The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's documentation and example conversations (found in references/examples.md and references/mandatory-rules.md) explicitly instruct the agent to prompt users for sensitive API keys like llm_api_key and WANDB_API_KEY in plaintext. The provided example dialogue shows a user providing a key (sk-abc123) directly in the chat, which is then passed into generated scripts.
[DATA_EXFILTRATION]: The skill transmits experiment metadata, hyperparameter configurations, and training results to external services such as Weights & Biases and LLM providers (NVIDIA NIM, OpenAI). This behavior is documented as a core feature for experiment tracking and optimization.
[COMMAND_EXECUTION]: The skill dynamically generates Python "runner" scripts and executes them via the command line. It also uses the Bash tool to install dependencies and manage the execution environment.
[EXTERNAL_DOWNLOADS]: The skill initiates the installation of multiple Python packages from the public PyPI registry, including nvidia-tao-automl, openai, and wandb.

tao-run-automl