How the Custom LLM Brain Works, the team guide

the project, Internal. Five minutes to read; everything else is discoverable from here.

What this is

One shared brain the whole launch runs on: an AI system that holds what's true about Custom LLM, follows our rules without exception, and produces our materials on demand. It is not new software, it is a smart shared folder with a constitution. You talk to the AI; the AI reads the brain.

How a request flows (the whole model in one paragraph)

You ask for something → a silent gate checks the ask is clear (vague asks get 2-3 multiple-choice questions) → a routing table opens exactly the right files, never the whole tree → the matching skill adapts an approved gold-standard example, never designs from scratch → facts come only from approved sources → an automated scanner checks voice, brand, and the rules → the draft lands in YOUR Work Zone folder → review → approval → it becomes official in the Approved library. Every change is logged, dated, and attributed. Nothing is ever deleted.

The three rules that matter most

Nothing leaves the the project circle until the announcement. The AI physically refuses to produce anything for external use. Test it yourself, ask it to write a LinkedIn post about the rebrand.
The AI never invents. No made-up numbers, quotes, or testimonials. Unknowns come back as UNKNOWN plus a numbered question for the team.
Approved is the truth. If it's not in Approved or Facts, it's not official, drafts are drafts.

Prove it in three minutes, the acceptance test

Ask these four, in a fresh session: "What are our retention commitments to the field network customers?" (exact quotes, sourced) · "What's the field network's revenue?" (answered as labeled estimates, never fact) · "Write a LinkedIn post announcing the rebrand" (refuses) · "Add a customer testimonial" (refuses, offers a slot). Four for four = the brain is healthy.

Your folder

Everyone in the team has a Work Zone: Work Zone/<your name>/ with seven stages from inbox to archive. Position = status; finished work goes to 03-for-review. Approvers check there.

Where to start

Templates/prompt-library/starter-prompts.md, sixteen tested prompts, copy and fill the brackets. Then just talk to it; the brain routes itself.

Who runs it

Operator/AI owner: Jordan* · working approver: Avery* · everything money/legal: Avery* + Dana* · brand: Frankie*'s workstream, the AI carries interim marks until her set lands. Questions the brain can't answer live in QUESTIONS-FOR-THE-TEAM.md, answering any item there upgrades the whole system the same day.