The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes code snippets that reference external libraries for 3D model decoding. Specifically, it points to Google's gstatic CDN for Draco decoders and the jsDelivr CDN for KTX2 basis transcoders. These are well-known and trusted sources within the Three.js ecosystem.
[DATA_EXFILTRATION]: Provides an example of using setURLModifier to prepend a CDN URL to asset paths. This is a standard technique for asset delivery optimization and uses a generic placeholder domain (cdn.example.com).
[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to load and process external 3D asset files like GLTF and textures. This ingestion surface is evaluated for risk:
Ingestion points: loader.load in SKILL.md and references/advanced-loading.md.
Boundary markers: Not present.
Capability inventory: Read, Glob tools allowed in frontmatter.
Sanitization: None specified for the asset data content. Despite the lack of sanitization, binary asset formats parsed by Three.js represent a low risk for instruction injection.

threejs-loaders