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

Create GPT

Train task-specific small GPTs (0.5B-1.7B) that fill the Tier 1.5 gap in the inference cascade. These are specialists, not generalists — they handle JSON validation, taxonomy triage, and "needs_review" gating. Cases with low confidence escalate to Chutes via /scillm.

Prompt Iteration Rule (NON-NEGOTIABLE)

System prompts in SFT training data MUST be validated through /prompt-lab before training. NEVER hand-craft system prompts in Python strings or JSONL files.

• Before training: /prompt-lab eval the system prompt against holdout ground truth
• Comparing prompt variants: /prompt-lab compare across models
• Finding minimum viable model: /prompt-lab find-minimum
• Only after prompt-lab validation → proceed to /create-gpt train

Inference Cascade