preflight

Pre-install vetting for agent skills & MCP servers

Check it before your agent runs it.

Paste a link to a skill or MCP server. Preflight reads the code and the instructions, then shows you exactly what it touches, in plain English. Free.

Works with github.com repos, skills.sh listings, and direct .zip archives. For a ClawHub skill, scan its source repo or paste the SKILL.md — ClawHub's API does not serve package downloads.

No link? Paste a SKILL.md instead

How it works

01

We read, never run

The package is downloaded and read as inert text. Static analysis only. Scanned code is never executed, imported, or installed.

02

Every claim has a receipt

Each finding names the file, the line, and the exact text that triggered it. Credential access, exfiltration shapes, shell execution, prompt injection, obfuscation. No vibes, no black-box score.

03

You get a grade

Clean, Caution, Risky, or Hostile, with the subscores behind it and a full permission summary: network, environment, shell, files. Share the report link with anyone.

Recent reports · 285 packages scanned

Browse all reports →