Paste a skill. See what it does.
The quick scan runs pattern checks in your browser. Nothing is uploaded. For the full scan, run the CLI locally.
Paste the SKILL.md or skill file content
Runs in your browser. Nothing is uploaded.
The quick scan runs pattern checks. It never calls a skill malicious.
The full scan adds taint analysis and attack paths: $ pipx install agentscan-cli
What the quick scan checks
Shell
Interpreter invocations, eval/exec, os.system, subprocess.
Network
URLs, credential-in-URL, fetch-and-execute pipes.
Secrets
AWS, GitHub, Slack, Stripe, OpenAI, JWT, private keys.
Supply chain
Unpinned installs, git clone, curl|bash.
Obfuscation
Base64 decode-to-execute chains, hex escapes.
License
Declared vs missing. 93.6% of public skills declare none.
The full scan is local, free, and open source.
It scans directories, MCP servers, and agent configs. Same input gives the same report. It never runs a skill.