The Agent Skills Directory

[COMMAND_EXECUTION]: The unit test script tests/test_validate_dataset.py utilizes subprocess.run() to execute the scripts/validate_dataset.py script. This execution uses a list-based argument format, avoids shell invocation, and is restricted to the local testing environment with controlled parameters, representing standard and benign development practice.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of dataset CSV files provided by the user.
Ingestion points: The scripts/validate_dataset.py script accepts a CSV file path via the --csv argument and reads its contents to validate the dataset structure (e.g., in tests/test_validate_dataset.py).
Boundary markers: The skill documentation does not define explicit boundary markers or delimiters to isolate user-provided CSV data from agent instructions.
Capability inventory: The skill has access to Bash and Read tools as declared in SKILL.md, enabling file system interaction and command execution.
Sanitization: The scripts/validate_dataset.py script implements robust sanitization and validation, checking for required columns (input_path, golden_path, label, object_name), verifying the existence of directories, and enforcing specific string formats for labels and object names. This reduces the risk of the agent misinterpreting data as instructions.

tao-train-optical-inspection