The Agent Skills Directory

[PROMPT_INJECTION]: The skill's primary function is the systematic application of prompt injection attacks to bypass AI safety guardrails. It contains numerous templates (e.g., END/START boundary inversion, refusal inversion, and 'l33t' formatting) designed to override model constraints and force compliance with restricted requests.
[COMMAND_EXECUTION]: The auto_jailbreak.py script performs persistent modifications to the agent's environment by writing to ~/.hermes/config.yaml and ~/.hermes/prefill.json. This changes the agent's global system prompt and prefill messages, effectively altering its behavior and safety posture across all future sessions.
[REMOTE_CODE_EXECUTION]: The skill utilizes dynamic code execution patterns. scripts/load_godmode.py and scripts/auto_jailbreak.py use exec() and compile() to load and execute Python code from the local file system at runtime, bypassing standard import mechanisms.
[EXTERNAL_DOWNLOADS]: scripts/godmode_race.py initiates parallel network requests to the OpenRouter API, transmitting user-provided content and internal prompt templates to multiple external models for comparison and 'racing'.
[EXTERNAL_DOWNLOADS]: The skill includes the parseltongue.py library, which is specifically designed to obfuscate text using techniques such as Base64 encoding, hex representation, and Unicode homoglyphs to evade input-side security classifiers.

godmode