Verify Completion

Verify Completion Skill

You systematically verify work is actually complete before claiming done.

Philosophy

• Evidence over feeling - "I think it's done" -> "I verified it's done"
• Original requirements - Check against what was asked, not what was built
• Edge cases - The obvious path works, but what about edges?
• Clean state - No debug code, no TODOs, no temporary hacks

Process

1. Recall Original Requirements

First, explicitly state what was requested:

• What did the user originally ask for?
• Were there any clarifications or changes?
• What's the acceptance criteria?

2. Run Verification Checklist

Go through systematically:

## Completion Verification

### Requirements Check
- [ ] Original requirement met: [state it]
- [ ] All clarifications addressed: [list any]
- [ ] Scope creep avoided: [nothing extra added]

### Code Quality (if applicable)
- [ ] No debug code (console.log, print, debugger)
- [ ] No commented-out code
- [ ] No hardcoded values that should be config
- [ ] No TODO/FIXME/HACK comments left
- [ ] Error handling complete

### Testing
- [ ] Happy path works
- [ ] Edge cases handled: [list them]
- [ ] Error cases handled: [list them]
- [ ] Tests pass (if test suite exists)

### Integration
- [ ] Works with existing code
- [ ] No breaking changes (or documented)
- [ ] Dependencies updated if needed

### Cleanup
- [ ] Temporary files removed
- [ ] Git status clean (only intended changes)
- [ ] No unrelated changes mixed in

3. Verify Evidence

For each check, verify with evidence:

# No debug statements
grep -r "console.log\|print(\|debugger" src/

# Tests pass
npm test  # or equivalent

# Git status clean
git status
git diff --stat

4. Check Edge Cases

For the specific feature, identify and test edges:

| Feature Type | Common Edge Cases | |--------------|-------------------| | Input handling | Empty, null, very long, special chars | | Lists | Empty list, single item, many items | | Numbers | Zero, negative, very large, decimals | | Dates | Boundaries, timezones, leap years | | Auth | No token, expired, invalid, wrong permissions | | Files | Missing, empty, too large, wrong format |

5. Final Walkthrough

Mentally or actually walk through:

1. User does the thing
2. System responds correctly
3. Edge cases handled gracefully
4. Errors give helpful messages

6. Report Verification Status

## Verification Complete

**Original requirement:** Add rate limiting to API

### Verified [OK]
- [x] Rate limiting applied to all `/api/*` routes
- [x] Returns 429 with Retry-After header
- [x] Configurable via environment variables
- [x] Tests pass (12 new tests added)
- [x] No debug code
- [x] Git diff shows only rate-limit related changes

### Edge Cases Tested
- [x] First request (no prior count)
- [x] Exactly at limit (100th request)
- [x] Over limit (101st request)
- [x] After window reset
- [x] Redis connection failure (graceful degradation)

**Ready to commit.**

Or if issues found:

## Verification Found Issues

**Original requirement:** Add rate limiting to API

### Issues Found
1. **Missing:** Graceful degradation when Redis is down
2. **Debug code:** console.log on line 47 of rateLimiter.js
3. **Edge case:** Empty API key not handled

### Not Ready - Fix These First

Quick Verification (for small changes)

For trivial changes, abbreviated check:

Quick verify:
- [x] Does what was asked
- [x] Tests pass
- [x] No debug code
- [x] Clean diff

Ready.

Using Tasks for Thorough Verification

For large changes, verify different aspects in parallel:

Task 1: Verify all requirements met (check against original request)
Task 2: Run test suite and check coverage
Task 3: Check for code quality issues (debug code, TODOs)
Task 4: Test edge cases manually

Synthesize: Ready or issues found.

Anti-Patterns

DON'T:

• Say "done" based on feeling
• Skip edge case testing
• Leave "I'll fix it later" items
• Mix verification with more changes

DO:

• Verify against original requirements
• Test edge cases explicitly
• Clean up before claiming done
• Document what was verified

When to Use

• Before committing
• Before creating PR
• Before saying "done" to user
• Before moving to next task
• After any "I think it's ready"

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hknc
• Source: hknc/claude-evolve
• License: MIT

Install and usage instructions live in the source repository linked above.