threejs-world-generation
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided 'world briefs' to define 3D scenes. 1. Ingestion points: 'text or structured briefs' mentioned in SKILL.md. 2. Boundary markers: The skill instructs the agent to maintain separate lists for 'explicit_constraints' (user facts) and 'inferred_details' (execution parameters). 3. Capability inventory: The ThreeJSWorld tool called in SKILL.md generates and writes executable files including world-runtime.js and index.html. 4. Sanitization: The workflow uses a structured world_spec intermediate (defined in references/world-spec.md) which provides a layer of schema validation between user input and generated code.
- [DYNAMIC_EXECUTION]: To create interactive environments, the skill generates and writes Javascript (world-runtime.js) and HTML (index.html) files at runtime. These files are assembled from the world_spec and internal templates.
- [EXTERNAL_DOWNLOADS]: The skill references an external research paper on arXiv and manages the installation of 3D asset catalogs (e.g., kenney-nature-kit). It implements best practices by requiring archive hash verification and explicitly forbidding remote asset fetches during the rendering phase.
Audit Metadata