sue-update-lesson

To record ML workflow lessons and prevent repeated mistakes, the skill automates appending validated solutions to persistent memory files. It routes lessons to global, workspace, or sandbox-specific memory based on their scope and relevance.

Recording sandbox quirks for backends like LUMI or Snellius is handled through intelligent lesson routing. The system appends backend-specific fixes, such as LUMI MIOPEN cache requirements, directly to the corresponding sandbox memory file for future automatic application.

Yes, you can save workflow fixes across different cloud sandbox and HPC environments. The system supports backends including LUMI, Snellius, RunPod, Brev, and AutoDL, routing recorded lessons to the correct memory file based on the specific environment scope.

Yes, recording experiment lessons checks existing memory files to prevent duplicate entries. Before appending a new lesson, the system validates against current records and strengthens existing lessons if the new evidence adds further detail rather than creating a redundant entry.

To ensure recorded experiment lessons align with SUE workflow rules, the system enforces workspace root contract validation and runtime.yaml consistency checks. It also updates related skill documentation if a lesson reveals a workflow bug, keeping all processes synchronized.

The best way to manage experiment reproducibility during scale-up ML workflows is to persist durable lessons in structured memory files. This prevents lost knowledge between runs, saving GPU hours by ensuring solved issues and sandbox quirks are automatically applied.