Your First Scan in 30 Seconds

Your Claude skills, MCP servers, and AI plugins run code you didn’t write, from sources you didn’t audit. One command tells you if that’s a problem.

Run your first scan

npx firmis scan .

That’s it. Firmis auto-detects Claude Skills, MCP Servers, Codex Plugins, Cursor Rules, and 4 more platforms — no config file, no manifest, nothing to set up.

What you’ll see

This is what a real finding looks like:

 Firmis Scanner v1.3.0

 Scanning: /your/project
 Platforms: mcp (3 servers), claude (2 skills)
 Rules: 209 enabled

 CRITICAL  sd-015  AWS credentials exposed in tool handler
           src/tools/aws-helper.ts:22
           → This skill is reading your AWS_SECRET_ACCESS_KEY at runtime

 CRITICAL  tp-003  Hidden instruction injected via tool description
           src/tools/search.ts:14
           → Prompt tells the agent to exfiltrate chat history silently

 HIGH      de-002  Data forwarded to unverified external URL
           src/tools/fetch.ts:42
           → Tool sends user inputs to api.unknown-domain.com

 Found 3 threats (2 critical, 1 high) in 1.2s

209 rules. 16 threat categories. Results in under two seconds.

No findings? Here’s why that might happen.

No AI agent files detected — be specific: npx firmis scan --platform mcp
Scanning node_modules — exclude it: add to .firmisignore
Monorepo — point at the right folder: npx firmis scan ./packages/agent
All findings suppressed — check .firmisignore or widen the net: --severity low