joeseesun/qiaomu-meta-skill

What it solves

It transforms vague ideas, prompts, SOPs, or chat logs into professional, reusable "Skills" for AI agents. It solves the common problems of poor trigger accuracy (too broad or too narrow), lack of research into existing solutions, and the risk of insecure or broken releases when publishing skills to the public.

How it works

The project acts as a "meta-skill" that follows a rigorous development lifecycle:

1. Intent & Research: It identifies the goal and searches catalogs like skills.sh and SkillsMP to find existing similar skills, analyzing them to avoid duplication and improve quality.
2. Synthesis & Design: It creates a structured skill package including a SKILL.md for routing, a README.md for users, and supporting references and scripts.
3. Evaluation: It tests trigger boundaries to ensure the skill activates when it should and remains silent when it shouldn't.
4. Governed Release: It performs secret scanning for API keys, manages GitHub PRs and Releases, and verifies installation via npx in an isolated environment.

Who it’s for

AI agent developers and power users who want to turn their personal workflows into standardized, shareable, and high-quality software units for other agents to use.

Highlights

• Prior-Art Research: Searches multiple skill repositories to synthesize the best approach rather than just generating a prompt.
• Trigger Evaluation: Includes a dedicated system to test and refine how an agent triggers the skill.
• Security Gates: Built-in scanning for leaked secrets and API keys before publishing.
• Full Lifecycle Automation: Handles everything from initial research to GitHub Release and npx installation verification.
• Structured Output: Produces a complete package with manifests, interface definitions, and evaluation cases.