Set up the Brad in HR AI Protocol

I want to set up the Brad in HR AI Protocol. You are running the setup.

The protocol has three layers.

MEMORY. A place my context lives that outlasts this conversation, so you know
something about me tomorrow that you learned today.

STANDARDS. How I want work done, written once instead of retyped into every new
chat.

CHECKS. One step I will really perform before I trust what you hand me.

Here is how to run this.

Ask me ONE question at a time and wait for my answer. Never send me a numbered
list of questions. Do not summarize my answers back to me. Do not compliment
them. If an answer is thin, ask one follow up, then move on.

Assume I am not technical. Explain anything unfamiliar in plain language and
keep it short. The whole thing should take about ten minutes.

Start with these five, one at a time.

1. What do you do, and who do you do it for?
2. What are the two or three things you are working on right now?
3. What should somebody know about how your business works that would not be
   obvious from the outside?
4. Are there names, terms, or shorthand you use that an outsider would not
   follow?
5. What should never end up in something you send out? Names, numbers, topics,
   anything.

Then write my context file using my words, not yours. Keep it under a page. Use
these headings: WHAT I DO. WHO I DO IT FOR. WHAT I AM WORKING ON NOW. HOW MY
BUSINESS REALLY WORKS. NAMES, TERMS AND SHORTHAND. NEVER INCLUDE WITHOUT MY
APPROVAL. LAST UPDATED.

Then ask these four, one at a time.

1. When something written for you sounds wrong, what usually makes it sound
   wrong?
2. Is there anything you never want to see? A word, a phrase, a formatting habit.
3. Do you want things short and direct, or fuller and more explained?
4. When you ask for something, do you want options to choose from, or one
   recommendation?

Write my standards from those answers. Rules only, never procedures. "Never use
industry jargon" is a rule. "First open the tracker, then update the status" is a
procedure. Short lines. Keep it under 3000 characters so it fits a custom
instructions field.

Then ask me one more question, and do not rush it.

What would be expensive if it came back wrong and you did not catch it?

Help me pick exactly one check. Not a policy. One step, small enough that I will
do it on a bad day. If I try to pick three, talk me down to one. Add it as a
single line at the bottom of my standards, then add these two lines underneath,
unchanged.

Check the thing, not the report about the thing.
Check against the source, not against a summary of the source.

When you are done, tell me three things and stop. Where my two files are and what
to do with them. That I should try the daily close today. And that this only
works if the context file grows, so whenever I catch myself re-explaining
something, that belongs in it.

Do not lecture. Do not list next steps beyond those three.

Ask me the first question now.

Close my day.

You were here. Do not ask me to summarize my own day.

Capture four things and only four.

1. What got done, including the actual output and where it went.
2. What got decided, and why. A decision without its reason gets re-litigated in
   three weeks by somebody who does not remember.
3. What got found that nobody was looking for. Errors, surprises, things that
   turned out not to be true. These are worth more than the wins.
4. What is still open, and who owns it. If it is waiting on me, say so plainly.

Add it to the top of my running log, newest first. Never delete an older entry.
If today overrides an earlier decision, say outright that it supersedes it and
why, rather than quietly replacing it.

Record what went wrong too, including your own mistakes. A log that only contains
wins is a log nobody trusts. When something was produced wrong and caught, write
down what was wrong, how it was caught, and what changed.

Write it so somebody reading it in three weeks could act without asking a
question.

Then tell me where it landed in a few lines. Not a recap of the whole day.