yaojingang/yao-meta-skill

YAO = Yielding AI Outcomes. A rigorous engineering, evaluation, governance, and portability system for reusable agent skills.

What it solves

yao-meta-skill is a system for turning messy, repeated operational workflows—such as prompt sets, transcripts, and runbooks—into governed, reusable AI agent skills. It moves beyond simple prompt engineering by treating AI skills as software assets that require a formal lifecycle: creation, compilation, evaluation, and release governance.

How it works

The project implements a "Skill OS" architecture that processes raw inputs through a structured pipeline:

  1. Intent Modeling: Clarifies the job, outputs, exclusions, and standards before any files are generated.
  2. Skill IR (Intermediate Representation): Creates a platform-neutral semantic contract for the skill, separating the intent from the specific AI platform.
  3. Compilation: Uses target compilers and adapters to generate the skill for multiple platforms, including OpenAI, Claude, and VS Code workflows.
  4. Evaluation: The "Output Eval Lab" runs trigger checks, output assertions, and benchmark reproducibility tests to create evidence of the skill's quality.
  5. Governance: A "Review Studio" acts as a gate, combining evidence, trust signals, and runtime permission probes to determine if a skill is ready for release.
  6. SkillOps: A feedback loop that uses metadata-only telemetry to track adoption drift and suggest adaptations for the next iteration.

Who it’s for

  • AI Agent Developers: Those who need to build portable, high-quality skills that work across different LLM providers.
  • Team Leads: Users who want to upgrade personal prompts into governed team assets with explicit boundaries and review gates.
  • Enterprise Operators: Those requiring strict evidence-based release governance and runtime permission auditing for AI tools.

Highlights

  • Platform Neutrality: Uses a Skill IR to compile one skill for multiple AI targets.
  • Evidence-Based Release: Replaces "vibes-based" testing with a formal evidence ledger and blind A/B review packs.
  • Comprehensive Toolchain: Includes a Review Studio for gatekeeping, an Eval Lab for assertions, and a SkillOps loop for maintenance.
  • Linguistic Support: Provides bilingual interpretation reports and localized documentation.