threejs-world-generation

Pass

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text briefs to generate complex world specifications. While the instructions emphasize separating user constraints from inferred construction details, the processing of untrusted natural language input to generate executable scene configurations presents a standard injection surface.\n
  • Ingestion points: Ingestion of user-provided world briefs in SKILL.md and agents/openai.yaml.\n
  • Boundary markers: Present; the skill instructs the agent to maintain separate explicit_constraints and inferred_details lists.\n
  • Capability inventory: The threejs_world tool generates JavaScript files (world-runtime.js) and writes them to the local project workspace.\n
  • Sanitization: Absent; the skill relies on structured JSON serialization rather than explicit content sanitization.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates JavaScript code (world-runtime.js) to facilitate Three.js scene construction in the browser. This generation is based on parameters derived from user input.\n
  • Evidence: Documentation in SKILL.md and references/world-spec.md confirms the tool writes deterministic scene construction scripts as part of the build operation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 14, 2026, 08:43 AM