ✅ Build verification passed

---

## Verification: Tests pass

1. **IDENTIFY**: All unit tests for UserService must pass
2. **RUN**: `npm test -- UserService.test.ts`
3. **READ**:

PASS tests/UserService.test.ts UserService ✓ registerUser creates user (45ms) ✓ registerUser throws on duplicate email (23ms) ✓ validateEmail accepts valid emails (5ms) ✓ validateEmail rejects invalid emails (4ms) ✓ hashPassword returns different hash each time (67ms)

Tests: 5 passed, 5 total Time: 1.2s

4. **VERIFY**: ✅ All tests green
- 5/5 passed
- No warnings or errors
- All assertions satisfied
5. **CLAIM**: ✅ UserService tests passing

✅ REQUIRED: Verification Checklist

Each task has explicit verification steps.

## Task: Implement login endpoint

**Implementation**: [code here]

**Verification Checklist**:

- [ ] 1. IDENTIFY: POST /auth/login must accept email/password and return JWT
- [ ] 2. RUN: `curl -X POST http://localhost:3000/auth/login -d '{"email":"test@example.com","password":"pass123"}'`
- [ ] 3. READ: Response body and status code
  ```json
  {
    "token": "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9...",
    "user": { "id": "123", "email": "test@example.com" }
  }
  • 4. VERIFY:
    • ✅ Status 200
    • ✅ Token is valid JWT
    • ✅ Token payload contains user ID
    • ✅ Password not in response
  • 5. CLAIM: ✅ Login endpoint verified

### ❌ NEVER: Skip Verification Steps

Skipping steps leads to undetected failures.

```markdown
# ❌ WRONG: Skipping VERIFY step
1. IDENTIFY: Tests must pass
2. RUN: npm test
3. READ: [didn't actually read output]
5. CLAIM: Tests passing ✅ ← DANGEROUS (step 4 skipped!)

# ✅ CORRECT: All 5 steps executed
1. IDENTIFY: Tests must pass
2. RUN: npm test
3. READ: Output shows "12 passed, 2 failed"
   - Failed: UserService.login() throws on invalid password
   - Failed: UserService.login() returns null on non-existent email
4. VERIFY: ❌ 2 tests failing
5. CLAIM: ❌ CANNOT claim completion - fix failing tests first

**Action**: Investigate failures, fix issues, re-run verification

❌ NEVER: Assume Without Evidence

# ❌ WRONG: Assuming success
✅ Tests passing (assumed - didn't run)
✅ Build succeeds (assumed - didn't verify)
✅ Lint clean (assumed - didn't check)

# ✅ CORRECT: Evidence for each claim
- Tests: ran npm test → 25/25 passed ✅
- Build: ran npm run build → dist/ created ✅
- Lint: ran npm run lint → 0 errors ✅

Decision Tree

Task completed?
  → Run gate function (IDENTIFY → RUN → READ → VERIFY → CLAIM)
  → Verification passed? → Mark complete
  → Verification failed? → DO NOT mark complete, fix issues

Making a claim?
  → Evidence exists? → Provide evidence + claim
  → No evidence? → Run verification first

Multiple verification points?
  → Create checklist with all 5 steps per item
  → Verify each item independently

Verification blocked (environment issue)?
  → Document blocker
  → Alternative verification method?
  → Wait for blocker resolution

Edge Cases

Flaky tests: If fails intermittently, run 3 times. If passes 3/3, consider stable. If fails 1+/3, investigate root cause.

Slow verification (>5 min): Note duration. Example: “Ran E2E suite (8 min) → 12/12 passed ✅”.

Manual verification: When automation unavailable, document manual steps explicitly.

1. IDENTIFY: Registration form accepts valid input
2. RUN: Manually open http://localhost:3000/register
3. READ: Form fields visible (email, password, confirm password)
4. VERIFY (manual):
   - Entered: test@example.com, password123
   - Clicked "Register"
   - Result: Redirected to /dashboard with success message
5. CLAIM: ✅ Registration form verified manually

Partial verification: When full verification blocked, verify what’s possible.

## Task: Deploy to staging

1. IDENTIFY: App must run on staging server
2. RUN: `ssh staging "pm2 start app"`
3. READ: "Process started"
4. VERIFY: ⚠️ PARTIAL
   - ✅ App started successfully
   - ❌ BLOCKED: Cannot verify public URL (DNS propagation pending)
5. CLAIM: ⚠️ PARTIAL - App deployed, public verification pending DNS

Checklist

  • Every claim has evidence
  • Gate function 5 steps followed (IDENTIFY → RUN → READ → VERIFY → CLAIM)
  • Verification output captured (not assumed)
  • Failed verifications prevent completion claims
  • Checklist provided for multi-step verification
  • Blockers documented if verification cannot complete
  • Evidence is reproducible (commands + outputs provided)

Example

# Feature Verification: User Authentication

## Task 1: Registration endpoint

### Verification

1. **IDENTIFY**: POST /auth/register must create user and return 201
2. **RUN**: `npm test -- auth-register.test.ts`
3. **READ**:

PASS tests/auth-register.test.ts POST /auth/register ✓ creates user with valid data (45ms) ✓ returns 201 status (12ms) ✓ returns user without password (8ms) ✓ rejects duplicate email with 409 (23ms) ✓ validates email format (7ms) ✓ enforces min password length (9ms)

Tests: 6 passed, 6 total

4. **VERIFY**: ✅ All tests passed
5. **CLAIM**: ✅ Registration endpoint verified

---

## Integration Verification

1. **IDENTIFY**: Full auth flow must work (register → login)
2. **RUN**: `npm test -- auth-integration.test.ts`
3. **READ**:

PASS tests/auth-integration.test.ts Authentication flow ✓ can register and immediately login (156ms) ✓ login fails before registration (45ms)

Tests: 2 passed, 2 total

4. **VERIFY**: ✅ Integration tests passed
5. **CLAIM**: ✅ Full auth flow verified

Resources