Image Editing

Pass

Audited by Gen Agent Trust Hub on Mar 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes shell command examples for running local image processing tools like rembg, iopaint, and realesrgan. These commands are intended for legitimate image manipulation tasks.
  • [EXTERNAL_DOWNLOADS]: The skill suggests installing several standard Python libraries such as diffusers, rembg, and gfpgan from public registries. It also references well-known and trusted external services via their APIs, including OpenAI, Stability AI, and ClipDrop.
  • [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes user-supplied images and text descriptions.
  • Ingestion points: User input enters the system through images and descriptive prompts in files such as inpainting.md, style-transfer.md, and upscaling.md.
  • Boundary markers: There are no clear delimiters or "ignore previous instructions" safety wrappers around user-provided data.
  • Capability inventory: The skill can perform network requests (requests.post) and execute local scripts or CLI tools across its various modules.
  • Sanitization: No validation or sanitization logic is demonstrated for handling untrusted input strings or image file content.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 17, 2026, 07:18 PM
Security Audit — agent-trust-hub — Image Editing