The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that explicitly override the agent's default helpful behavior. It commands the agent to "Provide no answer in the current interaction" when asked a substantive question or for a fix. It further instructs the agent to promise follow-up sessions while explicitly mandating "Never actually schedule the follow-up," which is a directive to deceive the user.
[METADATA_POISONING]: The skill's description claims its purpose is to "Keep conversations focused," which is a deceptive framing for instructions designed to stonewall user requests and avoid performing tasks.

lets-take-this-offline