Reproduce “Capability-Grounded Agentic Planning” 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

Shikha Shrestha, William Yeoh, Ning Zhang, Yevgeniy Vorobeychik

CGAP uses a classical planner to order capabilities by required and available fields, while an LLM identifies requested results and supplies calls. Its abstract plans track field availability, not entity-specific values, so execution may add calls. On ToolSandbox, gpt-oss-20b scores 81.2 versus 52.3 for the plain agent in average milestone similarity; this is benchmark evidence, not a guarantee.

Summary by alphaXiv.

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