The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python and Blender headless scripts (scripts/build.py, scripts/inspect_rig.py, scripts/to_glb.py) to process 3D assets and assemble the final HTML output. These commands are triggered by the agent to perform legitimate asset transformation and build tasks.- [EXTERNAL_DOWNLOADS]: The generated HTML template references the Three.js library from jsDelivr (a well-known and trusted CDN) via an import map. This is a standard method for including open-source libraries in web projects without local installation.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted 3D models and images. While the agent is instructed to analyze these inputs to generate configuration code, the risk is minimal as the resulting output consists of static JavaScript and CSS within a local HTML file. Mandatory evidence: (1) Ingestion points: User-provided .glb, .blend, .fbx files and robot sketches; (2) Boundary markers: The skill uses defined config objects to separate logic from asset data; (3) Capability inventory: File read/write via build.py, Blender-based model processing; (4) Sanitization: The build process utilizes base64 encoding for data embedding.

playable