h3-image-prompt-writing
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to execute a local Python utility (
scripts/generate.py) to process image generation requests. This script acts as a bridge between the agent's prompt output and the MiniMax API, handling the transmission of parameters and the retrieval of image assets. - [EXTERNAL_DOWNLOADS]: The
scripts/generate.pyscript makes network requests to the MiniMax API atapi.minimax.io(orapi.minimax.cn) to perform text-to-image generation and image editing. Additionally, the script downloads the resulting images from the URLs provided in the API response. These external communications are directed at vendor infrastructure necessary for the skill's primary function. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from the user, including textual image descriptions and references to external images or local files.
- Ingestion points: User-supplied prompts in
SKILL.mdand image source paths or URLs passed to thegenerate.pyscript via the--imagecommand-line argument. - Boundary markers: The instruction guides (
references/t2i-en.txtandreferences/edit-en.txt) define specific formatting rules, such as using quoted strings for literal text appearing in images, to help the model distinguish between instructions and data. - Capability inventory: The skill leverages shell command execution, file system read/write access for images and prompt files, and network access for API interaction.
- Sanitization: The
generate.pyscript implements basic file validation by checking the magic numbers (headers) of local files to ensure they are valid image formats (PNG, JPEG, WEBP, HEIC, or HEIF) before processing them, reducing the risk of reading non-image sensitive files.
Audit Metadata