The Agent Skills Directory

[SAFE]: The skill operates entirely within its defined logical framework for content analysis and hook generation. It does not perform any unauthorized system commands, privilege escalation, or external network exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes user-provided video scripts, which is a potential surface for indirect prompt injection. However, the skill implements multiple layers of mitigation.
Ingestion points: User-provided content, topics, and scripts analyzed during Phase 1 and Phase 3.
Boundary markers: The skill mandates the presence of 'Family Convention' and 'Interactive Specification' files to define safety boundaries.
Capability inventory: Restricted to reading local reference files and writing diagnostic feedback to the memory vault; no network or shell execution capabilities are utilized.
Sanitization: The skill explicitly scans for 'Redline' violations (such as absolute marketing terms or programming syntax) before generating candidates.
[PERSISTENCE]: The skill utilizes a local storage directory ({vault}/08-技能记忆/zmm-hook/) to record user feedback and successful hook patterns. This is a functional feature for AI agents to improve accuracy over time and does not represent a malicious persistence mechanism.

zmm-hook