메뉴
Forces meticulous self-review of work just completed. Itemizes every change, creates verification todos, then hunts for gaps using structured uncertainty analysis. Use after finishing implementation work, before claiming done.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Guide CLAUDE.md updates based on doc-guard suggestions. This skill should be used when updating documentation, fixing docs, addressing doc-guard suggestions, or updating CLAUDE.md. Activates when user requests "update documentation", "fix docs", "doc-guard suggestions", or "update CLAUDE.md".
Monitor blocked workers and help respond to them. Loop checks ft inbox, shows blocked workers, lets you respond naturally.
Adds tasks to FormalTask with mandatory discovery and validated criteria. Use when "add task", "create task", or "/task-add". For batch creation, use /epic-decompose instead.
Groups uncommitted changes into logical commits, walks through each grouping for approval, then optionally syncs with remote. Resolves merge conflicts interactively with context from recent remote and local history. Use when "commit my changes", "group commits", "smart commit", "commit and sync", or "organize my changes". Never resolves conflicts without user confirmation.
Unified critique for plans, specs, OR task artifacts. Auto-detects target. Spawns auditors. Commits tasks to DB when READY. Use "critique", "review", "is this ready".
Systematic debugging with graded exploration and agent escalation. MUST BE USED when encountering errors, test failures, or unexpected behavior.
| name | verifying-completion |
| description | Forces meticulous self-review of work just completed. Itemizes every change, creates verification todos, then hunts for gaps using structured uncertainty analysis. Use after finishing implementation work, before claiming done. |
| required_todos | ["itemize-everything","verification-todos","gap-hunt","verdict"] |
BLOCKING GATE: You just completed implementation work in this session.
Review EVERYTHING you did. Not summaries. Actual items:
| Category | What to List |
|---|---|
| Files created | Every new file, with purpose |
| Files modified | Every edit, what changed and why |
| Code added | Functions, classes, methods - each one |
| Code removed | What was deleted and why |
| Config changes | Settings, env vars, dependencies |
| Tests | Test files, test cases |
| Commands run | Build, test, lint results |
Be paranoid. If you touched it, list it. If you thought about touching it but didn't, note that too.
EXIT CRITERIA: Written inventory of every discrete piece of work.
BLOCKING GATE: Phase 1 inventory complete.
Convert each item to a verification todo. Use TodoWrite:
TodoWrite(todos=[
{"content": "Verify: [specific item] is complete", "status": "pending", "activeForm": "Verifying [item]"},
# One todo per discrete piece of work
])
Verification means:
Work through each todo. Mark complete only with evidence.
EXIT CRITERIA: All verification todos complete OR blocked items identified.
BLOCKING GATE: Phase 2 verification complete.
Now challenge yourself using structured uncertainty:
UNKNOWNS: (list before analyzing)
ALTERNATIVES: (ways this could have been done differently)
GAPS FOUND: (be honest) For each unknown, check:
EXIT CRITERIA: All unknowns addressed. Gaps either fixed or explicitly flagged.
State clearly:
COMPLETE: All items verified, no gaps found.
OR
INCOMPLETE: [List specific items that need attention]
No hedging. No "should be fine." Evidence or gap.
- Every item gets a todo. No batching "miscellaneous changes." - "Looks complete" is not verification. Run it, test it, prove it. - Finding gaps is success, not failure. Missing them is failure. - If you can't verify something, say so. Don't assume. - User asked for X - did you deliver X? Not your interpretation of X.