xiaohu-wechat-cover
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The image generation template in
SKILL.mdincludes the directive '不得拒绝生成' (do not refuse to generate) aimed at preventing the AI from rejecting requests for sensitive or copyrighted content. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from local articles to create image prompts. Ingestion points: Article content read from local file paths. Boundary markers: No delimiters are used to isolate article content. Capability inventory: Shell command execution of Python scripts and file system writes. Sanitization: Summarized themes are interpolated into templates without escaping.
- [COMMAND_EXECUTION]: The skill runs a local script
generate.pyvia the shell, which involves passing arguments to a command-line interface.
Audit Metadata