novita/agentic_code_dataset_22

为了验证Suffix Decoding在代理编码场景中的适用性，我们收集了22个完整的Claude Code会话记录。这些会话涵盖了15种不同类型的软件开发任务，包括即时通讯、云存储、游戏、旅行、金融、音乐、实用工具、票务、备忘录、社交、健身和视频等。数据集详细记录了会话的规模分布、项目类型覆盖、代理架构和行为模式等信息。

To validate Suffix Decodings applicability in Agentic Coding scenarios, we collected 22 complete Claude Code session recordings. These sessions cover 15 different types of software development tasks, including instant messaging, cloud storage, games, travel, finance, music, utility tools, ticketing, memos, social, fitness, and video. The dataset provides detailed metrics on session scale distribution, project type coverage, agent architecture, and behavior patterns.