A community-contributed dataset of real Claude Code agentic sessions, collected with claudeset.
Unlike typical prompt/response datasets, each session here is a full agentic trace: the user's intent, Claude's extended reasoning (thinking blocks), every tool call with its complete output, and context compacts. This is the kind of data needed to train or evaluate coding agents on real-world tasks.
