菜单
Expensive multi-perspective self-review using parallel subagents. Use when asked for a deep self-review, when the reviewer explicitly escalates to the parallel path, or when you need broader code-quality coverage than the default reviewer-owned pass.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | self-groom |
| description | Expensive multi-perspective self-review using parallel subagents. Use when asked for a deep self-review, when the reviewer explicitly escalates to the parallel path, or when you need broader code-quality coverage than the default reviewer-owned pass. |
| argument-hint | [file or directory to focus on] |
Get the git diff of recent changes, classify the change, then launch parallel review subagents. Synthesize their feedback into a prioritized report.
This is the expensive path. Use /self-groom when the user explicitly wants a deeper self-review or when the reviewer asks for the parallel pass.
Launch all applicable reviewers simultaneously as subagents.
The subagent should read all CLAUDE.md files (user-level, project-level, repo root) and check every change against documented guidelines. Flag violations with specific file:line references. If you are Codex, then check for AGENTS.md compliance instead. (Often AGENTS.md is aliased to CLAUDE.md, but if both exist, check AGENTS.md for Codex and CLAUDE.md for Claude.)
Apply Software Design Principles to identify unnecessary complexity:
Also fix other basic code quality issues: naming, formatting, comments, dead code, duplication.
For each issue: state the problem, which principle it violates, and a concrete fix.
Check if new code paths are tested. Are tests testing behavior (not implementation)? Obvious edge cases missing? Brittle coupling to implementation details?
Performance (O(n²) where O(n) possible, unnecessary copies). Correctness (off-by-one, race conditions, unhandled edge cases). Security (injection, hardcoded secrets, unsafe deserialization).
Combine all findings into a deduplicated, prioritized report:
file: link format, not plain file:line text.[TopBar.tsx:162](file:web/src/components/TopBar.tsx:162).file: links only when a relative path would be ambiguous or unavailable.## Groom Report
### Critical (must fix)
- [TopBar.tsx:162](file:web/src/components/TopBar.tsx:162) Issue — Reviewer: X
### Recommended (should fix)
- [TopBar.tsx:162](file:web/src/components/TopBar.tsx:162) Issue — Reviewer: X
### Suggestions (nice to have)
- [TopBar.tsx:162](file:web/src/components/TopBar.tsx:162) Issue — Reviewer: X
### Looks Good
- Aspects that passed with no issues
Dispatch workflow for leader/orchestrator sessions. Use when dispatching a quest to a worker, choosing which worker to assign, spawning new worker sessions, or deciding whether to reuse vs spawn. Triggers: 'dispatch', 'send quest', 'assign worker', 'spawn worker', 'which worker', 'reuse or spawn'.
Cross-session orchestration for Takode. Use when you need to interact with other sessions: listing active sessions, peeking at session activity, reading messages, sending instructions to workers, or spawning new sessions. Triggers: 'check on workers', 'send to session', 'orchestrate', 'coordinate agents', 'list sessions', 'peek at session', 'what are my sessions doing', 'check session status'.
Confirm quest intent before creating a new quest or refining an idea into a worker-ready quest. Invoke when a user asks to create, file, refine, scope, or prepare a quest, or before materially rewriting quest title/description/tags as part of refinement. Do not use for routine quest feedback, claiming, completion, verification checks, or other bookkeeping.
Port changes from a git worktree to the main repository. This is the skill behind `/port-changes`; `worktree-rules` remains the underlying skill slug/directory. Use when asked to 'port changes', 'sync to main', 'push to main repo', '/port-changes', or when porting worktree commits.
Quality review for another agent's change. Invoke it with a concise description of the change to review, then inspect that worker's status and diff and later verify whether required findings were addressed.
Adversarial review of a worker's completed task. Independently evaluates whether the worker did thorough, honest work or took shortcuts. Returns ACCEPT or CHALLENGE with specific questions. Invoked by the reviewer session after being spawned by the leader.