A ShareGPT-format training dataset aggregated, deduplicated, and converted from community-contributed dataclaw datasets (real Claude Code session recordings). Intended for SFT fine-tuning of Qwen3 models to learn Claude Code-style agentic coding behavior: chain-of-thought reasoning combined with tool use.
