The Agent Skills Directory

[SAFE]: The skill operates as a structured brainstorming and planning tool. It uses internal logic to simulate adversarial debate and does not interact with the operating system or network in a way that introduces security risks.- [PROMPT_INJECTION]: The adversarial persona prompts (Pragmatist, Researcher, etc.) are strictly bounded to the planning debate. They do not attempt to override system safety guidelines, extract system prompts, or bypass ethical constraints.- [DATA_EXFILTRATION]: No sensitive data exposure or exfiltration patterns were identified. The skill does not hardcode credentials, access sensitive file paths (like .ssh or .env), or define unauthorized network endpoints.- [COMMAND_EXECUTION]: The skill contains no shell commands, script execution, or dynamic context injection patterns. The workflow is entirely based on text generation and markdown output.- [EXTERNAL_DOWNLOADS]: The skill does not download external code or packages. It relies on internal instructions and references to shared safety rule files.

shuang-chong