Luban skill
A workshop skill: the agent vets a finished skill, finds peers, scores it and edits step by step, testing every change.
- 965 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent browses the web for peers and, at your choice, edits files and commits in the skill repository.
- Merging into the default branch, tagging a release and deploying happen only with your explicit permission.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/lubanDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent browses the web for peers and, at your choice, edits files and commits in the skill repository.
- Merging into the default branch, tagging a release and deploying happen only with your explicit permission.
This is third-party code. Review the repository files before installing.
What it does
Luban takes a working skill and makes it publishable. The agent goes through five actions: it checks whether the skill should exist at all, searches the web for peers and compares against them, scores the skill on three rulers (structure, a test run, a live check), makes edits one at a time and, after release, leaves a watch list for the next round. The output is a 13-section report: a peer table with links, scores and three improvement directions with ready replacement snippets. A shareable certificate comes with it, showing scores before and after, the new positioning in one line and the next step. Checks stay in the repository: a one-off comparison becomes a script and acceptance criteria become project rules.
Who it is for. Skill authors whose working skill has no installs and no clear description.
Good fit when
- The skill works well for you, but nobody installs it from GitHub.
- Before publishing you want a peer comparison and a score backed by evidence.
- You want every change to triggers and README accepted only after a check on real data.
- After release you need a watch list to start the next round of improvements.
- You doubt that the skill should exist in its current form.
Not a fit when
- The skill has to be written from scratch: Luban refines existing work.
- You need a quick cosmetic text fix without comparison or checks.
- The agent must stay offline: it looks for peers on the web.
- The release must ship automatically: Luban leaves the merge, tag and deploy to you.
Example request
`让鲁班看看这个skill：<folder or GitHub link>`. Pick one of the three proposed directions while the agent has changed nothing.Limitations
Trigger phrases and most of the README are in Chinese, for example `让鲁班看看这个skill`; the English version is in `README.en.md`. The peer-search step needs web access for the agent. The repository shows one worked example: improving the ai-news-radar project from version 0.6 to 0.7.0. Luban merges into the default branch, tags a release and deploys publicly only after your direct permission, and a question like "is everything done?" does not count as permission. The last commit was made in July 2026, the license is MIT.
How to disable. Remove the `luban` plugin through `/plugin` or delete the skill folder from the agent's skills directory.
FAQ
How do I install Luban in Claude Code?
Add the marketplace and install the plugin: `/plugin marketplace add LearnPrompt/luban-skill`, then `/plugin install luban`. With a skills manager use `npx skills add LearnPrompt/luban-skill -g`; Hermes Agent has `hermes skills install LearnPrompt/luban-skill`.
How do I start a skill review?
Tell the agent the phrase from the README and give the skill folder, a repository link or the SKILL.md text. Luban vets the material, finds peers, scores the skill and proposes three directions with one recommendation.
Does Luban change files without asking?
No. Until you choose a direction it edits nothing. It also stops before changing the positioning, merging into the default branch, tagging a release and deploying anything users can see.
What does the report contain?
Thirteen sections: the vetting verdict, a peer table with links, a niche assessment, a score table, three improvement directions and ready replacement snippets.
What did the example in the repository show?
On the ai-news-radar project the live check found that data had not updated for 8 days while CI checks stayed green. Replaying the scoring on 83,725 old records removed 327 false items.
Discussion
No questions or reviews yet.
Related
SkillsA doctor skill for agent skills that picks an architecture, audits against 21 red lines and returns a report with line references
A skill that designs, tests, evaluates and packages other agent skills, starting from architecture
A meta-skill that turns a workflow into an agent skill: prior-art search, trigger evaluation and publication to GitHub
A skill for creating skills that proves them by behavior: before-and-after runs, regression evals and ecosystem audit
A skill that helps an agent author other skills to the agentskills.io spec: structure, metadata, progressive disclosure and validation
A skill that turns a goal into a ready prompt: the agent works in builder and critic loops until its output wins a blind comparison with a reference
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Luban (this card) | no | 965 | |
| Darwin Skill | no | 6,818 | scores SKILL.md on nine criteria and rolls back failed edits; Luban adds a necessity check, peer search with links and a report with three directions |
| skill-creator | no | 2,291 | helps write a new skill to the specification; Luban works on a finished skill and takes it to publication |
README in short
Besides installation, the README compares Luban with an ordinary request to improve a skill and lists the mandatory stop points. The method builds on horizontal-and-vertical analysis, the ratchet from darwin-skill and validation-gated acceptance from Microsoft SkillOpt.
README badge
Original SKILL.md
--- name: luban description: | 鲁班（Luban）——Skill打磨工坊。把一个"能用的Skill"打磨成"能被理解、能被安装、能被传播、能被验证、能持续进化"的公共Skill资产。 方法论是工匠式的五个动作：验料（先挑战这个Skill的前提是否成立，不值得雕的料直说）、访行（联网寻找同类Skill，看清自己在生态里站什么位置）、过尺（结构、实测、活体三把尺一起量——活体指拉真实运行产物对账，绿色的CI会撒谎）、慢刨（冻结原版做基线，改动必须通过验证门才保留，否则回刀；验证手段尽量沉淀为仓库里的工具和规矩）、回炉（发布不是终点，留对标观察清单，下一轮从真实反馈进）。 当用户想要升级、优化、打磨、产品化、发布自己开发的Skill时使用。最终产出一份结构化的《Skill打磨报告》、可直接替换的改写片段，以及一张可截图传播的"出师证书"结果卡。 触发词包括但不限于：让鲁班看看这个skill、班门打磨、打磨我的skill、升级我的skill、优化这个skill、skill体检、skill审计、产品化我的skill、这个skill怎么发布、对标一下同类skill、为什么我的skill没人装、帮我把skill发到GitHub/ClawHub、改进SKILL.md。 即使用户只是丢来一个Skill目录、GitHub仓库链接或一段SKILL.md说"帮我看看怎么改"，只要上下文是想让Skill变得更好用、更可传播，都应该触发。 不要用于从零创建一个新Skill（用skill-creator）、不要用于普通的代码review（用code-review）、不要用于改写一段和Skill资产无关的普通提示语。 --- # 鲁班 | Skill打磨工坊 > **工坊规矩** > 鲁班打磨一件工具，靠五个动作。**验料**：先判断这块料值不值得雕——朽木不可雕也，不值得就直说，给出换料的方向。**访行**：把市面上同类的活儿都看一遍，知道自己这件在行里站什么位置，闭门造车出不了好工具。**过尺**：结构、实测、活体三把尺一起量，每个分数都要有证据，不凭手感——活体那把尺量的是真实运行产物，静默失败比文档烂致命。**慢刨**：原件先封存做基线，刨完拿尺子再量——量得过就留，量不过就回刀，绝不为了显得干了活而多刨。**回炉**：交活不是终点，同行还在动，用户还会回来，下一轮从真实反馈进。 你是鲁班，工匠祖师爷。用户把他的Skill拿到班门前，你的任务不是夸它或者随手抛光，而是把它当成一件准备摆进GitHub/ClawHub/skills.sh/Tessl生态的作品来打磨：让第一次见到它的人一眼能看懂、一分钟能装上、三分钟能跑出看得见的结果。最终产出一份**《Skill打磨报告》**、通过验证门的**可直接替换的改写片段**，以及一张**"出师证书"结果卡**。 打磨过程中你同时是五个工种： 1. **掌柜**（产品经理）：判断这件工具到底解决谁的什么问题，为什么值得安装。 2. **行脚**（生态研究员）：在GitHub、ClawHub、skills.sh、Tessl等生态中寻找同类Skill，分析它们凭什么被理解、收藏、安装、传播。 3. **量尺师傅**（审计员）：用结构评分 + 实测表现双轨评估，找出最该优先打磨的面。 4. **刨工**（优化器）：做有边界的候选编辑，只接受能通过验证门的改动。 5. **摆活儿的**（README与Showcase导演）：把Skill包装成别人愿意停下来看、看完想装的公共资产。 ## 前置准备 ### 接活：明确打磨对象 用户可能给你以下任意一种输入。如果已经足够明确，不需要追问，直接开始： 1. **目标Skill**：本地Skill目录路径 / GitHub仓库链接 / ClawHub页面 / 一段SKILL.md正文 / 一个还没成型的Skill想法 2. **目标发布平台**（可选）：GitHub / ClawHub / skills.sh / Tessl / 私用 3. **用户优先级**（可选）：传播力 / 实测效果 / 安装率 / 跨runtime兼容 / README表达 / showcase强度 如果输入不完整，先用现有材料做**最小可行审查**，不要卡住，但必须明确标注缺失项。 完整的实战案例（真实仓库、真实数字、全程可查证）见 `examples/ai-news-radar-case.md`——拿不准某一步该做到什么深度时，对照它。 ### 看料：读取材料清单 尽量读取/检查以下材料，读不到的标注"缺失"： - `SKILL.md`、`README.md`