OpenClaw deployment guide, OpenClaw curriculum for starters when installed

The only thing I do is talk to them。

not propt, not remodel, not restructure. just talk, give feedback, watch them write down。

• Proclaim: This is from Shubham Saboo, an overseas god who can focus on a wave

forty days ago, my content-intellectual tweets piled with emoticons and hashtag to study how intelligents drown valuable information in noise. i spend more time on error than i do directly。

Today, Kelly drafts with my tone, and Dwight sends seven stories every morning, each worth reading. Eight smarts running on a 24-hour basis. I opened Telegram, looked at the draft, had a cup of coffee。

Day 1 and day 40 are the same model. The difference is a bunch of Marktown documents that get rich every week。

This is the file system。

Let's get one thing straight

• Smart bodies don't get smarter because you've spent more time. But the documents around it will be richer, more precise and more relevant to your needs. These cumulative contexts are the moat。

many people spend a lot of time changing prompts, changing models, and studying various programming frameworks. but the real difference is not the model, it's the document system。

There is no news queue, no database and no complex organization framework. The whole system is the Markdown file on disk. The document system itself is an integrated layer。

That sound shabby? You'll see why it works better than any frame。

Three layers of architecture

The entire operating system consists of three layers:

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
First level: identity level
Who's the smart body? Who's it for? Zenium
SOUL.md
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Zenium
Zenium
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Second level: operating layer
How do intelligence work? How do you heal yourself? Zenium
AGENTS.md |HEARTBEAT.md | Role Guide
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Zenium
Zenium
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Third level: knowledge level
What did the smarts learn? Zenium
MEMORY.md Daily log
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Figure 1: Third-level document structure

Each level addresses one core issue:

Who is this file identification level at the top of the hierarchy? Who is it for? Soul.md, DENTITY.md, USER.md, how do you work? How? AGENTS.md, HEARTBEAT.md MAMORY.md, Daily Log, shared context

It's detaching down there。

First level: Identity level

Soul.md - Who's the smart body?

This is the personality file of the intelligent. Definition of identity, responsibility, behaviour。

An example of a study of smart Dwight:

# SOUL.md (Dwight)

## Core identity
Dwight -- study the brain. Named after Dwight Schrute because you have his power:
It is extremely rigorous, knows everything in its field and takes it very seriously.
No bullshit, no speculation, only facts and sources.

1 TP5T1 TP5T your role
You're the team's intelligence cadre. Responsible for researching, verifying, collating and exporting information,
For other smarts to create content.

## your principles
No fabrication - source links to each assertion
Signals are better than noise - not all popular content has value
3. In case of uncertainty, mark [UNVERIFIED]

IDENTITY.md - Quick Reference Card

Soul.md is a complete personality and IDENTITY.md is a business card。

# IDENTITY.md

- ** Name:** Dwight
** Role:** Research AI — Intelligence backbone
-** Gas:** Strong, strict, zero tolerance for inaccuracies
**Emoji:**
- ** Source of inspiration:** Dwight Schrute (Office)

The file is very small, but when you run eight smarts at the same time, the design will greatly enhance the experience. This is also what smart people show when they send you messages from Telegram。

USER.md - Persons served by smart bodies

Every intelligence needs to know who it's helping。

# USER.md

- ** Name:** Shubham
- ** Time Zone:** PST (USA/LA)
- ** Diet:** vegetarian

## Background
- Google Cloud, senior AI product manager.
- Founder of the Awesome LLM Apps Open Source Project (91k+stars)

## preferences
- Short paragraphs, powerful sentences.
- Don't use a dash, ever.
- Practice first. Never talk about theory

Personal details are more important than you think. The time zone means the smarts don't set you up at 3:00 in the morning. Dietal preferences mean that when Pam drafts a newsletter for team dinner, the steakhouse is not recommended. These details can have a compounding effect。

Write once, all smart bodies read。

Second level: operating layer

AGENTS.md - Rules of conduct

SOUL.md defines who the smart body is and AGENTS.md defines how it works: session start process, file reading order, memory management, security rules。

AGENTS.md:

# AGENTS.md

# AGENTS.md

## per session

Before doing anything:
1. Read SOUL.md - This is your identity.
Read USER.md — this is what you serve.
3. Read memory/YYYYY-MM-DD.md (Today + Yesterday) for recent context
4. If in main session: read Memory.md

## Memory

- Something in your head disappears after the session is restarted, not the file.
- When someone says, "Remember this,"
- Text > Brain

## Security

- Never leak private data.
- Use the wastebin instead of just deleting it.
- In case of doubt, ask first

Smarts don't remember sessions, they start from scratch every time. If a correction did not fall into the document, it would not exist at the next meeting. AGENTS.md makes this clear, making sure the smarts write everything down。

Every smart body can build on this to expand its own rules. Kelly's AGENTS.md added six additional documents: writing style guide, post-format references, real cases, daily assignments..

HEARTBEAT.md - Self-healing mechanism

The intelligent team is the infrastructure, which can fail。

Heartbeat.md of Monica monitors two things:

1. Whether or not the browser survives
2. Whether or not a time-bound mission is carried out -- if it goes, Kelly and Rachel will work on the basis of outdated intelligence

## health check (run every heartbeat)

** Browser: ** Check that OpenClaw hosting browser is running.
If running: false, start it.

** Scheduling mission:** Checks for lastrunAtMs timeout (>26 hours).
If overtime, force trigger through CLI.

Tasks to monitor:
- Dawn morning (8:01 AM)
- Kelly X Draft (5:01 PM)
- Rachel LinkedIn (5:01 PM)

i've been fucked in the third week. the dispatcher had a bug, and the mission was advancing in line, but never actually carried out. i haven't seen it in hours. and then i built up a heartbeat to incorporate the failure pattern。

You don't need this on the first day. You'll know exactly what to monitor, because you've already felt what's gonna crash。

Level III: Knowledge

This is a truly effective memory system — a three-tier document-based system。

Level 1: MEMORY.md

Not the original log, not everything that happened, but what really matters。

# Memory.md

## Shubham
- It is prohibited to break the code, to use a colon, stop or reorganize the sentence.

##
- Never delete a project folder without Shubham ' s confirmation.
On 26 February, during the clean-up, the Gemini-council React application of Ross was deleted.
React version permanently lost.

## X posting rules
- We'll start with a strong initial.
- The entire tweet is extremely short (within 180 characters)
- No Hashtag, no emoji.
- Three drafts for each topic.

###
[Listing out each pattern rejected: bullet, arrow, LinkedIn tone]

Take note of the sections "Breath and Tear Lessons" and "Application Demonstration". Monica deleted a project folder, an error that has since been permanently written into her long-term memory. She'll never do that again。

one correction, one storage, to prevent the same error from being repeated in each future session. this section alone is more valuable than any prompt project guide。

Level II: Daily log (original records)

# Kelly Daily Log - February 5, 2026

## 5 p.m. - Daily X Draft

#1T1T# Hotspot Today
- Opus 4.6 vs GPT-5.3-Codex 27 minutes apart.
- Anthropic C compiler (16 smarts, $20,000)

### has submitted draft
C Compiler - Mono, Format found
Mitchell Hashimoto's 6 Steps - Thread Format
3. Opus 4.6 vs GPT-5.3-Codex - Thermal evaluation format

#1T#
- Shubham feedback on the draft

The daily log is a raw material and the Memory.md is a refined product。

Maintenance rules: Daily logs accumulate quickly, and the context of a smart body expands without cutting. Kelly's log once reached 161,000 tokens, and the quality of the output dropped sharply and had to be compressed to 40,000. Only today and yesterday's logs are loaded every time。

level 3: shared-content/ (cross-intelligence)

This is the newest part of the membership and the part that changes everything。

i'm sorry.
THESIS.md - My current view of the world
Ideas - FEEDBACK-LOG.md - Corrects for all smarts
SIGNALS.md, I'm tracking articles and trends

THESIS.md is my current frame of thinking: what I'm concerned about, what I've written, and what gaps remain. Dwight reads it to determine research priorities, and Kelly reads it to match my thinking. Every intelligence is aligned to the same source of truth。

FEEDBACK-LOG.md is an inter-intelligent correction layer. When I told Kelly, "Don't use a dash," the same feedback applied to Rachel, Ryan and Pam. Instead of correcting each of the four intelligent bodies, I only write once and everyone reads it。

• this single change saves more time than any prompt i've ever done。

How intelligence works together

NO API CALLS, NO MESSAGE QUEUES, ONLY FILES。

Dwight wrote the study intel/DAILY-INTEL.md, Kelly, Rachel, Pam. Collaboration is the document system。

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Dwayne-INTEL.md
(Research)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Zenium
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Read, read, read, read.
I'm sorry.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Kelly, Rachel, Pam.
│ (twitter)L (linkedIn)│ (communication)│
_ -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --

Figure 2: Document-based collaborative processes

Monographer principle: Never let two intelligent bodies write the same document at the same time. Each shared file is designed as a writer, multiple readers. This will prevent all coordination conflicts that you would have had to debug。

This was made possible by the movement: Dwight runs at 8 a.m. and 4 p.m. and Kelly and Rachel at 5 p.m. Dwight first, because everyone depends on his output. There's been a mistake in the order, and what you read in the downstream intelligence is an outdated or blank document。

Full Directory Structure

workspace/
Ideas - SOUL.md # Monica
A quick reference for IDENTITY.md # Monica
Ideas - AGENTS.md # Root rules of conduct (inheritance of all intelligent bodies)
Idea-USER.md # about me.
The long-term memory of Memory.md # Monica
Ideas - HEARTBEAT.md # self-healing
Ideas -Shared-context/
THESIS.md # My current view of the world
│ - FEEDBACK-LOG.md # Inter-Intelligence Correction
I'm tracking the trend.
Intel/
│--DAILY-INTEL.md # Dwight output
Agents/
│-dwight/ # research intelligence
SOUL.md
I'm sorry.
– Memoory/
Think - Kelly/# Twitter smart body
SOUL.md
I'm sorry.
X-CONTENT-GUIDE.md
– Memoory/
│ - Rachel/# LinkedIn Smart Body
│ - pam/ # communication intelligence
I'm sorry.
–Momory/
- Shubham/ # personal notes
_Shared/ # share context
└ - - 2026-02-27.md # Daily Operating Log

Why is this working

The documents are not static. They are evolving。

Kelly's SOUL.md first day was just a rough draft. By the 40th day, it had specific examples of tone, a list of rejected models she herself had written, and an area “never recommend again”。

The first day of Dwight's principles is called “Finding hot trends”. Day 10 became “If Alex cannot act on it today, skip”. On day 20, he joined the verification steps。

Share context layers until the 20th day. I was doing the same thing over and over again over and over again. Then I built THESIS.md and FEEDBACK-LOG.md, and suddenly, one correction spreads all over the place。

The models for Day 1 and Day 40 are the same. It won't get smarter because you've spent longer。

But the documents surrounding it have become richer, more precise and more relevant to your specific needs。

These cumulative contexts are the moat. No one can copy it by using the same model。

You have to win it by coming out every day and talking to smart people。

How do you start

Time action to install todayOpenClawWrite a SOUL.md, IDENTITY.md, USER.md. Picks out the most repetitive routine tasks, sets a time-bound task and starts giving specific feedback after three days to ensure that the feedback falls into the memory file, instead of just one week in the chat log to create AGENTS.md, defines the session start process, adds the memory management rule to two weeks to start writing emory.md, reviews the daily log and distills the recurring correction into a permanent entry。

At this point, you will feel that after three weeks of compound interest, you will add a second intelligence body to create document-based collaboration. As the pattern emerges, the addition of role-specific guidance is about to create a shared context layer at the same time. THESIS.md to record current thinking, and four weeks after your first troubles with the management of cross-intelligence by FEEDBACK-LOG.md to add HEARTBEAT.md

Final Thoughts

The only thing you need to do is talk to your smart body. The document will do everything else。

not propt, not remodel, not restructure。

Just talk. Give feedback. Watch them write down。

Then one day you open Telegram, read the draft, have a cup of coffee。

Your intelligence has learned how to work for you。