baoyu-cover-image
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted article content and reference images to derive visual metaphors and prompt details for image generation.
- Ingestion points: Processes local markdown files provided by the user and reference images saved to a dedicated
refs/directory. - Boundary markers: Utilizes structured Markdown headers within the generated prompt template (e.g.,
# Content Context,# Visual Design) to separate user-provided content from generation instructions, though it lacks explicit instructions to ignore potentially embedded commands in the source article. - Capability inventory: The skill is designed for image generation and local file management (saving prompts, source backups, and images). It does not exhibit capabilities for network exfiltration or arbitrary code execution.
- Sanitization: The skill provides instructions for the agent to generate identifiers (slugs) in 'kebab-case', which serves as a structural constraint for generated filenames and directory paths.
Audit Metadata