blockbench-gpt-image-textures
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill interacts with fal.ai endpoints (
fal.run) to generate images and downloads the resulting texture files from the service's media storage (fal.media). - [COMMAND_EXECUTION]: The skill requires the execution of local Node.js and Python scripts (
uv_layout_prompt.mjs,fal_gpt_image.mjs,resize_to_atlas.py) to process UV layout data, manage API calls, and perform image resizing operations. - [INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection as it incorporates user-derived model metadata, such as cube and mesh names, directly into the prompts sent to the image generation model.
- Ingestion points: Layout data and identifiers are ingested from a local
inspection.jsonfile. - Boundary markers: Prompts are structured as JSON objects, but model-derived strings are interpolated without explicit safety delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill can execute local scripts, perform file system writes, and make network requests to a specific service provider.
- Sanitization: No explicit sanitization or filtering is performed on model metadata before it is included in the generated prompt.
Audit Metadata