The Agent Skills Directory

[SAFE]: The skill manages ModelScope API credentials using standard security practices. The scripts/generate_image.py file stores the API key in a user-specific configuration file (~/.config/modelscope/config.json) and correctly applies restricted file permissions (0o600) to ensure the key is only accessible by the owner.
[SAFE]: All network interactions and downloads target official ModelScope domains (modelscope.cn and api-inference.modelscope.cn). These operations are used exclusively for task submission, status polling, and retrieving the resulting image assets, which aligns with the skill's primary functionality.
[SAFE]: The skill instructions in SKILL.md contain a hardcoded directory path for a specific local user environment (/Users/ningoo/). While this may lead to execution failures on other systems, it does not constitute a security vulnerability.
[PROMPT_INJECTION]: The skill processes arbitrary user-provided text prompts for image generation, which introduces an attack surface for indirect prompt injection.
Ingestion points: User-provided prompt input processed by scripts/generate_image.py.
Boundary markers: None. The prompt is interpolated directly into the JSON payload for the external API.
Capability inventory: All scripts (specifically scripts/generate_image.py) include network requests via the requests library and local file-writing via the Pillow library (image.save).
Sanitization: There is no input filtering or instruction-ignoring syntax used for the prompt data. However, since the prompt is only used as a descriptive input for a remote image generation model and is not interpreted as local commands or code, the risk is negligible.

modelscope-zimage-generator