Reproduce “Rethinking Agent Trajectory Compression: A Workflow-Aware Perspective” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Qijia Zhuang, Rui Liu, Wenqi Hu, Yuxiang Ren

WAAC reduces model-visible history for long-horizon tool agents by linking fields to downstream dependencies and combining current requirements with policies from prior trajectories. It assigns fields exact, summarized, referenced, or dropped forms while retaining artifacts in runtime state. On PinchBench, its Overall score is 87.6 versus Vanilla’s 72.6, with input-token use 24.6% lower; this evidence applies to represented dependencies and schemas.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.