rekurse

Rekurse helps coding agents avoid repeating the same failed approach. When an agent gets stuck, Rekurse records the session and creates checkpoints before each user message. It replays the session from earlier checkpoints, with and without candidate lessons, to find where the agent stops recovering on its own. A lesson is stored only after it improves recovery against a fresh baseline and is validated on a different task. MongoDB Atlas is the system of record for sessions, checkpoints, replay results, lessons, and final reports. Atlas Vector Search detects duplicate lessons so memory stays compact as the system learns over time. Instead of treating every agent reflection as truth, Rekurse turns memory into a measurable hypothesis: replay it, compare outcomes, validate it on new work, then adopt or reject it.