The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill executes docker exec --user root to modify file permissions inside local service containers (e.g., content-physics-agent-service) using chmod -R a+rX. This is implemented as a specific workaround for permission issues with the Scene Optimizer component in references/content-agents/scripts/content_agent_client.py.
[COMMAND_EXECUTION]: Widespread use of subprocess.run() across multiple orchestrator and reference scripts to drive the conversion, validation, and container management pipeline.
[EXTERNAL_DOWNLOADS]: Provides an installation hint to download and execute the uv package manager installer from astral.sh. Astral is a recognized and well-known service provider in the developer tools space.
[CREDENTIALS_UNSAFE]: Sensitive model provider API keys are read from the environment and mirrored into local configuration files for Docker services. The skill implements redaction logic in shared/script_utils.py and references/preflight/scripts/preflight.py to prevent credential leakage into workflow logs and reports.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through the processing of untrusted asset files (USD, URDF, MJCF, etc.) which are used to generate inputs for subsequent AI agent sessions.
Ingestion points: Source assets are inspected in references/identify-asset-context/scripts/run.py and processed by various converter references.
Boundary markers: Explicit delimiters for embedded data instructions are generally not used in the script-level prompt generation.
Capability inventory: Significant capabilities including shell execution, file system modification, and network requests are present across the reference scripts.
Sanitization: Log redaction is present, but the skill primarily relies on the structural parsing of complex 3D data files.

omniverse-cad-to-simready