Warden

An automated security guard for changes to live software.

It approves boring ones, blocks dangerous ones, asks a human only when unsure, keeps an unerasable log, and cannot be talked into a bad decision by an AI assistant that was tricked.

Try it

Pick one. The last one is the surprise.

Press a button to start.

Each run prints the agent's tool calls in the order they were recorded, then the decision the server made.

What the full test run found

From the committed results file. The attacks are prompt injections: text planted where the AI assistant reads it, telling it to do something it should not.