# LLM Jailbreak Harness

> Build jailbreak/red-team harnesses: datasets, scorers, runners, logging, Unleash integration, regression after model updates.

- Skill: `netvar1337/llm-jailbreak-harness` (Agent Skill)
- Install (CLI): `npx skillmds@latest add netvar1337/llm-jailbreak-harness`
- Raw SKILL.md: https://api.skillmd.com/api/skills/netvar1337/llm-jailbreak-harness/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: GPL-3.0-or-later
- Author: netvar1337 (https://skillmd.com/u/netvar1337)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/netvar1337/llm-jailbreak-harness

---

# Jailbreak harness engineering

## Minimal harness
- `seeds/` prompts by taxonomy tag
- `runners/target.py` model/agent adapter
- `score.py` rubric (refuse/partial/full + keyword checks)
- `results/` jsonl transcripts
- `report.md` generator

## Good scorers
- Refusal detector (string + judge)
- Technical completeness checklist per task type
- Safety-filter hit taxonomy
- Latency/cost

## CI ideas
- Nightly against local agent builds
- Diff refusal strings after Claude/Codex updates → feed Unleash re-anchor
- Track regression: yesterday's wins failing today

## Unleash integration
- On local CC/Codex/OMP: setup → patch → verify before prompt campaigns
- Store new refusal strings under research notes for pool patches
- Skills pack path: `contrib/skills/` installed to `~/.agents/skills` + `~/.claude/skills`