# drako

An offline AI agent security and governance scanner that checks a codebase and its installed agents against 80 deterministic rules, with no LLM involved in the evaluation.

drako scans source code, and the agents installed in Claude, Cursor, VS Code and Windsurf, to produce a governance score and a determinism score, and separates reachable risks, such as a dangerous tool an agent actually calls, from theoretical ones through static analysis.

Its 80 rules span 16 categories: security (hardcoded keys, prompt injection, code execution, filesystem and network access), governance (missing audit logging or human-in-the-loop review), EU AI Act compliance gaps, missing determinism controls such as temperature or retry limits, vendor concentration, and multi-agent topology risks, among others, all through pure AST analysis with no network access.

A standalone `drako bom` command produces an inventory of an agent's tools with no runtime or network use; full runtime enforcement is available as a separate paid tier. It installs with pip.

Website: https://www.getdrako.com/