Verification Skill Skill

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:

- 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.

When This Skill Applies

Before saying "done", "complete", "finished", "working"
Before committing code
Before creating PRs
Before marking tasks complete
Before reporting success to user
Before transitioning to next task

[!CRITICAL] IF YOU ARE ABOUT TO CLAIM SUCCESS, YOU MUST VERIFY FIRST.

Verification Assertions

Before any completion claim, these assertions must be TRUE:

assertions:
  - name: command_executed
    check: "verification command was run (not remembered)"
    fail_action: "Run the command now"

  - name: output_read
    check: "full output examined (not truncated)"
    fail_action: "Read complete output"

  - name: exit_code_checked
    check: "exit code is 0 (or expected non-zero)"
    fail_action: "Check exit code"

  - name: binary_verdict
    check: "verdict is YES or NO (not 'probably')"
    fail_action: "Make binary decision"

  - name: evidence_attached
    check: "claim includes [command] → [output]"
    fail_action: "Attach evidence to claim"

Assertion Checklist

Before claiming completion, verify each assertion:

| # | Assertion | Check | Pass? | |---|-----------|-------|-------| | 1 | command_executed | Did I run the command THIS session? | ☐ | | 2 | output_read | Did I read the FULL output? | ☐ | | 3 | exit_code_checked | Is exit code 0 (or expected)? | ☐ | | 4 | binary_verdict | Is my answer YES or NO (not "maybe")? | ☐ | | 5 | evidence_attached | Does my claim include [command] → [output]? | ☐ |

[!CRITICAL] IF ANY assertion fails → STOP. Do not claim completion.

Why Assertions > Prose Rules

| Prose Rule | Assertion | |------------|-----------| | "Make sure to run the command" | command_executed: TRUE | | "Read the output carefully" | output_read: TRUE | | "Check if it worked" | binary_verdict: YES or NO |

Assertions are:

Testable - Can be verified mechanically
Binary - TRUE or FALSE, no interpretation
Actionable - Each has a fail_action

Prose rules require interpretation. Assertions require compliance.

Agent Skills: Verification Skill

Install this agent skill to your local

Skill Files