The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill utilizes the jsDelivr CDN to load the Three.js library and its associated addons. While jsDelivr is a standard and reputable CDN for web development, loading remote scripts is categorized as a LOW risk. The use of explicit version pinning (@0.171.0) significantly mitigates the risk of supply chain attacks or unexpected breaking changes.
[PROMPT_INJECTION] (INFO): The skill provides patterns for loading external 3D models and textures (e.g., via GLTFLoader and RGBELoader). This constitutes an ingestion point for untrusted data (Category 8). However, the capability of the skill is strictly limited to visual rendering on the GPU, which does not allow the external data to influence the agent's logic or privileged operations. Evidence: 1. Ingestion via GLTFLoader/TextureLoader in SKILL.md; 2. Boundary markers are absent in the examples; 3. Capability inventory is limited to GPU rendering and console logging; 4. Sanitization is performed internally by the Three.js library loaders.
[SAFE] (SAFE): Technical review of the source code confirms there are no attempts at prompt injection, data exfiltration, obfuscation, or persistence. The 'hit.object.name' string flagged by automated scans is confirmed as standard Three.js property access for object identification and does not contain phishing or malicious URLs.

threejs