stable-diffusion-image-generation
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text prompts and image data which are interpolated into model inference pipelines. Ingestion points: Text prompts (prompt, negative_prompt) and image files (image, mask_image, init_image) are ingested across multiple examples in SKILL.md and references/advanced-usage.md. Boundary markers: No specific boundary markers or 'ignore instructions' warnings are present in the prompt templates. Capability inventory: The skill performs file system writes (image.save()) and network operations to download model weights (DiffusionPipeline.from_pretrained()). Sanitization: No sanitization or filtering of the input prompts or image metadata is performed before processing.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install several external packages and model weights from well-known repositories. Package installation: Instructions include installing diffusers, transformers, accelerate, torch, xformers, fastapi, and others via pip. Remote code execution: references/troubleshooting.md includes an instruction to install xformers directly from Meta's official GitHub repository. Model downloads: Numerous examples fetch pre-trained weights from the official Hugging Face Hub (e.g., stabilityai/stable-diffusion-xl-base-1.0, runwayml/stable-diffusion-inpainting).
- [COMMAND_EXECUTION]: The skill includes shell commands for installation and environment configuration. Evidence: pip install and huggingface-cli commands are provided in SKILL.md and references/troubleshooting.md for setting up the local environment.
Audit Metadata