build-rigged-game-assets
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external asset data, including 3D model files (FBX/GLB) and JSON manifests that can include URLs or task descriptions. This creates an attack surface where instructions embedded in asset metadata or external source descriptions could potentially influence the agent.
- Ingestion points: The
sourceandprovenancefields in the character and monster manifests (assets/*.template.json). - Capability inventory: Includes local script execution for validation and general repository interaction (build, lint, git).
- Boundary markers: The instructions do not specifically warn the agent to ignore or delimit embedded natural language instructions within external 3D model files.
- Sanitization: While the
validate_asset_manifest.pyscript enforces schema constraints, it does not sanitize the contents of external provenance links or model metadata. - [COMMAND_EXECUTION]: The skill utilizes a local Python script,
scripts/validate_asset_manifest.py, to verify that asset manifests conform to the project's requirements and that referenced files exist. - The script performs legitimate validation logic using
pathlibandjsonand does not contain any patterns associated with malicious activity, such as dynamic code evaluation or remote command execution.
Audit Metadata