菜单
Unified code review router. Detects patterns, proposes reviewers, spawns in parallel. Use "review", "review my changes", "review PR
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
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 | review |
| description | Unified code review router. Detects patterns, proposes reviewers, spawns in parallel. Use "review", "review my changes", "review PR |
| argument-hint | <target> [--quick|--thorough|--security] |
| inherit | ["review"] |
| tools | ["auggie","warpgrep"] |
| spawns_subagents | true |
| required_todos | ["detect-target","analyze-patterns","propose-reviewers","spawn-reviewers","synthesize"] |
quick: Auto-detect target. Skip to Phase 3 with code-quality-auditor only.
full:
| Input | Action |
|---|---|
123 or PR #123 | gh pr diff 123 |
file.py | git diff HEAD -- file.py |
| (none) | git diff HEAD |
Extract: files changed, file types, line counts.
quick: Skip pattern analysis. Use code-quality-auditor only.
full: Scan diff for signals → map to reviewers:
| Signal | Reviewer |
|---|---|
sqlite3, cursor.execute | sqlite-reviewer |
subprocess, Popen, os.system | subprocess-reviewer |
Path(, os.path, open( | path-security-reviewer |
os.environ, getenv, API keys | configuration-auditor |
try:, except, raise | error-handling-reviewer |
BaseModel, Field(, validator | schema-reviewer |
hooks/, PreToolUse | hook-reviewer |
test_, pytest, assert | test-quality-auditor |
Enum, status transitions | state-machine-reviewer |
tmux, TUI widgets | tui-reviewer |
requests, httpx, API calls | api-client-reviewer |
| Large loops, recursion | performance-auditor |
input(, args., CLI args | input-validation-reviewer |
Always: code-quality-auditor
Build: detected (pre-checked) + available (unchecked).
quick: Skip AskUserQuestion. Use detected reviewers automatically.
full:
AskUserQuestion(questions=[{
"question": "Which reviewers?",
"header": "Review",
"options": [
{"label": "code-quality (Recommended)", "description": "Quality, antirez style"},
{"label": "sqlite-reviewer", "description": "Detected: SQL in diff"},
# ... detected first, then available
],
"multiSelect": True
}])
Profiles: --quick (quality only), --thorough (all detected + test), --security (security + path + input + config)
quick: Review diff yourself applying code-quality checks. Report findings inline.
full: ALL in SINGLE message (parallel):
Task(subagent_type="code-quality-auditor", model="sonnet", run_in_background=True,
prompt=f"Review diff:\n{diff}")
Task(subagent_type="sqlite-reviewer", model="sonnet", run_in_background=True,
prompt=f"Review diff:\n{diff}")
Wait for all. Collect outputs.
quick: Present findings inline with verdict. Skip formal synthesis.
full:
REVIEW: {target} | {N} findings | {reviewers}
─────────────────────────────────────────────
P0 - Blocking
file.py:42 [sqlite] SQL injection risk
P1 - Should Fix
file.py:15 [quality] Function >50 lines
─────────────────────────────────────────────
Verdict: PASS | FIX_REQUIRED | BLOCKED
Verdict: P0>0 → BLOCKED, P1>2 → FIX_REQUIRED, else PASS
- Always include code-quality-auditor - Detected = pre-checked, available = unchecked - Spawn ALL in single message - P0 blocks, no exceptions