library:review — Completion Validation

Verify that claimed work is actually done. Evidence before assertions.

When to Use

• Before closing a PM task
• Before claiming a feature is complete
• When reviewing another session's work
• Before creating a PR or merging

Process

Step 1: Identify Claim

What is being claimed complete? Get specific:

• Task ID (if PM tracked)
• Feature name
• What "done" means for this item

Step 2: Gather Criteria

From the spec and vault, identify what "complete" requires:

• Code exists?
• Tests exist and pass?
• Endpoint/UI responds correctly?
• Edge cases handled?
• No regressions?

Step 3: Run Verification

For each criterion, run the actual check:

Criterion: "Auth middleware validates JWT"
Command: npm test -- --grep "auth" 
Result: [actual test output]

Do NOT rely on:

• Code review alone (code existing ≠ code working)
• Previous session's claims
• Memory entries without fresh verification

Step 4: Verdict

VERIFIED — All criteria pass. Evidence attached. PARTIAL — Some criteria pass, others fail. List what's missing. FAILED — Critical criteria fail. List failures with evidence.

Step 5: Update Records

• Call library:pm:update with verdict and evidence
• If VERIFIED: transition task to done
• If PARTIAL/FAILED: add comment with findings, keep task open
• Update vault wiki article if it had a [VERIFY] tag for this item

Iron Rule

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

If you haven't run the verification command in THIS session, you cannot claim it passes.

Token Budget

Weight: Light-Medium Estimated context cost: ~1200 tokens Subagent delegation: No

MCP Tools Used

• library:pm:update — update task with verdict
• library:vault:parse — check for related [VERIFY] tags
• library:config:get — project settings