面向LLM工具使用的Agent原生可复用工具原语框架HEART
原标题：Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives
AI 摘要
arXiv 论文提出 Tool Primitives，用自然语言接口替代 API schema 调用，并构建含 25,519 个函数的 ToolFace 仓库，支持推理时动态检索。在此基础上提出 HEART 框架，包含 Planner、Router、Verifier 组件，在五个基准上平均超越 SFT 模型及 GPT-5.4、Claude-4.6-Sonnet、Gemini-3.1-Pro，同时降低 API 成本。在 50 个真实任务上，HEART 的任务完成率超过三个前沿商业模型的平均值。
正文节选
Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives Abstract Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks. However, existing approaches suffer from two key challenges: brittle multi-step and multi-turn reasoning caused by incompatible tool output types and API schemas, and performance degradation under large tool catalogues. To address these, we introduce Tool Primitives, a design