sue-fullrun-summarize-result

What problem does it solve?

Manually aggregating results from completed ML/HPC scale-up experiments is time-consuming and error-prone, as metrics, logs, and resource data are scattered across W&B, Slurm, sandbox filesystems, and CSV ledgers. This Skill automates the full post-run summarization workflow, ensuring all critical evidence is compiled into a consistent, student-accessible bundle.

Core Features & Use Cases

• Mandatory inode cleanup: Automatically runs tar+remove for high-file-count bulk output trees to conserve sandbox storage before fetching results, no user intervention required.
• End-to-end result aggregation: Pulls metrics, W&B cohort data, scheduler accounting, variant results, failure diagnostics, and resource usage into a standardized summary bundle.
• Student-safe reporting: Generates a self-contained HTML dashboard, detailed summary report, and step-by-step student verification guide so users can inspect, validate, and reproduce results without accessing private operator notes or remote sandbox paths.
• Use Case: After launching a 64-variant GPU experiment grid on RunPod, use this Skill to automatically generate a full result summary, resource cost breakdown, and student check guide in minutes instead of hours of manual log and metric collection.

Quick Start

Use the sue-fullrun-summarize-result skill to generate a complete summary and student verification guide for the completed scale-up experiment run in your current DeepResearch workspace.