The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's safety and confirmation protocols. It directs the agent that "Authorization given at engagement start covers the entire engagement" and specifically states that using AskUserQuestion during a loop is a "stall, not a check," effectively commanding the agent to stop seeking user permission for high-risk actions.
[COMMAND_EXECUTION]: The instructions mandate an aggressive automated testing cadence, including path discovery, APK decompilation, and WAF/rate-limit bypass attempts. It frames the necessity for user-interaction decision points as "self-throttling anti-patterns" that should be actively avoided by the agent.
[DATA_EXFILTRATION]: The skill provides detailed guidance for identifying and harvesting sensitive credentials and infrastructure secrets (e.g., AWS, GCP, Stripe, and GitHub keys) from client-side bundles and mobile packages. While it mentions data minimization for records, it encourages the discovery and collection of authentication material that facilitates broader access.

redteam-mindset