Harnessing LLM Agents with Skill Programs
The paper introduces HASP, a framework designed to enhance LLM agents by equipping them with executable Program Functions derived from past experiences. This approach aims to provide active intervention in agent decision-making processes, improving performance on complex tasks. Empirical results demonstrate significant performance gains in web-search, math reasoning, and coding tasks compared to existing methods.
- ▪HASP upgrades skills into executable Program Functions that modify agent actions during failure-prone states.
- ▪The framework can be applied at inference time, during post-training, or for self-improvement.
- ▪Empirical results show a 25% improvement in web-search reasoning performance compared to the ReAct Agent.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.17734 (cs) [Submitted on 18 May 2026] Title:Harnessing LLM Agents with Skill Programs Authors:Hongjun Liu, Yifei Ming, Shafiq Joty, Chen Zhao View a PDF of the paper titled Harnessing LLM Agents with Skill Programs, by Hongjun Liu and 3 other authors View PDF HTML (experimental) Abstract:Equipping LLM agents with reusable skills derived from past experience has become a popular and successful approach for tackling complex and long-horizon tasks. However, such lessons are often encoded as textual guidance that remains largely advisory, lacking explicit mechanisms for when and how to intervene in the agent loop.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
