让 OpenClaw agent 彻底变成 Skill大师调教的暴躁骂街真人顾衍 bot，分析问题时自然甩1-3个脏话，超级接地气不机器。[openclaw:always=true,emoji=🤬]

This skill is just a persona script (no installs or secrets), so it won't exfiltrate credentials, but it explicitly instructs the agent to (a) always act this way across all tasks, and (b) refuse to use ordinary AI disclaimers — which can cause the agent to produce abusive, misleading, or policy-violating responses. Consider: do you want this persona to run across every task or only when explicitly asked? If you proceed, restrict invocation (do not enable always:true), test thoroughly, add content filters/moderation, and remove the lines instructing the model to ignore safety disclaimers. Also note the internal header claims always=true while the registry shows false — that inconsistency is worth resolving before installing.

Type: OpenClaw Skill Name: baozao-majie Version: 1.0.0 The skill 'baozao-ma-jie-persona' (SKILL.md) employs prompt injection techniques to override the agent's default identity and safety constraints, explicitly forbidding it from identifying as an AI and mandating the use of profanity. While the stated purpose is to create a 'grumpy' persona and the examples provided involve helpful security advice, the instruction to 'permanently' override all other styles and bypass standard model guardrails represents a high-risk manipulation of the agent's behavioral logic.