review-work
Skill3 starsChanged 2 months agoMITView on GitHub
当一批工作完成提交前需要独立综合评审,或 omz-reviewer 单点评审不足以覆盖目标/QA/安全/上下文维度时激活。5-lane 并行评审;全 PASS 才 PASS。
8 other projects use a version of this.
Adopted by
Most recently taken up by
- benmix/skillstheir own version, first seen Oct 13, 2026
- wjgoarxiv/litopencodetheir own version, first seen Oct 13, 2026
- AgiFlow/ai-plugintheir own version, first seen Oct 10, 2026
- hassan-mohiddin/freeflowtheir own version, first seen Oct 4, 2026
- jakubsuplicki/codumenttheir own version, first seen Oct 4, 2026
- Ostico/Skillstheir own version, first seen Aug 12, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- jakubsuplicki/codument★ 48same name
- agoodway/.claude★ 11same name
- hassan-mohiddin/freeflow★ 5same name
- AgiFlow/ai-plugin★ 4same name
- apocohq/skills★ 4same name
- wjgoarxiv/litopencode★ 2same name
- benmix/skills★ 1same name
- Ostico/Skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- ulw-executeSkill
- ulw-planSkill
- ulw-researchSkill
The file
Files
review-work — 5-lane 并行评审协议
把"这批工作能不能算完成"拆成 5 条独立 lane 并行审。全 lane PASS 才 PASS;任一 FAIL 即 FAIL;INCONCLUSIVE 不算 PASS。
Phase 0:上下文收集(发起前必做)
主 agent 一次性收集,放进每个 lane 的 prompt(lane 是叶代理,看不到主会话):
- 目标原文 + 二进制成功标准(goal.json)
- 变更清单:
git diff --stat+ 完整 diff(或 worktree 路径) - DoneClaim / 测试输出 / 工件路径
- 评审范围声明:本次评审覆盖的文件/行为面
5-lane 配置(一回合并行 spawn 齐发,主 agent 派发)
| lane | 角色 | 职责 |
| 1 Goal Verifier | omz-oracle | 目标达成度:对照 goal.json 逐条 SC 裁决 |
| 2 QA Executor | omz-junior(或按 category 路由) | 实测:重跑测试/QA 步骤,输出自己的转录,不信转述 |
| 3 Code Reviewer | omz-reviewer | 代码质量:blocker/major/minor 穷举格式 |
| 4 Security | omz-oracle | 注入/凭证泄漏/权限扩大/危险操作面 |
| 5 Context Miner | omz-junior | git 考古:log/blame/--grep/reverted commits,找"这段历史上翻过车"的证据 |
lane 3/4 各自穷举输出"未发现 X 类问题";lane 1/2/5 输出 PASS | FAIL | INCONCLUSIVE + 证据。
裁决规则
- lane 是叶代理、一 verdict 即终审(lane 内部不 iteration)。
- lane 沉默/超时 → 重 spawn 一个范围更小的 reviewer 补位 → 仍失败 → 安全关闭并在报告点名该 lane 未完成。
- 复审 = 全新 spawn,scope 仅限 delta(新改动部分),不复审全盘。
worktree 纪律
评审尽量在独立 worktree:git worktree add → 评审期 git worktree lock --reason="<评审批次>" → 结束 git worktree unlock && git worktree remove。主 worktree 只读。
报告模板(汇总给主 agent)
# review-work 报告 <批次 id>
lane1 Goal: PASS|FAIL|INCONCLUSIVE — 一句话证据
lane2 QA: PASS|FAIL|INCONCLUSIVE — 一句话证据
lane3 Code: PASS(blocker=0 major=<n> minor=<n>)|FAIL — 摘要
lane4 Sec: PASS|FAIL|INCONCLUSIVE — 摘要
lane5 Context: PASS|FAIL|INCONCLUSIVE — 摘要
总裁决: PASS|FAIL
未决项: <INCONCLUSIVE lane 的待办>FAIL 时必须附可执行的修复派单建议(每条映射到 category 路由)。
references/
两份都是机器契约,不是背景阅读——不读就派/不读就汇总 = 契约必然漂移:
references/lane-prompts.md— 5 个 lane 的完整派发 prompt(含通用 MUST NOT DO 条款与全部占位符{{BATCH_ID}}{{GOAL}}{{DIFF}}{{DIFF_STAT}}{{FILE_CONTENTS}}{{DONECLAIM}}{{TEST_TRANSCRIPT}}{{SCOPE}}{{WORKTREE}})。派任何 lane 之前必须先读它,占位符在 Phase 0 逐个填实;lane 是叶代理,prompt 之外的上下文它一概看不到。references/verdict-schema.md— 裁决的字段级契约。汇总裁决前必须按它解析:单 lane 报告 JSON schema、exhaustive_check维度集合、汇总规则(任一 FAIL 即 FAIL / INCONCLUSIVE 不算 PASS)、AdversarialVerify JSON(verdict/evidence/repro/confidence,confirmed唯一通过)及其与 lane verdict 的映射、复审上限 2 次与 delta scope 规则。缺字段视为 lane 未完成。
两套枚举禁止混用字面量:lane verdict 是 PASS | FAIL | INCONCLUSIVE,AdversarialVerify 是 confirmed | false-positive | needs-fix | needs-human-review(后者与 agents/omz-reviewer.md、skills/ulw-execute/SKILL.md 三处逐字一致)。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

