Summary

Luban addresses the challenge of transforming individual, often unshareable "agent skills" into robust, public assets. It targets developers creating agent skills, providing a systematic workshop to polish them for understanding, installation, distribution, verification, and continuous evolution, thereby enhancing their utility and adoption.

How It Works

Luban employs a five-action methodology: "Material Inspection" (验料) to challenge the skill's premise, "Competitor Analysis" (访行) to understand its ecosystem, "Measurement" (过尺) via structure, empirical tests, and live data, "Slow Carving" (慢刨) to freeze baselines and gate changes by verification, and "Re-smelting" (回炉) to establish post-release observation lists. This approach ensures skills are rigorously validated and designed for long-term maintainability, moving beyond superficial improvements.

Quick Start & Requirements

Installation is via npx skills add LearnPrompt/luban-skill -g Claude Code or through the plugin marketplace (/plugin marketplace add LearnPrompt/luban-skill, then /plugin install luban). To use, instruct the agent: "Let Luban review my skill: [Your Skill directory / GitHub repo link / SKILL.md content]". The tool requires an agent environment supporting Claude Code plugins. Official documentation is within the SKILL.md file, and a detailed case study is available in skills/luban/examples/ai-news-radar-case.md.

Highlighted Details

• Systematic Skill Polishing: Utilizes a unique five-action process (验料, 访行, 过尺, 慢刨, 回炉) for comprehensive skill refinement.
• Evidence-Based Validation: All claims and improvements are supported by verifiable data, competitor analysis with URLs, and dedicated verification scripts.
• Real-World Impact: Successfully upgraded ai-news-radar (v0.6 to v0.7.0) with merged PRs, uncovering silent failures and improving key metrics like data accuracy and rendering performance.
• Iterative Development: The "Slow Carving" principle ensures changes are only integrated after passing rigorous verification gates, promoting stable evolution.

Maintenance & Community

Developed by "LearnPrompt," the project draws inspiration from other open-source efforts like khazix-skills and darwin-skill. Community engagement is facilitated through the WeChat Official Account "卡尔的AI沃茨," X handle @aiwarts, and the website learnprompt.pro. Several related projects from the "LearnPrompt" collective are also listed.

Licensing & Compatibility

The project is released under the MIT License, permitting free use, modification, and distribution. No specific restrictions for commercial use or integration with closed-source projects are noted, aligning with the permissive nature of the MIT license.

Limitations & Caveats

Luban enforces strict safety boundaries, requiring explicit user authorization for critical actions like merging to default branches or deploying changes. It avoids writing sensitive credentials or private paths into public artifacts and ensures all modifications are auditable. The tool's initial step is to challenge the skill's premise and perform analysis before making any edits, preventing immediate, unverified modifications.