mujoco-studio-sidecar
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation instructs the agent to install the "mujoco" package from PyPI. This is a verified library from a trusted organization (Google DeepMind).
- [COMMAND_EXECUTION]: The skill launches a background daemon ("studiod.py") to manage simulation state, hot-reloading, and communication with the MuJoCo Studio Web sidecar. This is the core functional requirement of the skill.
- [DYNAMIC_EXECUTION]: The daemon utilizes the "mujoco" library to load and hot-reload Python-based simulation and viewer plugins from the user's workspace. This mechanism is designed to facilitate live development and iteration on MuJoCo models.
- [INDIRECT_PROMPT_INJECTION]: The agent reads simulation state snapshots from JSON files generated by the daemon. These snapshots contain data (such as joint names and numeric states) originating from workspace files. While this creates a surface for potential indirect injection if workspace files are untrusted, the risk is inherent to the developer-focused task where the agent is the primary creator of the content.
- [SAFE]: The daemon hosts a local HTTP server on the loopback interface (127.0.0.1) and uses "Access-Control-Allow-Origin: '*'" to permit interaction with the MuJoCo Web viewer. This exposure is restricted to the local machine and is a standard requirement for web-based development sidecars.
Audit Metadata