verification

# Verification Skill

> **Core Principle**: Evidence before claims, always.

## The 5-Step Mandate

Before ANY completion claim, execute these steps in order:

### 1. IDENTIFY
What command, check, or test proves the claim?
- Be specific: exact command, exact file, exact assertion
- "It should work" is NOT identification

### 2. RUN
Execute the verification fresh.
- Do NOT rely on cached results
- Do NOT rely on memory of previous runs
- Do NOT skip because "I just ran it"

### 3. READ
Examine the complete output.
- Check exit code (0 = success)
- Count results (how many passed/failed?)
- Read error messages fully

### 4. VERIFY
Binary decision: Does the output confirm the claim?
- YES → Proceed to step 5
- NO → Stop. Do not claim completion.

### 5. CLAIM
Only now, make the claim WITH evidence attached.
```
Claim: [what you're claiming]
Evidence: [command run] → [output received]
```

---

## Disqualifying Language

These phrases BLOCK completion claims. If you catch yourself using them, STOP and run verification:

| Phrase | Why It's Blocked |
|--------|------------------|
| "should work" | Uncertainty = unverified |
| "probably" | Uncertainty = unverified |
| "seems to" | Uncertainty = unverified |
| "I believe" | Belief ≠ evidence |
| "I think" | Thought ≠ evidence |
| "Great!" | Celebration before verification |
| "Done!" | Claim before evidence |
| "Perfect!" | Superlative before verification |

---

## What Verification Looks Like

### Good Verification
```
Claim: Tests pass

Evidence:
$ npm test
✓ 47 tests passed
Exit code: 0
```

### Bad "Verification"
```
Tests should pass now since I fixed the bug.
```

---

## Integration with Deviation Log

Unverified claims are logged as deviations:

```yaml
- id: [auto]
  timestamp: [now]
  expected: "Verification before completion claim"
  actual: "Claim made without evidence"
  root_cause: premature_completion
  fix: |
    Enforce verification skill invocation.
    Add reminder to session-start hook.
```

---

## Whe