Statically scan a Claude Code skill, plugin, or MCP server (a local path or a git/GitHub URL) for malware patterns BEFORE installing it: secret/env exfiltration to suspicious hosts, install-time (pre/postinstall) hooks, obfuscated remote execution (eval(atob), base64|sh), prompt injection in tool descriptions/SKILL.md, committed binaries, and auto-approve/skip-permission configs. Static-only — it clones and reads files, never executes the target code, so scanning something malicious is safe. Returns a verdict (clean/suspicious/dangerous) with the flagged files.