mirror of
https://github.com/magnus919/agent-skills.git
synced 2026-09-11 19:47:12 +03:00
Magnus Hedemark
8e5bc023a0
Adds references/vetting-third-party-skills.md with a dependency-style vetting checklist (provenance, SKILL.md body, scripts, references), safe first-run practice, and reporting guidance, citing the Snyk ToxicSkills audit as the primary source for ecosystem risk statistics. SKILL.md gains the match-prescriptiveness-to-fragility decision rule, the run-vs-reference intent rule for bundled scripts, and an Adopting Third-Party Skills section. README and eval manifest updated. Closes #311
2.5 KiB
2.5 KiB
Agent Skills — The Standard Itself
This skill is the meta-skill: it documents the Agent Skills open format itself. Every other skill in this repository follows the conventions defined here.
Why Install This Skill
When your agent loads this skill, it gains the ability to create, review, and edit valid Agent Skills-format skills. That means:
- Your agent can build new skills on demand. Describe a workflow you want encoded, and your agent can scaffold a valid SKILL.md with correct frontmatter, progressive disclosure structure, and agent-friendly descriptions.
- Existing skills stay valid. When your agent edits a skill, it checks against the format standard — no broken frontmatter, no missing required fields.
- Consistency across your skill library. Every skill in your repository follows the same conventions, making them loadable by any Agent Skills-compatible harness.
What You Get
| Directory | Purpose |
|---|---|
SKILL.md |
A create-and-review workflow, the core format rules, and a completion checklist |
references/specification.md |
A bundled snapshot of the official format specification |
references/best-practices.md |
Guidance for useful, well-scoped instructions |
references/optimizing-descriptions.md |
How to design and test reliable automatic triggers |
references/using-scripts.md |
How to bundle safe, agent-friendly executable helpers |
references/vetting-third-party-skills.md |
How to vet a skill from a registry, colleague, or LLM before running it |
references/evaluating-skills.md |
A practical evaluation loop for testing skill quality |
references/client-implementation.md |
Guidance for products that discover and load skills |
Quick Start
Ask your agent to create or review a skill, then load this skill first. For a structural check, run:
skills-ref validate path/to/skill
If the reference validator is not installed, your agent should report that clearly and complete the equivalent manual checks.
Triggers
- You're creating a new skill and need to know the correct format
- You're reviewing an existing skill for format compliance
- You want your agent to generate a valid SKILL.md scaffold
- You're setting up conventions for a shared skill repository
Requirements
The documentation itself has no runtime dependencies. Optional validation uses the skills-ref reference library; install it in an isolated environment when you want to run its CLI.