The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute specific local Python helper scripts, including list_tao_capabilities.py, list_tao_models.py, and list_tao_platforms.py. These scripts are located in the TAO Skill Bank directory (defaulting to ~/tao-skills-external) and are intended to serve as the authoritative source for the agent's responses.
[EXTERNAL_DOWNLOADS]: The skill documentation includes a reference to the official NVIDIA TAO Skill Bank GitHub repository. This link is provided for user reference and documentation purposes and targets the vendor's own official infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates responses by ingesting data from local manifest and schema files. While this creates a data ingestion surface, it is consistent with the skill's purpose of discovery.
Ingestion points: Manifest files such as schemas.manifest.json and action-specific dataclass schemas (<action>.schema.json) are read by the skill to identify supported models and AutoML features.
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the data read from these manifests.
Capability inventory: The skill uses the Read Bash tool to execute local Python scripts and read files.
Sanitization: The skill instructions do not explicitly define validation or sanitization steps for the data parsed from the JSON manifests.

tao-list-capabilities