sue-goal-pipeline

What problem does it solve?

This Skill eliminates the error-prone manual work of coordinating the complex, ordered sequence of SUE scale-up experiment stages across multiple HPC and cloud sandbox backends, preventing out-of-order execution, missed prerequisite checks, and lost pipeline state when runs are interrupted.

Core Features & Use Cases

• Canonical Stage Order Enforcement: Automatically normalizes any user-requested or partially completed stage sequence back to the hard canonical SUE pipeline order, preventing invalid stage jumps or skips.
• Interrupted Pipeline Resumption: Reads existing pipeline state to resume from the earliest missing or blocked stage instead of rerunning completed work, saving time and compute resources.
• Backend-Aware Orchestration: Automatically selects the correct environment installer for the target sandbox (LUMI, Snellius, NM5, RunPod, etc.) and enforces all workspace root and sandbox repo root contracts before any compute-intensive steps.
• Use Case: An ML researcher running a scale-up training experiment on LUMI can invoke this skill once to handle all prerequisite checks, environment setup, script writing, dryrun, fullrun, and post-run summary without manually tracking stage order or resuming from failures.

Quick Start

Use the sue-goal-pipeline skill to run the complete end-to-end SUE scale-up experiment workflow for your active workspace, covering all prerequisite checks, launch stages, and post-run reporting.