ML

This page lists the safe and destructive patterns in the ML Experiment Tracking shell guard pack. See Shell Tool for how these patterns are evaluated.

ML Experiment Tracking

Pack ID: ml.experiment_tracking

Guards against destructive ML experiment and model tracking operations

Safe patterns

Rule ID	Pattern
`ml.experiment_tracking.nvidia_smi`	`^nvidia-smi\b[^\|><]*$`
`ml.experiment_tracking.gpustat`	`^gpustat\b[^\|><]*$`
`ml.experiment_tracking.wandb_status`	`^wandb\s+status\b[^\|><]*$`
`ml.experiment_tracking.mlflow_models_list`	`^mlflow\s+models\s+list\b[^\|><]*$`

Destructive patterns

Rule ID	Reason	Suggestion	Confidence
`ml.experiment_tracking.mlflow_gc`	Permanently deletes expired MLflow runs and artifacts	Use `mlflow experiments search` to review before garbage collecting	High
`ml.experiment_tracking.mlflow_delete_experiment`	Deletes an MLflow experiment and all its runs	Use `mlflow experiments search` to review the experiment first	High
`ml.experiment_tracking.wandb_artifact_delete`	Permanently deletes a Weights & Biases artifact version	Use `wandb artifact get` to inspect the artifact first	High
`ml.experiment_tracking.wandb_sweep_cancel`	Cancels a running hyperparameter sweep and all its agents	Review sweep progress in the W&B dashboard first	Medium
`ml.experiment_tracking.clearml_task_delete`	Deletes ClearML tasks and associated artifacts	Review tasks with `clearml-task list` before deleting	High

This documentation was generated from rust/agents/src/tool_guard/shell/packs/ml.rs.