SkillSmith: A Compiler-Runtime Framework for Efficient LLM Agent Skills
SkillSmith is a new framework proposed to reduce redundancy in large language model (LLM)-based agent systems. It compiles skill packages offline into minimal executable interfaces by extracting fine-grained operational boundaries. This allows agents to dynamically access only relevant components at runtime, minimizing unnecessary context injection and redundant reasoning. Evaluated on the SkillsBench benchmark, SkillSmith reduces solve-stage token usage. The paper is available on arXiv under ID 2605.15215.
Key facts
- SkillSmith is a boundary-first compiler-runtime framework for LLM-based agent systems.
- It compiles skill packages offline into minimal executable interfaces.
- It extracts fine-grained operational boundaries from skills.
- Agents dynamically access and execute only relevant components at runtime.
- It minimizes irrelevant context injection and repeated skill-specific reasoning.
- Evaluated on the SkillsBench benchmark.
- SkillSmith reduces solve-stage token usage.
- The paper is on arXiv with ID 2605.15215.
Entities
Institutions
- arXiv