The problem
Every AI chat starts from zero. You explain the product, the customer, the
quarter’s goal and what “activation” means on your team. Tomorrow you
explain it again. The assistant gives generic advice because it knows
nothing specific.
The same gap hurts people. A decision made in March lives in one person’s
memory and a lost Slack thread.
Jyothi Nookula showed her own version of the fix on
Aakash Gupta’s podcast
in July: put the context in files, point the assistant at them, and
schedule a morning brief. This entry is that setup, kept as small as it
can be.
What you’ll build
Two things.
- A context folder. Plain text files, stored in git so every change
is saved with a date. A short
CLAUDE.md file at the top tells the
assistant what is in the folder and how to use it.
- A morning brief. A scheduled task in Claude Cowork that runs each
weekday. It reads the folder, your calendar and your assigned tickets,
and writes one page.
Example output
Brief for Tuesday 11 August
Meetings (3)
10:00 Pricing review with finance
Context: decisions/2026-07-22-credits-vs-seats.md says we
chose credits and left overage caps open. That is today's
open question.
13:30 Customer call, Northwind (design partner)
Last call 28 July: they asked for SSO on the mid tier.
No ticket exists for it.
16:00 Sprint review
Needs you today
PROJ-412 Blocked 4 days, waiting on your answer about empty states
PROJ-398 Spec review requested Friday, not started
Not in the folder
Nothing on file about Northwind's renewal date. Ask in the call.
Prerequisites
- A paid Claude plan with Cowork, which is in the Claude desktop app.
- Node.js and Claude Code. Setup 101 walks through both.
- Permission from your company to connect your calendar and ticket tool.
Build it
-
Create the folder. In the terminal:
npx mogkit init product-context
cd product-context
mkdir context decisions customers
mogkit init creates the folder, turns on git, and adds a starter
CLAUDE.md.
-
Write four short files. Keep each under a page. You are writing for
a smart new hire.
context/strategy.md what we are building this year and why
context/glossary.md what our terms mean ("activation" = ...)
context/team.md who owns what
context/metrics.md the numbers we watch and their definitions
-
Add this to the bottom of CLAUDE.md.
## Product context
Before you answer a product question, read the files in context/.
Decisions live in decisions/ as YYYY-MM-DD-short-name.md, each
with the decision, the reason, and who made it.
Customer notes live in customers/, one file per customer.
If the folder does not cover something, say "not in the folder".
Do not fill the gap with a guess.
-
Log your first decision. Create
decisions/2026-08-11-example.md with three lines: what you decided,
why, and who agreed. Do this after every real decision from now on.
-
Save your work.
git add . && git commit -m "Start product context folder"
-
Open the Claude desktop app, go to Cowork, and choose the
product-context folder as the working folder.
-
Connect your tools. In Cowork’s connector settings, connect Google
Calendar and Jira or Linear. Sign in to each when asked.
-
Create the scheduled task. Ask Cowork to schedule this prompt for
every weekday at 8:00:
Write my brief for today and save it to briefs/YYYY-MM-DD.md.
1. List today's meetings from my calendar. For each, find related
files in decisions/ and customers/ and say what is still open.
2. List tickets assigned to me that are blocked or waiting on me.
3. End with "Not in the folder": things today's meetings need
that the folder has no file for.
Keep it to one page. Facts only. Quote the file name for anything
you take from the folder.
-
Read tomorrow’s brief and correct it. When it gets something
wrong, the fix is almost always a missing or outdated file. Update the
file.
How it works
An AI model knows nothing about your company until something puts that
information in front of it. A context folder is the simplest way to do
that. The assistant reads the files at the start of the task, the same way
a new hire reads the wiki.
CLAUDE.md is the index. Claude reads it first in every session. It
stays short and points to the other files. A long CLAUDE.md gets
skimmed, by people and by the model.
- Files beat memory. The model does not remember yesterday. The
folder does. Each decision file you add makes every later answer
better.
- Git gives you history. Every change is saved with a date and an
author, so “when did we decide that?” has an answer.
- The schedule runs the same prompt daily. Connectors fetch today’s
calendar and tickets. The prompt joins them to the folder. The “not in
the folder” section tells you which file to write next.
Variations & next
- Share it. Push the folder to a private GitHub repository and invite
the product team. Changes go through review like code does.
- Add your skills. The workspace already has the mogkit skills in
.claude/skills/. Anyone who opens the folder can run
spec-stress-test or tradeoff-frame with the team’s context loaded.
- Add a Friday review. Schedule a second task that lists the week’s
new decision files and asks which decisions never got written down.
Limits & honesty
The brief is as current as the folder. If you stop logging decisions, it
will quote old ones with full confidence.
A scheduled task that works in a folder on your computer runs only while
the computer is awake and the Claude desktop app is open. If your laptop
is closed at 8:00, the brief does not run until you open it.
The folder will hold sensitive material. Keep the repository private,
leave out anything about individual performance or pay, and follow your
company’s rules for customer data.
The brief saves prep time. It does not attend the meeting for you or tell
you what to decide.