yaojingang/yao-meta-skill — reverse-engineered prompt

Build me a Python based Skill OS that helps turn repeated AI workflows into reusable agent skills people can actually install, review, and improve over time.

I want to be able to start from messy inputs like notes, prompts, transcripts, or a runbook, and have the tool generate a clean skill package with a simple skill doc, clear triggers, inputs and outputs, references, optional scripts, and evaluation files. It should support compiling that same skill into formats that work across common agent platforms like OpenAI, Claude, and more generic skill targets, instead of locking it to one tool.

Please also include a strong review and governance layer. I want generated HTML pages that show an overview of the skill, a review gate with warnings and fixes, evidence for what has been tested, release readiness checks, install simulation, permission and trust checks, failure tracking, and a ledger for what is proven versus still pending. It should feel beta ready, practical, and honest about gaps. If needed, look up current docs online and wire things together in a sensible way.