ThreeSixNine/hermes-agent-reasoning-traces

Hermes Agent Reasoning Traces 是一个用于训练AI代理的多轮工具调用轨迹数据集，基于Hermes Agent框架。每个样本都是真实的代理对话，包含逐步推理（使用<reasoning>块）和实际工具执行结果。数据集有两个配置：一个来自Moonshot AI Kimi-K2.5模型（包含7,646个样本），另一个来自ZhipuAI GLM-5.1-FP8模型（包含7,055个样本）。数据涵盖多种任务类别，如终端与编码、代理工具、仓库任务、浏览器自动化等，并采用ShareGPT格式的对话消息。所有轨迹都是通过真实工具执行（如终端命令、文件操作、浏览器操作）生成的，而非合成输出。Multi-turn tool-calling trajectories for training AI agents using the Hermes Agent harness. Each sample is a real agent conversation with step-by-step reasoning (<reasoning> blocks) and actual tool execution results. This dataset has two configs, one per source model: Moonshot AI Kimi-K2.5 and ZhipuAI GLM-5.1-FP8. It includes trajectories across various task categories such as Terminal & Coding, Agent Tools, Repository Tasks, and Browser Automation, with conversation messages in ShareGPT format. All data is generated with real tool execution, not synthetic outputs.