Skill Author Toolkit

Three skills for people who build OpenClaw skills: audit what you've already published, lint a new SKILL.md before you ship it, and run the publish flow without copy-pasting flags from memory.

What's inside

Skill Doctor. Audit, debug, and clean up installed skills. Detects duplicate or conflicting triggers, inline security red flags, and out-of-date versions. Predicts which skill will fire for any prompt.

SKILL.md Linter. Runs nine concrete checks against a SKILL.md (or a folder of them): frontmatter completeness, kebab-case naming, semver version, license declared, description length and trigger-phrase count, declared permissions block, "when NOT to use" clause, vague-trigger detection. Outputs a report card with before/after fix snippets.

Publish Helper. Bumps version, appends a dated changelog entry, validates the frontmatter, and prints the right clawhub publish or clawhub package publish command with all flags filled in. Runs the command on your confirmation.

What you need

Local-only. No connectors, no API keys. The Publish Helper shells out to the clawhub CLI for the actual publishing step; install it via npm i -g clawhub and run clawhub login once.

Quick start

1. Install the plugin.
2. Try one of:
   • "audit my skills" (Skill Doctor)
   • "lint my SKILL.md at [path]" (Linter)
   • "publish [skill] as a patch release" (Publish Helper)

Roadmap

Planned for v0.2:

Eval runner. Run a skill against a test-prompt suite and grade the outputs.

Skill scaffolder. Generate a starter SKILL.md (plus references) from a one-paragraph description.

License

MIT. Built by Chris (chris-openclaw on ClawHub).