An instruction-tuning dataset that teaches an LLM to operate
burpwn — a transparent intercepting
proxy and rootless sandbox for AI-driven web pentesting on Linux — across three
interfaces: instruction-style CLI prose, real Bash tool calls (how an
agent actually runs burpwn from a CLI session / under the PreToolUse hook), and
the MCP (Model Context Protocol) tool interface. Roughly half the records are
genuine multi-turn… See the full description on the dataset page:
https://huggingface.co/datasets/own2pwn-fr/burpwn-usage.
