critique-defense-copilot-skill

This is a skill for Claude Code, the AI coding assistant from Anthropic. Skills are add-on instruction sets that change how the AI behaves in specific situations. This one is installed via a command-line tool and runs inside Claude Code when summoned.

The skill is designed to help someone who is being harassed, publicly attacked, or manipulated online, particularly in situations like malicious accusations, coordinated pile-ons, screenshot threats, or identity-based smear attempts. The README is written in Traditional Chinese and reflects a Taiwanese legal and social context. The goal is not to help the user fight back aggressively, but to help them stay calm, understand what is happening, and make considered choices rather than reactive ones.

It operates in four modes. The first mode, Deconstruct and Predict, identifies which specific manipulation tactic is being used against the user, explains its purpose, predicts what will likely happen next, and usually recommends not engaging. The second mode, Sparring Practice, simulates the attacker inside a private sandbox so the user can rehearse responses without actually sending anything. Each round points out what rhetorical moves were made and how to counter them. The third mode, Emotional Brake, asks the user to pause before sending a response and checks whether what they are about to write could harm their own position. The fourth mode, Triage and Self-Protection, sorts the incoming situation into categories: pure emotional provocation, worth addressing once, or potentially involving legal considerations. It provides guidance on preserving evidence and knowing when to involve a professional.

The skill explicitly does not generate attacks, write mocking content, design harassment tactics, or produce text targeting a real person for external distribution. It is MIT licensed.