
from yao-meta-skill297
Create, refactor, evaluate, and package reusable agent skills — a lightweight framework for turning processes into reliable skills.
Provides a structured playbook for turning repeated workflows, prompts, or transcripts into reusable SKILL.md packages. The meta skill defines router rules, operating modes (Scaffold, Production, Library), and a compact workflow to guide skill creation, evaluation, and packaging.
Use when you need to convert a manual process into a reusable agent skill, set up governance/QA ladders for skills, or scaffold a production-ready skill directory. Useful for teams standardizing skill design and delivery.
references/.Meta and authoring workflows are agent-agnostic but most useful to developer-facing agents and tools (Copilot, Codex, Claude Code) that produce or edit repository content.
Yao Meta Skill is a comprehensive framework for creating, evaluating, and packaging agent skills. It features 39 scripts covering eval suites, governance checks, cross-platform packaging, description optimization, and CI testing. The SKILL.md is lean with good progressive disclosure into references/. Architecture follows the AgentSkills spec well. However, most scripts fail in isolation because they depend on internal modules (trigger_eval), data files, and a Makefile — they're designed for the full repo, not standalone use. Only 4/39 scripts ran successfully in isolation.
trigger_eval (internal module, not packaged standalone)Well-architected meta-skill with excellent separation of concerns. The main weakness is that scripts are deeply coupled to the repo structure and can't run standalone. Security is solid — URL allowlisting, env var filtering, path traversal protection. The skill would benefit from bundling trigger_eval as a script or making it installable.