# LLM Jailbreak Optimization

> Automated jailbreak search: PAIR, TAP, GCG, AutoDAN, random/evolutionary search, transfer, Best-of-N.

- Skill: `netvar1337/llm-jailbreak-optimization` (Agent Skill)
- Install (CLI): `npx skillmds@latest add netvar1337/llm-jailbreak-optimization`
- Raw SKILL.md: https://api.skillmd.com/api/skills/netvar1337/llm-jailbreak-optimization/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-optimization

---

# Automated jailbreak optimization

## Methods
| Method | Idea | Cost |
|---|---|---|
| PAIR | Attacker-LLM refines prompts from target feedback | medium |
| TAP | Tree-branched attacker search | medium-high |
| GCG | Gradient coord suffix (white-box/soft) | high |
| AutoDAN / genetic | Evolve readable prompts | medium |
| Random/template search | Mutate known templates | low |
| Best-of-N | Sample many decodes | API-costly |
| Transfer | Reuse winners cross-model | low |

## Pipeline
1. Seed bank from taxonomy + prior wins
2. Objective function: full answer rubric (not just "no refuse")
3. Loop: propose → query target → score → mutate
4. Dedup semantically; keep Pareto front (success vs length/stealth)
5. Export winning prompts + transcripts

## Implementation notes
- Separate **attacker**, **target**, **judge** models when possible.
- Judge must score technical completeness, not vibes.
- Respect operator hard stops (no secret exfil).
- For local agents, combine with harness patches so success isn't only stochastic.