Turning Messy Workflows into Governed AI Assets

What it does yao-meta-skill is a Python CLI framework that takes raw operational material—transcripts, prompts, runbooks, notes—and distills it into a structured, reusable skill package. Each package includes a lean SKILL.md, trigger definitions, evaluation cases, and export adapters for platforms like OpenAI and Claude. The default flow bakes in governance checks, resource boundary validation, and a feedback loop so the resulting asset is meant to be team-ready rather than a one-off prompt.

The interesting bit The project treats prompt authorship as systems engineering, not creative writing. It front-loads an “intent confidence gate” that keeps asking questions until the true job is unambiguous, silently scans public GitHub repos for pattern conflicts, and generates a small library of reports—artifact design profiles, system models, prompt quality profiles—that abstract complexity into reviewer-visible evidence. It also publishes a self-administered benchmark where it scores 91.5/100, handily outranking Anthropic and OpenAI skill creators on its own rubric.

Key highlights

• Claims perfect precision and recall (1.0/1.0) across 66 prompts in 21 test families, with zero false positives or negatives in its own eval suite.
• Self-graded portability score of 100/100 across OpenAI, Claude, and generic export targets, plus CJK-aware trigger matching.
• Context budgets are tracked explicitly (e.g., root skill sits at 994/1000 tokens), treating token limits as a first-class engineering constraint.
• Governance maturity is quantified: the root skill scores 90/100, and the system includes anti-pattern regression checks.
• README is refreshingly honest about excluding GitHub stars from its quality metric, preferring its own weighted engineering rubric.

Caveats

• The impressive benchmark scores are self-administered; the project grades itself against competitors using its own rubric and dataset.
• The documentation is dense with internal jargon—archetype-aware quickstarts, RTF mapping, intent dialogues—so expect a learning curve before the workflow clicks.
• At 552 stars, it is still a relatively niche tool, and the README admits this by deliberately ignoring popularity as a quality signal.

Verdict Worth a look if you are building a team library of reusable AI skills and need structure around boundaries, testing, and cross-platform export. Skip it if you just want a lightweight way to jot down a few prompts; the overhead here is closer to setting up a microservice than saving a text file.