stable-diffusion-image-generation

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill follows standard practices for using well-maintained machine learning libraries from trusted organizations like Hugging Face, Meta (xformers), and StabilityAI.
  • [SAFE]: External downloads and references (e.g., Hugging Face Hub, PyTorch registries) are directed toward well-known, established technology services.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides an interface for processing untrusted user input (prompts and images) through diffusion pipelines.
  • Ingestion points: User-provided text prompts and initialization/mask images in SKILL.md and references/advanced-usage.md.
  • Boundary markers: Absent in the provided boilerplate code.
  • Capability inventory: The skill facilitates file system writes (image.save) and network communication for retrieving model weights and deploying APIs.
  • Sanitization: No specific prompt sanitization is implemented; the troubleshooting guide notes that pipe.safety_checker = None can be used to bypass model safety filters, which should be evaluated by the user based on their deployment context.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 03:11 PM
Security Audit — agent-trust-hub — stable-diffusion-image-generation