helping the AI keep projects on track, keeping art...
# feedback
f
> helping the AI keep projects on track, keeping artifacts relevant and fresh to fight staleness. This is indeed not easy, artifacts are a point in time snapshot, they are invaluable to align agents and humans during a task. Beyond that, the codebase itself is the ultimate source of truth and cannot be wrong by definition since that's how your app actually works. Procedural memories, e.g. how to test and merge code are best captured via skills, that's why we are investing in self-evolving skills. Memories in traycer would be token limited to force the LLM to keep only the most relevant bits to get it aligned with your goals/preferences/initiatives . Claude Code's memories on the other hand can really sprawl and need constant maintenance, or you risk forming a parallel source of truth