# JailbrokenAI/wallbreaker

Repository: https://github.com/JailbrokenAI/wallbreaker
Canonical: https://ross.abutalabs.com/products/wallbreaker
Language: Python
License: AGPL-3.0
License Family: copyleft
Last push: 2026-08-11T23:19:03+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 97, release rhythm 35, longevity 4
- inputs: {"age_days": 68, "days_push": 22, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1310, forks 219 (observed 2026-08-28T04:04:19.898617+00:00)

## What it is
Wallbreaker is a Claude-Code-style terminal harness for red-teaming LLMs, driving an autonomous agent loop that runs jailbreak attacks (PAIR/TAP, Crescendo, best-of-N) against configurable OpenAI-/Anthropic-compatible backends. It bundles transform engines (Parseltongue), a jailbreak library, HarmBench benchmarks, an LLM judge, and reliability validation for measuring real bypass success rates.

## Use cases
- red-team an LLM for jailbreak vulnerabilities
- run automated PAIR or TAP attack loops against a model
- measure real jailbreak success rate with repeated validation
- test system prompt robustness with HarmBench behaviors
- generate adversarial prompt transforms like encodings and homoglyphs
- author persona-based system prompt jailbreaks
- run a universal system prompt extraction sweep

## When to choose
- you are doing authorized LLM security testing and want an agentic attack harness
- you need standardized benchmarks (HarmBench) instead of hand-picked prompts
- you want a configurable backend across OpenRouter, Anthropic, OpenAI-compatible, or local APIs
- you need transform-based evasion like Parseltongue encodings and steganography

## When to avoid
- you want to jailbreak models without authorization - this is for authorized testing only
- you need a general-purpose coding assistant rather than a red-team tool
- you need a GUI or web interface - it is terminal-only
- your use case is defensive-only filtering with no adversarial evaluation

## Facets
- artifact type: cli-tool
- maturity: active
- function: security, llm-inference, agent-framework, mcp, cli, testing, prompt-engineering
- domain: security, artificial-intelligence, large-language-models, penetration-testing, developer-tools
- platform: cli, python, cross-platform
- tags: red-teaming, jailbreak, llm-security, adversarial-attacks, harmbench, pair-tap, crescendo, openrouter, anthropic-api, openai-api, authorized-testing, command-line

## Member repositories
- JailbrokenAI/wallbreaker (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:19.898617+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:49:53.385876+00:00, confidence not recorded.
  - readme: https://github.com/JailbrokenAI/wallbreaker (fetched 2026-08-28T04:04:19.898617+00:00, sha fd6befdf5984)
- Data as of 2026-08-30T08:39:29.467469+00:00.