build-crucible-scene

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation guides the agent to start a server using the @codepawlhq/crucible npm package via npx. This is a vendor-owned package and is central to the skill's intended use.
  • [DYNAMIC_EXECUTION]: The skill supports the use of GLSL "shader hooks" within material definitions. This allows for custom fragment rendering logic, which is a standard feature for 3D modeling and rendering tools.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data sources, creating a potential surface for indirect prompt injection.
  • Ingestion points: Ingests 3D assets (GLB files) from a catalog and receives human feedback (notes/pins) via the scene_ask_user tool.
  • Boundary markers: The skill relies on the agent to interpret the 3D scene visually and follow specific modeling recipes, which inherently limits the impact of embedded instructions in asset metadata.
  • Capability inventory: The agent has capabilities to write files (exporting recipes, rendering jobs) and perform complex 3D geometry transformations.
  • Sanitization: The skill focuses on visual output and structured JSON recipes, reducing the risk of accidental instruction execution from processed data.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 10:49 AM