The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes dataset annotation files (e.g., annotations.txt) which represent an external data ingestion surface for the agent.\n
Ingestion points: The skill specifies data file paths in references/skill_info.yaml for training, evaluation, and inference actions.\n
Boundary markers: The instructions do not explicitly provide boundary markers or warnings to ignore embedded instructions within the annotation files.\n
Capability inventory: The skill has the capability to execute training, evaluation, and inference commands via Docker as described in references/setup-and-run.md.\n
Sanitization: There is no evidence of explicit sanitization or validation logic for the content of annotation files within the skill's instructions.\n- [EXTERNAL_DOWNLOADS]: The skill references official NVIDIA TAO toolkit container images from the NVIDIA Container Registry (nvcr.io).\n
Evidence: Image references such as nvcr.io/nvidia/tao/tao-toolkit:7.3.0-pyt and nvcr.io/nvidia/tao/tao-toolkit:7.3.0-deploy are found in references/skill_info.yaml and references/tao-deploy-fast-foundation-stereo.md.\n
Context: These resources originate from a well-known service associated with the skill's vendor and do not escalate the security verdict.\n- [COMMAND_EXECUTION]: The skill documentation includes a command for cleaning up Python bytecode caches during local development.\n
Evidence: A find command with rm -rf is provided as a tip in references/setup-and-run.md to clear __pycache__ directories.\n
Context: This is a standard development maintenance command provided for manual execution by a user and does not represent an automated high-risk execution pattern within the skill's logic.

tao-train-fast-foundation-stereo