framefields-compose
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references the
framefieldspackage available on npmjs.com for installation and setup. It also mentions deterministic physics dependencies from the@dimforgeorganization. - [DYNAMIC_EXECUTION]: The SDK supports custom procedural logic through signal builders and physics force fields, allowing users to define per-frame behavior via function callbacks.
- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing external data such as text, captions, and media files.
- Ingestion points: Text content in
Layer.text, remote assets inLayer.videoorLayer.image, and subtitle data inLayer.caption. - Boundary markers: The documentation does not specify explicit delimiters or warnings for preventing the agent from following instructions embedded in these asset sources.
- Capability inventory: The skill utilizes Node.js for rendering, involving local file reads for assets and potential network operations for remote resources.
- Sanitization: No specific sanitization or filtering logic is described for the ingestion of external text or metadata into the composition nodes.
Audit Metadata