CheMLFlow

Create or review a single CheMLFlow runtime config YAML. Use when Codex is asked to build one config for a dataset, choose regression vs classification settings, handle SMILES or non-SMILES CSV inputs, choose curation/drop-row rules, set split/CV/random seed/scaler/model options, or explain when a single config should become DOE fanout.

Design and review CheMLFlow DOE YAMLs and generated DOE artifacts. Use when Codex is asked to create, modify, or audit CheMLFlow DOE specs, search spaces, model/feature/scaler/split compatibility, manifest skip reasons, parent/child CV shape, or expected valid/skipped case counts.

Audit and curate CheMLFlow analysis outputs. Use when Codex is asked to validate `analysis.py` results, inspect `report.json`, compare `all_runs_metrics.csv` with `all_runs_metrics_by_execution.csv`, verify scaler/feature/model/split balance, find failed or incomplete folds, or decide whether a CheMLFlow result bundle is complete and trustworthy.

Coordinate end-to-end CheMLFlow studies across dataset profiling, runtime config design, DOE generation, local or Slurm execution, analysis, and audit. Use when a user asks an agent to run or improve a CheMLFlow experiment workflow rather than only build one config, review one DOE, or audit one analysis bundle.