Paper Overview
Field: Machine Learning (ML) Authors: Duling Xu, Zheng Chen, Zaifeng Pan Published: 2025-05-15 arXiv: 2505.10889
Summary
Skills have been widely adopted in large language model (LLM)-based agent systems across various domains. In existing frameworks, skills are typically injected into the agent reasoning loop as contextual guidance once matched to a runtime task, enabling specialized task-solving capabilities.
The authors identify two major sources of redundancy in this execution paradigm:
1. Irrelevant context injection — entire skill packages are injected even when only parts are relevant. 2. Repeated skill-specific reasoning and planning — the agent re-derives the same planning logic across runs.
SkillSmith
SkillSmith is a boundary-first compiler-runtime framework that compiles skill packages offline into minimal executable interfaces. By extracting fine-grained operational boundaries from skills, SkillSmith enables agents to dynamically access and execute only the relevant components at runtime, minimizing unnecessary context injection and redundant reasoning overhead.
Results on SkillsBench
Compared with using raw skills, SkillSmith achieves:
- 57.44% reduction in solve-phase token usage
- 42.99% reduction in thinking iterations
- 50.57% reduction in solve time (2.02x faster)
- 57.44% reduction in token-proportional monetary cost
Source code and data are available via the arXiv page: 2505.10889.
*Auto-collected on 2026-05-19.*