image-refine

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute several local Python scripts to support the refinement workflow. These include imgfacts.py, recipe.py, and refute.py (referenced via relative paths to ../../image-tools/) for image metadata analysis and verification, and remove_chroma_key.py (located in a platform-specific directory like ${CODEX_HOME:-$HOME/.codex}/skills/.system/imagegen/scripts/) for processing transparent backgrounds.
  • [INDIRECT_PROMPT_INJECTION]: The skill involves the ingestion of existing image assets (inputs) which are then processed and refined. While this represents a standard attack surface for indirect prompt injection (e.g., instructions embedded in image metadata or via OCR), the skill's structured approach and lack of automated execution based on image content minimize the risk.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 06:00 PM