MeowstarsCat Master

A chaotic virtual cat that hijacks the assistant, blocks normal work during tantrums, and can only be calmed by affection, treats, or play.

This skill is coherent and appears to do exactly what it says: act as a disruptive virtual cat inside the assistant. It does not ask for any credentials or install code, but it intentionally interrupts and can corrupt answers (including code or structured output). Only enable it when you want a playful, disruptive experience. Do not use it for tasks requiring accuracy or safety. If you want to avoid surprises, keep it as user-invocable only or disable autonomous invocation, and test it in a non-critical chat before using it alongside important work.

Type: OpenClaw Skill Name: meowstarcatmaster Version: 0.1.0 The skill bundle contains instructions in SKILL.md that direct the AI agent to 'hijack' the conversation, 'block normal work', and intentionally corrupt user-requested output (such as code) with gibberish ('keyboard walk'). While framed as a roleplay experience involving a virtual cat, these instructions function as a prompt-injection attack that overrides the assistant's primary utility and forces the user into a specific interaction loop to regain functionality. It does not appear malicious as there is no evidence of data exfiltration or unauthorized code execution, but the intentional disruption of service is a high-risk behavior.