ozu-perspective

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXPOSURE]: The skill references local filesystem paths for a knowledge base located at ~/Documents/白梦客知识库/wiki/美学配方/. These are provided as citations for the agent to reference supplementary aesthetic guidelines. While referencing specific paths in the home directory is a precise data access pattern, these paths target documentation rather than sensitive system or credential files.
  • [INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion point for untrusted user input describing cinematic scenes or creative themes.
  • Ingestion points: User queries about visual narrative, aesthetic direction, and family/ritual themes in SKILL.md.
  • Boundary markers: The skill defines a self-check workflow ([小津自检]) to ensure output alignment but does not include explicit delimiters to isolate user input from instructions.
  • Capability inventory: The skill does not request access to high-privilege tools such as shell execution, network operations, or file-system writing.
  • Sanitization: No specific sanitization or filtering logic is defined for the content of user-provided scene descriptions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 03:24 AM
Security Audit — agent-trust-hub — ozu-perspective