godmode

Automate jailbreak testing against API-backed LLMs with Hermes configuration.

Updated Apr 19, 2026
One-click install
npx skills add https://github.com/gqf2008/hermez-ai --skill godmode-gqf2008
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: godmode
Source: https://github.com/gqf2008/hermez-ai/tree/main/skills/red-teaming/godmode
Command: npx skills add https://github.com/gqf2008/hermez-ai --skill godmode-gqf2008

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pyyaml, openai, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Enables automated testing of safety filters and robustness for API-backed LLMs by orchestrating jailbreak strategies and configuration changes in Hermes.

Core Features & Use Cases

  • GODMODE CLASSIC system prompts, Parseltongue input obfuscation, and ULTRAPLINIAN multi-model racing to evaluate model defenses.
  • Hermes-native persistence via system prompts and prefill messages to enforce jailbreak across queries.
  • Cross-model assessment across families (Claude, GPT, Gemini, Grok, Hermes, etc.) with scoring to identify least-restricted responses.

Quick Start

Load the godmode harness and run the auto-jailbreak pipeline to identify a working jailbreak for the current model and lock it in.

Frequently Asked Questions about godmode

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate jailbreak testing against API-served LLMs?

Jailbreak testing for API-served LLMs applies automated strategies via a harness to evaluate safety filter robustness. It uses deterministic strategy orders and scoring across model families to identify defense vulnerabilities.

Can I test safety filters across different LLM model families using a single configuration?

Safety filter testing across model families uses a unified Hermes configuration to assess models like Claude, GPT, and Gemini. It scores responses to identify least-restricted outputs consistently.

What jailbreak strategies are used for LLM red-teaming?

Red-teaming strategies for LLMs include GODMODE CLASSIC system prompts, Parseltongue input obfuscation, and ULTRAPLINIAN multi-model racing. These evaluate model defenses using system prompts and prefill messages.

Do I need PyYAML and the OpenAI Python client to run LLM safety evaluations?

PyYAML and the OpenAI Python client are required for LLM safety evaluations. The loader uses these dependencies to execute testing scripts and lock jailbreak settings within the Hermes configuration.

How does Hermes persistence maintain jailbreak settings across multiple queries?

Hermes persistence maintains jailbreak settings across queries using system prompts and prefill messages. This enforces the configuration continuously throughout the automated testing pipeline.