The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches model checkpoints, encoders, and datasets from Hugging Face, NGC, and GitHub. These sources include resources from established organizations like NVIDIA, Google, and Meta, which are handled according to standard developer workflows for machine learning model retrieval.
[COMMAND_EXECUTION]: The skill executes shell commands for environment setup, Docker container management, and launching machine learning training and inference tasks via torchrun. These operations are consistent with the primary purpose of the TAO AnomalyGen pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including defect specifications (JSONL) and image datasets.
Ingestion points: dataset_dir (images/masks) and defect_spec (JSONL).
Boundary markers: The skill utilizes standard YAML and JSONL parsing for configuration files.
Capability inventory: The skill has the ability to execute shell commands, write to the file system (results and configuration directories), and launch subprocesses via Docker.
Sanitization: Data is processed through standard structured parsing libraries (YAML/JSONL).
[SAFE]: The skill follows secure practices for secret management, instructing users to provide the Hugging Face access token (HF_TOKEN) via environment variables or .env files rather than hardcoding credentials. Troubleshooting steps for Docker volume permissions involve standard chmod and chown commands.

tao-generate-anomalies