sue-run-preparation

What problem does it solve?

This Skill eliminates failed ML/HPC scale-up experiment runs caused by missing runtime environments, misconfigured paths, or unprepared datasets, removing the need for error-prone manual preflight checks that are inconsistent across different HPC and cloud sandbox backends.

Core Features & Use Cases

• Parallel Preflight Dispatch: Spins up separate subagents to handle environment installation and dataset preparation simultaneously, cutting down preflight time for large experiments.
• Strict Contract Enforcement: Validates workspace root paths, output directory rules, and backend-specific constraints (like LUMI's container-only runtime policy) to prevent inode waste, path mismatches, and quota violations.
• Safe Configuration Updates: Returns proposed runtime.yaml patches instead of editing configuration files directly, ensuring all changes are reconciled and applied consistently after preflight tasks complete. Use Case: Before launching a large-scale computer vision training run on the LUMI supercomputer, this skill automatically builds a custom Singularity container with required dependencies, prepares the ImageNet-derived dataset in inode-safe LMDB format, and verifies all environment imports and path contracts are met before the dryrun phase begins.

Quick Start

Use the sue-run-preparation skill to complete all preflight setup for your upcoming scale-up experiment dryrun or fullrun.