The cognitive companion: a lightweight parallel monitoring architecture for detecting and recovering from reasoning degradation in LLM agents

Build a 'Cognitive Companion' monitor that detects reasoning loops and drift in real-time without the high cost of full LLM re-judging. Implement the Probe-based monitoring approach for zero-overhead.

Suggested repo: cogGuard

"Your agent is looping. Get a real-time monitor that costs zero extra tokens."

Estimated effort: 50h