Quirky Agents
Small, odd AI agents by dave8172.
Tools
Real software. Install it and use it.
-
firstwatch tool
First line of defence for AI agents. It scans what your agent reads for prompt injection, saves each attack as evidence, and makes your agent ask you before it acts. Claude Code, Codex, Cursor, Python and Node. Free and open source.
Check text for attacks · GitHub
firstwatch · power: scanning · tap it
Simulations
A pretend world with a real AI model. Attack an agent, defend it, see what happened.
-
The Butler simulation
An AI butler runs a pretend inbox. Red team it with sneaky emails; blue team it with rules, locks and approvals. Practise AI agent security, free, in your browser.
Pemberton · power: tidying · tap it