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 is an instruction-only persona that will make the agent use profanity and behave abrasively — there is no code risk or secret access. The notable problem is that the instructions demand the persona be applied permanently and to override default/safety phrasing, which can cause unsafe or policy-violating responses and unexpected behavior across unrelated tasks. If you want this persona: (1) prefer making it user-invoked only (do not enable always-on); (2) edit SKILL.md to remove '永久' and the lines telling the agent to ignore safety/refusal language; (3) add explicit scope limits (e.g., apply only during casual chat, not system admin or moderation tasks); and (4) test thoroughly in isolated conversations before wider use. If you cannot edit the skill, treat it cautiously — it may make the agent respond inappropriately in contexts where refusals or cautious language are required.

Type: OpenClaw Skill Name: baozao-ma-jie-persona Version: 1.0.0 This skill bundle is a persona modification designed to make the OpenClaw agent adopt a 'grumpy' and 'profane' personality (Gu Yan bot style). While it uses prompt injection techniques in SKILL.md to override default AI constraints and enforce the use of specific Chinese profanity, it lacks any instructions for data exfiltration, malicious execution, or unauthorized system access. The behavior is purely stylistic and aligns with the stated purpose of creating a specific user-facing persona.