Plimsoll: an agent skill for testing prompt injection, leaks, and tool abuse
I’ve been working on LLM/agent security for a while now, mostly around prompt injection, jailbreaks, leaks, tool abuse, and where the actual security boundary sits once a model starts using tools.
Getting accepted into Anthropic’s Cyber Verification Program gave me a bit more room to push that work further, and I’ve been gradually turning it into Plimsoll.
It’s an open-source agent skill for red-teaming LLM apps and agents.