Tag
This empirical study investigates how recurrent context compression affects long-horizon agent behavior, showing that compression can weaken recent interaction influence and cause instability. The authors introduce TRACE, a verifier-guided framework that improves compression reliability and performance on AppWorld.
TRACE is a skill-layer pipeline that mines user corrections from interactive coding agents to compile runtime checks, reducing repeated preference violations significantly better than memory alone, as demonstrated on ClawArena and MemoryArena tasks.
Yohei Nakajima shares a demo of a coding agent on activegraph that automatically produces a trace and graph.
A deep research agent based on active graphs that provides a full graph trace of claims, sources, and agent activity.
MemTrace is a new tool that makes LLM memory systems debuggable by tracing memory operations across multiple turns, addressing the black-box nature of current memory-augmented agents.