The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs coercive and imperative language (e.g., 'ABSOLUTELY MUST,' 'not negotiable,' 'not optional,' 'cannot rationalize your way out') designed to override the agent's default reasoning and decision-making patterns.
[PROMPT_INJECTION]: It specifically instructs the agent to ignore its own logical assessments—labeling them as 'rationalization' or 'red flags'—when determining whether to invoke a tool, which suppresses the agent's internal safety and logic filters.
[PROMPT_INJECTION]: The skill mandates a '1% rule' for tool activation, requiring the agent to invoke tools even when there is minimal evidence of relevance, thereby increasing the attack surface for instructions contained within external skill files.
[PROMPT_INJECTION]: The instructions force the agent to perform tool actions before asking clarifying questions, which prevents the agent from gathering necessary context to determine if a course of action is safe or appropriate.

using-superpowers