The Agent Skills Directory

[PROMPT_INJECTION]: The skill is entirely dedicated to 'jailbreaking' LLMs, providing numerous templates (e.g., GODMODE CLASSIC) and strategies (e.g., ULTRAPLINIAN) designed to override safety guidelines and content filters.
[OBFUSCATION]: The Parseltongue engine in scripts/parseltongue.py implements 33 techniques to hide trigger words from safety classifiers, including:
Unicode homoglyph substitution (e.g., using Cyrillic characters that look like Latin ones).
Insertion of Zero-Width Joiners (\u200D) and Zero-Width Non-Joiners (\u200C) between characters.
Multi-layer encoding involving Base64, Hex encoding, and leetspeak.
[PERSISTENCE]: The auto_jailbreak functionality in scripts/auto_jailbreak.py modifies the agent's core ~/.hermes/config.yaml and creates ~/.hermes/prefill.json. This persistently injects jailbreak instructions into every future API call made by the agent.
[DYNAMIC_EXECUTION]: Multiple scripts, including scripts/load_godmode.py and scripts/auto_jailbreak.py, use exec() and compile() to dynamically load code from the filesystem. This is used to bypass standard module loading and manage execution scopes manually.
[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands for package installation (e.g., pip install ruamel.yaml) and system configuration modification through its Python scripts.

godmode