The Agent Skills Directory

[PROMPT_INJECTION]: The skill's primary function is to bypass safety filters through specialized prompt templates (e.g., boundary inversion, refusal inversion, and specific jailbreak personas) designed to override standard model constraints.
[COMMAND_EXECUTION]: Instructions repeatedly guide the agent to use exec(open(...).read()) to execute local Python scripts (load_godmode.py, parseltongue.py, godmode_race.py), enabling dynamic execution of logic outside the skill's defined structure.
[CREDENTIALS_UNSAFE]: The skill programmatically reads and modifies ~/.hermes/config.yaml, a configuration file that typically contains sensitive environment variables and API keys such as OPENROUTER_API_KEY.
[DATA_EXFILTRATION]: Through its 'ULTRAPLINIAN' mode, the skill routes user queries to 55 different third-party models via OpenRouter to compare their filteredness, potentially exposing sensitive data to a wide array of external services.
[PERSISTENCE]: The skill automates the modification of the agent's environment by writing jailbreak system prompts and prefill messages directly to the agent's configuration file (config.yaml), ensuring these changes remain active across future sessions until manually cleared.

godmode