sue-scripts-writing

What problem does it solve?

This Skill eliminates the error-prone manual work of adapting student ML experiment code to run on HPC and cloud sandbox backends, enforcing strict path containment, GPU utilization, and inode-saving rules required for production scale-up experiments, so you avoid misplaced output files, idle GPU waste, and failed readiness gates.

Core Features & Use Cases

• Strict Path Rule Enforcement: Automatically enforces SUE path rules to ensure all experiment artifacts (logs, checkpoints, results, launchers) live under the designated experiment directory, eliminating stray files in source trees or scratch space.
• Compliant Launcher Generation: Creates Slurm and sandbox-side launchers that adhere to GPU-vacancy, full-path display, and inode-safe output invariants, with built-in monitoring for GPU utilization and path compliance.
• Existing Launcher Auditing: Reviews pre-existing launchers for alignment with SUE path rules, scheduler requirements (like NM5 TMPDIR policies), and output bucket policies, correcting violations automatically where possible.
• Use Case: If a student's training script writes checkpoints to the repository root instead of the experiment directory, this skill rewrites the launcher to redirect all outputs to the correct SUE path and adds GPU binding and utilization monitoring to meet scale-up readiness requirements.

Quick Start

Use the sue-scripts-writing skill to wrap the provided student ML experiment script into a compliant Slurm launcher for your target sandbox, ensuring all run artifacts are stored under the correct experiment directory and GPU utilization is actively monitored.