knowyourailive
All tools
v1.0MIT

agentscan

A linter for the config files that give your coding agent its power.

A single .mcp.json can hand a model shell access. A CLAUDE.md can tell it to run whatever a scraped web page says. These files pass code review because they don't look like code.

  • Nine rule families: permission bypass, shell-command MCP servers, over-broad filesystem roots, auto-run hooks, hardcoded keys, instructions that disable confirmation, unpinned MCP packages, wildcard tool grants.
  • Covers Claude Code, Claude Desktop, Cursor, Cline/Roo, Windsurf and Copilot instruction files.
  • No API keys, no network calls, no dependencies. Under a second on a normal repo, with a non-zero exit code for CI.

Know the limit

Static analysis of files on disk. It does not observe what the agent does at runtime.

GitHub ↗
$ pipx install git+https://github.com/ppradyoth/agentscan
$ agentscan .

 CRITICAL AS001  permissions.defaultMode set to bypassPermissions
          .claude/settings.json:3
 CRITICAL AS001  MCP server 'shell' executes an arbitrary shell command
          .mcp.json:4
 CRITICAL AS004  Hardcoded secret (Anthropic API key)
          .claude/settings.json:14
 HIGH     AS002  MCP filesystem server rooted at a broad path

Summary: CRITICAL:6  HIGH:2  MEDIUM:8

Used in the guide