How to check the quality of Claude Code and Codex skills with NVIDIA SkillEvaluator
NVIDIA's SkillEvaluator is a CLI that checks an agent skill folder (SKILL.md and its supporting files) and reports a score and what to fix. Run it before you hand a skill to your team or publish it. This covers 0.9.0, released November 13; it needs Python 3.12 to 3.14.
- Install the CLI.
pip install skillevaluator
- Point it at your skill folder and run only the Tier 1 checks. No API key is needed.
skillevaluator validate ./my-skill --tiers 1
Tier 1 downloads pinned releases of the SkillSpector and Gitleaks security scanners when they are missing. If a scanner cannot run, the result is INCOMPLETE and the exit code is nonzero. To get just a score without scanners, use quality-check. The passing threshold defaults to 70.
skillevaluator quality-check ./my-skill -r cli -r json -o reports/first-quality
- To ask for it in plain language from Claude Code or Codex, add the SkillEvaluator skill.
npx skills add NVIDIA/SkillEvaluator
Without Node.js, clone the repository and copy the skill folder. The first copy is for Claude Code, the second for Codex.
git clone --depth 1 https://github.com/NVIDIA/SkillEvaluator.git
cp -r SkillEvaluator/skills/skillevaluator ~/.claude/skills/
cp -r SkillEvaluator/skills/skillevaluator ~/.codex/skills/
Then ask something like "Is my skill any good?" The agent picks the right tier, runs the CLI, reads the report and tells you what to fix.
- To check changed skills on every pull request in a GitHub repository, run the command below. It opens a setup pull request.
skillevaluator ci init
There are three tiers. Tier 1 checks that the skill is safe and well-formed. Tier 2 looks for guidance that repeats or overlaps and uses embeddings and a chat model. Tier 3 compares agent results with and without the skill, and needs a model provider, a supported agent and a sandbox (Docker by default). Tiers 2 and 3 can incur model charges. The license is Apache-2.0.
Sources
Get notified when a new tip lands
Tip alerts only go out in the app. Install it and turn notifications on.