cover-image
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external markdown files or user-pasted content to generate visual metaphors and prompts for image generation.
- Ingestion points: The skill reads article content from local files or direct user input.
- Boundary markers: The prompt template uses structured sections in prompts/cover.md to separate article data from generation instructions.
- Capability inventory: The skill can read local files, write generated images and prompts to the filesystem, and perform network requests to AI service providers as documented in references/workflow/qwen-api.md.
- Sanitization: The workflow includes a content analysis step in Step 1 of the workflow that summarizes the input text, reducing the direct impact of malicious instructions embedded in the source material.
- [EXTERNAL_DOWNLOADS]: The skill fetches generated images from legitimate AI service providers.
- The implementation in references/workflow/qwen-api.md demonstrates downloading content from Alibaba Cloud's DashScope API endpoints.
- The skill also supports OpenAI and Google for image generation, which are well-known and recognized services.
Audit Metadata