yao-meta-skill/SKILL.md

name, description, metadata

name	description	metadata
yao-meta-skill	Create/improve/evaluate agent skills from workflows, prompts, SOPs, scripts. Use for migration/release/package, routing, evals/tests, install checks, 优化已有 skill, 补 trigger 评测. Exclude no-skill summary/translation/docs.	author Yao Team

Yao Meta Skill

Router Rules

• Route by frontmatter description.
• Keep SKILL.md lean; put guidance in references/, logic in scripts/, and evidence in reports/.
• Use the lightest reliable process.

Modes

• Scaffold: exploratory/personal. Production: team reuse. Library: shared infra. Governed: high-trust/release-critical.
• Rules: Method, Operating Modes, Resource Boundaries.

Compact Workflow

1. One-off/no reusable process: Do not create a skill; near-neighbor; require repeated use + reusable output contract.
2. Capture job, output, exclusions, constraints, standards, lightest fit.
3. Scan up to 3-5 references: external, user, local; skip tiny edits; surface only uncertainty/conflict.
4. Write description early; route/boundary edits need trigger_eval.py; releases need risk-matched gates before folders.
5. Add output-risk, artifact-design, prompt-quality, system-model, next directions only when earned.

Playbooks: Method, Intent, Skill IR, Output Eval, Review Studio.

Skill OS 2.0 Gates

For production/library/governed/team releases, run Skill IR, compiler, trigger/output eval, Skill Atlas, conformance, trust, registry/package/install, upgrade, drift, waiver, Review Studio.

Governed Package Boundary

For file-backed/release-critical/governed packages, name input_files as file-backed fixture; include owner, review cadence, input_files, output contract, rollback boundary; require trust report and reports/output_quality_scorecard.md; mark unavailable telemetry/approvals/metrics/benchmarks as missing evidence; do not fabricate evidence.

Preserve labels literally when they apply: file-backed fixture, input_files, output contract, rollback boundary, trust report, reports/output_quality_scorecard.md, missing evidence.

First-Turn Style

• Start from the user's work/outcome before structure.
• Ask only 2-3 key questions unless enough detail exists.
• In Chinese, sound soft and companion-like; use Intent Dialogue.

Output Contract

Create/refactor/package: produce SKILL.md, aligned agents/interface.yaml, justified assets, boundary/exclusion/gate summary. Audit/evaluate-only: findings + proposed fixes; edit files only if asked. No-skill: no files.

Reference Map

Primary: Method, Artifact Design, Systems, Governance, SkillOps.