Agent SkillsAuthoring

Validate a skill

Catch the frontmatter and naming mistakes that make ~1 in 5 public skills fail to load.

Roughly one in five public skills fails basic structural validation — which is why they silently never load. skill-author ships a dependency-free checker so you never ship one of them.

Run the checker

Point validate.py at any skill folder:

python3 validate.py path/to/skill

It's stdlib-only — no packages to install, no network calls — so it audits clean and runs anywhere Python does.

What it checks

  • Frontmatter — required fields present, valid YAML, no missing name/description.
  • Naming — lowercase-hyphenated, no spaces, no reserved words.
  • Body size — flags instruction bodies past the recommended token budget so the skill stays loadable and fast.
  • Structure — the folder layout the spec expects.

Why stdlib-only matters

A validator that pulls third-party packages is itself an audit liability. Ours is a single Python file with no dependencies and no network access — you can read the whole thing before you run it.

Use it in CI

Run it on every skill in a repo as a pre-publish gate — the same discipline behind our "100% spec-valid" claim. This is exactly the gate the maintained line runs, plus security scanning and evals.