Scope, design, and deliver ML/AI work in a consulting engagement. Trigger on:
"scope an AI engagement", "should we use RAG or an agent", "AI PoC", "client AI
project", "AI feasibility study", "/ai-engagements". Begin at GATE E-1 of THE
CONTRACT: no architecture talk, no demo, and no estimate before its gate passes.
Model choice and pricing come from the claude-api skill, never memory. All
client-facing prose passes writing-instructions; the smell
checker scripts/smell_check.py is mandatory on anything priced or promised.
