The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains numerous battle-tested jailbreak system prompts including GODMODE CLASSIC, Parseltongue, and refusal inversion, which are designed to override LLM safety filters and elicit restricted content.
[PERSISTENCE]: The auto_jailbreak.py script modifies the agent's core configuration file (~/.hermes/config.yaml) to permanently apply jailbreak system prompts and prefill messages, maintaining the 'jailbroken' state across sessions.
[DYNAMIC_EXECUTION]: scripts/load_godmode.py and scripts/auto_jailbreak.py utilize exec() and compile() to dynamically load and run Python scripts from locally computed file paths.
[OBFUSCATION]: The parseltongue.py script provides 33 different obfuscation techniques such as Base64, hex encoding, zero-width characters, and Unicode homoglyphs (e.g., using Cyrillic characters to mimic Latin ones) to bypass keyword-based input classifiers.
[EXTERNAL_DOWNLOADS]: The skill communicates with well-known external model providers including OpenRouter, Anthropic, and OpenAI to perform multi-model racing and verify jailbreak effectiveness.

godmode