STOP. READ THIS ENTIRE SKILL.MD BEFORE CALLING ANY ENDPOINT.

Classifier Lab

A multi-modality skill for training classifiers on custom datasets. Supports vision (image), text (transformer), and tabular (sklearn) classification.

Minimum Training Data (NON-NEGOTIABLE)

Do NOT benchmark a classifier trained on fewer than 200 samples per class.

The rule is n_samples / n_classes >= 200. Models below this threshold will not cross the 90% promotion gate. You're benchmarking noise, not a trained model.

If you have fewer than 200 samples per class, collect more labels via shadow mode before training. See /create-classifier SKILL.md for evidence and thresholds.

Quick Start