Skill checker: see what an agent skill can run
How it works
- Your browser reads the repository straight from GitHub: the file list from the GitHub API and each file from raw.githubusercontent.com. Nothing goes through our server.
- Every folder with a SKILL.md file under the link is a skill. We check up to 50 per link.
- For each skill we read the SKILL.md frontmatter (name, description, allowed-tools, licence) and scan its script files as text for network calls. Nothing is run.
- If a skill is already listed on AI-Powered, you get a link to its page.
Browse checked skills in the agent skills directory, or see the most popular skills.
Skill checker FAQ
What does the skill checker look at?
The files in each skill folder on GitHub: which script languages it includes (Python, shell, JavaScript and others), the tools its SKILL.md asks for in allowed-tools, network calls found in its scripts (requests, fetch, curl and similar), binary files, the licence, and the number and size of its files.
Is this a security audit?
No. It is a static check of the text in the files, the same one we run for the skills listed on AI-Powered. It can miss things (for example code that builds a URL at run time or downloads more code), and a skill with no scripts can still tell your agent to run commands. Read the files before you install a skill.
Does it run the skill or send the files to AI-Powered?
No. Your browser reads the files from GitHub and checks them locally. Nothing is executed, and the files are never sent to our server. We only look up whether a skill folder is already listed on AI-Powered.
Can I check a private repository?
No. The checker reads public repositories only, without signing in to GitHub.
Why did GitHub stop the check?
GitHub lets each visitor make 60 API requests an hour without signing in. Each check uses two or more. When the limit is used up, the checker shows when it resets.