task-evaluator
Use this skill to review a task result after implementation and before write-back.
Inputs
Provide:
implementation_brief- changed files or diff
- checks run and results
- optional touched docs or restart artifacts
Workflow
More from tosukui/myskills
swarm-meeting
Run structured multi-role design reviews and architecture debates for technical decisions. Use when Codex needs to compare options, pressure-test tradeoffs, recommend an MVP path, or simulate a meeting with distinct evaluation roles such as moderator, skeptic, pragmatist, minimalist, maximalist, retrieval architect, Granary workflow lead, semantic purist, lightweight contrarian, context economist, or workflow conservative.
13swarm-research-pack
Build compact task-scoped evidence packs from local code, docs, notes, and diffs before implementation or design review. Use when Codex needs to triage many local artifacts, prepare briefing material for swarm-meeting, separate known facts from assumptions, or narrow what should be loaded into live context.
12task-writeback
Update restart notes, task records, and plan deltas after a task without bloating AGENTS.md. Use when Codex should write back the minimum durable context needed for the next fresh session.
10task-implementer
Implement a bounded change from a task-scoped brief without reloading the whole project history. Use when Codex already has an implementation brief or research pack and should keep coding context narrow.
10task-supervisor-loop
Coordinate a task through task-scoped research, implementation, evaluation, and write-back phases so live context stays small. Use when Codex should avoid one long session and instead pass compact artifacts between phases or subagents.
10bootstrap-project-state
Fill or refresh AGENTS.md, docs/project/current_state.md, and docs/project/critical_path.md from repository evidence without inventing facts. Use when a repo has thin state templates or stale restart docs and Codex should reconstruct the durable project summary from local files.
8