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

assistant

Shared GPT + classifier + regressor inference gateway for persona monitor tasks.

The "warm pond" — /assistant can autonomously evolve its inference stack via /assistant-lab (self-improvement workbench: create-*, *-lab, model factory) while /monitor-skills provides observability across the skill ecosystem.

Prompt Iteration Rule (NON-NEGOTIABLE)

All system prompts for /assistant models MUST be iterated through /prompt-lab before being baked into training data via /create-gpt. NEVER hand-craft system prompts in Python strings.

• New model prompt → /prompt-lab eval against ground truth first
• Prompt plateau (80-90% shadow agreement) → /prompt-lab compare across variants
• Model retraining → validate prompt with /prompt-lab find-minimum before /create-gpt

Tier Cascade