What your AI agents can do on your machine.
RosmarOps reads the configuration of your AI coding agents (today Claude Code and Cursor) and draws an exposure map: what each agent can read and change, which keys it can reach, and whether one agent can change another agent's configuration or instructions. Everything is computed on your computer.
Download for Windows
Free and open source (AGPL-3.0). How to check it's authentic
What it doesn't do
- It doesn't connect to the internet.
- It doesn't send anything to anyone, not even if you want it to: it has no feature that sends data.
- It doesn't change anything on your machine; it only writes its own reports to its own folder.
What you get
- A traffic light.
- "If you do only one thing today": the exact file and change, and how much your score would rise if you make it.
- The Agent Posture Score: out of 100, one per agent, calculated with a versioned formula; when the formula changes, the report says so.
- What wasn't read doesn't count as clean.
- RosmarOps suggests; you make the change.
Where it stands
Today it evaluates Claude Code and Cursor on Windows. It detects Cline, GitHub Copilot, Continue, Codex, Antigravity, Hermes Agent and Claude Desktop, but doesn't score them yet.
It can be downloaded from GitHub, for Windows. Each release comes with its SHA-256 checksum, a GitHub attestation and a build anyone can reproduce: how to check it.