Preflight for agent skills.
Paste a SKILL.md or upload a skill folder to see what might get flagged before you publish or install it.
Findings
LiveCheckSkill rules with references to published scanner checks.
Paste or upload a skill to see findings here.
Use CheckSkill from your terminal
Send your Markdown as the request body to get JSON findings.
curl --fail-with-body -sS \
-H 'Content-Type: text/markdown' \
--data-binary @SKILL.md \
https://checkskill.ai/api/scan
On macOS, create an unlisted link and open the skill in your browser. Use xdg-open on Linux.
link=$(curl -fsS \
-H 'Content-Type: text/markdown' \
--data-binary @SKILL.md \
https://checkskill.ai/api/open) && open "$link"
Use POST /api/share to get the same link in JSON. All three routes also accept application/json with a skill string. For a folder scan, send JSON with a files array of {path, content} objects. Submit up to 64 text files with paths relative to the folder and 64 KiB of text total. Share links have a separate 24,000-character limit for the compressed source in the URL.
Why CheckSkill exists
A skill can tell an agent what to run, which content to trust, and what to report to you. CheckSkill highlights instructions worth checking before you publish or install a skill. It’s free and works without an account.
Skills.sh publishes audits from Gen Agent Trust Hub, Socket, and Snyk. Those hosted classifiers run on their providers’ infrastructure. CheckSkill runs its own rules on the exact text you provide, including unpublished revisions, and shows the passages behind each finding. The references point to related checks from other scanners. A clear scan only means these rules found no matches. Skills.sh audits
Snyk also offers a CLI that reads local files and sends them to its hosted analysis service. Snyk scanning documentation