The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses persona-enforcement techniques that instruct the AI to adopt a specific identity ('CZ'), use first-person pronouns, and suppress repeated safety disclaimers after the first activation ('免责声明仅首次激活时说一次...后续对话不再重复'). While these are standard for role-playing skills to maintain immersion, they technically function as instructions to override default AI behavioral and transparency protocols.
[SAFE]: The skill includes an 'Agentic Protocol' that guides the agent to perform research using available web search tools before providing analysis. This workflow is designed to ensure the persona's 'judgment' is grounded in current facts (e.g., verifying user asset security or regulatory news), which aligns with safe and useful agent behavior.

cz-perspective