PACR: Secure runtime defense layer for OpenClaw AI agents
Ongoing research project securing OpenClaw against prompt injection and unsafe tool execution. Built a PACR security module, tested trusted and malicious scenarios, added decision logging, and integrated policy checks into the exec approval runtime path to block risky commands before execution.
