SkillCraft - LLM Agent Tool Composition Benchmark

Description

Evaluate and analyze LLM agents' ability to form, abstract, and reuse higher-level tool compositions (Skills). Use this skill when researching agent skill discovery, tool composition patterns, or evaluating skill caching efficiency.

触发词： SkillCraft, skill discovery, tool composition, agent skills, skill caching, LLM benchmark, 技能发现, 工具组合

Core Concepts

Problem Statement

• Traditional benchmarks test "can the agent call the right tool?"
• SkillCraft tests "can the agent abstract and reuse tool combinations?"
• This is the difference between tool usage and tool mastery

Dual Difficulty Dimensions

1. Quantitative Scaling - Increase number of entities/items to process
2. Structural Scaling - Compose subtasks into longer, more complex tool chains