Which of your skills is stealing the request?

Your skill passed eval. Does it fire in your setup, next to everything else you installed?

Official skill evals run each skill alone. In real use it sits next to dozens of other skills and plugins, and one of them quietly takes the request. In our own test, 21/21 prompts fired in an isolated eval; the same prompts missed 5 times in the real harness.

skillcheck runs your prompts through your real claude and codex CLIs, with all your skills loaded, and tells you which skill fired, which one stole the request, and whether your fix helped.

Why skills don't trigger, and how to fix it · See an example report — two skills went from 0/4 and 1/4 to 4/4 after rewriting their descriptions.

Try it (one file, Python 3 stdlib — source on GitHub)

curl -O https://raw.githubusercontent.com/Noem836/skillcheck/main/skillcheck.py
python3 skillcheck.py scan                      # static checks, no model calls
python3 skillcheck.py gen --skills my-skill     # draft should-fire / near-miss prompts
python3 skillcheck.py run --agent claude --tag before
python3 skillcheck.py report --tag before       # fire rate, false fires, who took it
python3 skillcheck.py upload --tag before       # optional share link

What it checks

• Discovery per agent: Claude Code (user, project, plugins, skillOverrides) and Codex skill folders
• Frontmatter, name/folder mismatch, nested SKILL.md, slash-only wording, disable-model-invocation
• Real trigger runs, repeated per prompt, so flaky skills show up as rates
• Competition: which other skill grabbed the prompt
• Before/after compare after you edit a description

Privacy

Runs use your own CLI login on your machine; nothing is sent to us. Runs are locked down: Claude gets only the Skill tool, Codex runs without your MCP servers, web search or plugins. upload is opt-in and sends only skill names and hit/miss rows, never prompt texts.