furniture-room-scenes

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The Setup section contains a shell command for users to add a vendor-specific MCP server. This command configuration allows the AI to utilize specific tools provided by the author to perform image manipulations. Evidence: claude mcp add --transport http apimage https://mcp.apimage.org/mcp ...\n- [EXTERNAL_DOWNLOADS]: The skill references a remote endpoint to provide extended tool capabilities. This is a vendor-owned resource used to fetch or interact with image processing infrastructure. Evidence: Connection to https://mcp.apimage.org/mcp.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted reference images, which serves as a potential surface for instructions embedded in data. This is documented as a standard architectural risk for image-processing workflows.\n
  • Ingestion points: reference_images parameter in generate_image calls within SKILL.md.\n
  • Boundary markers: No specific delimiters or "ignore embedded instructions" markers are used to wrap the reference data.\n
  • Capability inventory: The skill utilizes generate_image, remove_background, create_brand_asset, and replace_background tools.\n
  • Sanitization: The instructions do not specify any validation or filtering steps for the ingested image data.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 12:36 AM