yao-meta-skill: a governed lifecycle system for modeling, compiling, and evaluating reusable AI agent skills

yao-meta-skill: a governed lifecycle system for modeling, compiling, and evaluating reusable AI agent skills

What it solves

yao-meta-skill is a framework for turning repeated workflows, prompts, and notes into reusable, governed AI agent skills. It moves beyond simple prompt engineering by treating skill creation as a software engineering process, providing tools for modeling, compiling, evaluating, and managing the full lifecycle of an AI skill.

How it works

The project implements a "Skill OS" (version 2.0) that follows a structured pipeline:

  1. Intent Modeling: Clarifies the job, outputs, and constraints of a skill through a dialogue process before generating files.
  2. Skill IR (Intermediate Representation): Creates a platform-neutral semantic contract for the skill, separating the intent from the platform-specific implementation.
  3. Compilation: Uses target compilers and adapters to generate the skill for various platforms, including OpenAI, Claude, and VS Code.
  4. Evaluation & Review: Employs an "Eval Lab" to test triggers and output quality, producing evidence that is reviewed in a "Review Studio" HTML gate page.
  5. Release & Governance: Verifies packages, simulates installations, and uses a "claim guard" to prevent premature public claims until evidence is verified.
  6. SkillOps Loop: Tracks adoption drift and telemetry to inform the next iteration of the skill.

Who it’s for

  • AI Agent Developers: Those who need to create portable, team-ready skills that work across different LLM providers.
  • Team Leads: Users who need governance, release gates, and evidence-based quality assurance for AI assets.
  • Power Users: Individuals turning personal productivity workflows into structured, installable skill packages.

Highlights

  • Platform Neutrality: Uses a Skill IR to compile skills for multiple targets (OpenAI, Claude, etc.) from a single source.
  • Evidence-Based Governance: Replaces blind trust with an evidence ledger and review studio for release gates.
  • Comprehensive Eval Lab: Includes assertion grading, blind A/B review packs, and runtime permission probes.
  • SkillOps: Integrated telemetry and adoption drift reports to guide iterative improvement based on real usage signals.

Sources