A manually curated SFT dataset for training autonomous bug bounty hunting agents. 1.5k hand-crafted penetration testing sessions spanning reconnaissance, exploitation, CVSS scoring, and remediation -- designed for fine-tuning models like Qwen 3.6 27B into capable, tool-driven security agents.
OpenAI SFT format -- each sample contains a messages array with system, user, assistant, and tool roles.
{"messages": [
{"role": "system"… See the full description on the dataset page:
https://huggingface.co/datasets/yamura4/warden01.
