CrabTrap: An LLM-as-a-judge HTTP proxy to secure agents in production

Create an extensible middleware for agentic workflows to intercept and sanitize prompts and tool outputs in real-time. This provides the 'human-in-the-loop' layer necessary for production-grade agent security.

Suggested repo: crab-proxy

"Stop your agents from hallucinating (or lying) to your production database."

Estimated effort: 40h