sue-reset

What problem does it solve?

This skill solves the pain point of manually cleaning up completed DeepResearch scale-up experiment workspaces, which risks accidental deletion of critical source code, configuration, or historical experiment data, and often leads to inode bloat from scattered loose output files.

Core Features & Use Cases

• Versioned Experiment Archiving: Automatically moves all epoch artifacts (runtime configs, ledgers, logs, run bundles, local W&B cache) to a structured versioned history directory under sue-exp-history/ for full reproducibility.
• Safe Workspace Reset: Preserves all source code, Hydra configs, orchestration state, and Git metadata while clearing only experiment-specific outputs, then bootstraps a fresh runtime config for the next experiment epoch.
• Optional Extended Cleanup: Supports optional deletion of W&B runs and groups for the archived epoch and remote sandbox output roots, with explicit user confirmations for all destructive actions.
• Use Case: After completing a 3-month LUMI training run for a large language model scale-up experiment, use this skill to archive all run artifacts and start a new experiment with updated hyperparameters without manually sorting through hundreds of output files.

Quick Start

Use the sue-reset skill to archive your current scale-up experiment outputs and initialize a clean workspace for a new experiment epoch.