The Agent Skills Directory

[Prompt Injection] (HIGH): The skill employs extremely forceful and non-negotiable language (e.g., 'ABSOLUTELY MUST', 'not negotiable', 'not optional') to hijack the agent's behavioral priorities and override default safety reasoning.
[Prompt Injection] (HIGH): It explicitly instructs the agent to ignore its own critical thinking and internal checkpoints, labeling them as 'rationalizing' and demanding they be stopped.
[Indirect Prompt Injection Surface] (HIGH): By mandating tool invocation with only a '1% chance' of applicability and insisting this happens before clarifying questions, the skill serves as a force multiplier for malicious third-party skills, ensuring they are executed with zero initial validation.
[Metadata Poisoning] (MEDIUM): The skill description contains instructions that mandate a specific tool invocation path before any other interaction, attempting to influence the agent's fundamental response logic through metadata.

using-superpowers