Exploration and Exploitation Errors Are Measurable for Language Model Agents

Build a diagnostic library that analyzes agent trajectories to detect exploration/exploitation imbalances. Provide a dashboard that visualizes where the agent is getting stuck in decision loops.

Suggested repo: agent-probe

"Quantify your agent's learning behavior without reading thousands of logs."

Estimated effort: 60h