mujoco-showcase-shot
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
render_showcase_shot.pyinvokesffmpegviasubprocess.runto encode video frames into an MP4 file. - Evidence: Subprocess calls are used for side-by-side video stitching and as a fallback for the
avlibrary. - Mitigation: The script uses a list for command arguments rather than a raw shell string, preventing shell injection vulnerabilities.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of MuJoCo model files (
.xml) and 3D mesh files (.obj) provided by the user. - Ingestion points: The script loads asset specifications via
mujoco.MjSpec.from_file()and parses Wavefront OBJ files. - Boundary markers: None explicitly used for these file types.
- Capability inventory: The skill can write files to the artifact directory and execute the
ffmpegbinary. - Sanitization: The script performs some validation on OBJ file integrity (checking for normals and UVs) but primarily relies on the MuJoCo engine's own parsers for security.
Audit Metadata