BoxPwnr — LLM-Driven CTF/Pentest Runner (Docker)

Practical Notes

• GitHub: 393 stars · 46 forks; pushed 2026-05-12 (verified via GitHub API).
• README requires cloning with submodules and running uv sync to create .venv before uv run boxpwnr ….
• README documents hard limits like --max-turns, --max-cost, and execution timeouts (default 30s, max 300s).

Main

A useful BoxPwnr pattern for teams:

1. Define a target catalog (labs/benchmarks) and run with consistent flags (--max-turns, --max-cost) so results are comparable.
2. Keep the executor boundary strict: everything runs inside the Docker environment; your host stays clean.
3. Use --generate-progress / --resume-from to create handoffs between attempts instead of restarting from scratch.
4. When a task is “almost solved”, switch to manual follow-up (or keep the target running) and treat the LLM as a coordinator, not a miracle worker.

This keeps experimentation fast while still producing artifacts you can review later.

FAQ

Q: Do I need Docker? A: Yes. README says BoxPwnr requires Docker to be installed and running.

Q: How do I control cost/time? A: Use --max-cost, --max-turns, and execution timeout flags described in the README.

Q: What’s the minimal run command? A: After uv sync, run uv run boxpwnr --platform htb --target meow (example from README).