Welcome to your Business Brain OS, You.

Skills are pre-built instructions that tell Claude how to handle a specific type of task — in your voice, for your audience, using your context files.

This module covers how the skill system works, how to call a skill, what's in your library, and how skills get better over time.

• The .claude Folder — skills, commands, and hooks explained
• How to Call a Skill — the three ways to run one
• What's in the Library — an overview of what's available
• Skills Cheatsheet — the quick reference for every skill
• How Skills Improve — the feedback loop that makes them better

There's one habit that makes the difference between a system that compounds and one that stays flat: running the wrap-up at the end of every session.

This module is short on purpose. One lesson, one habit.

• End Every Session — what the wrap-up does and why it matters

There are three ways to call a skill.

The slash commands and run syntax are shortcuts for when you already know what you want.

The Business Brain OS comes with 52 skills. Some you'll use every day. Others you'll reach for occasionally when you need something specific. The full list is in your skills cheatsheet (see the next lesson), but here's a sense of the range:

• Writing copy in your voice: social posts, emails, sales pages, blog posts, Threads, Instagram captions
• SEO: keyword research, blog writing, page optimisation, rank tracking
• Research: trends, competitor content, topic and hook ideas
• Strategy: content routing, weekly planning, end-of-day review
• Website & page building: service pages, sales pages, HTML page design built to your brand
• Pipelines: one command that produces multiple outputs (a transcript into a full week of content, a YouTube video into a blog post plus social posts plus email)
• Utility: fact checking, formatting, PDF generation, publishing

The cheatsheet lists all 52 of your skills with a one-line description of what each one does. Keep it open in a separate window when you're starting out, as it's the fastest way to find the right skill for any task.

After every skill run, the system asks if you have any feedback. If you do ("make the emails shorter", "skip the bullet lists", "don't open with a question"), that gets saved to a file called learnings file inside the skill folder. The next time you run that skill, it reads those notes first and applies them before it starts.

You never have to repeat yourself. The system gets better the more you use it.

At the end of every session, running wrap-up also checks whether anything from your conversation is worth saving to your permanent memory files: preferences, decisions, context that matters next time. Two minutes at the end of each session, and nothing gets lost.

The one thing to do before you close Claude every time.

Run this every time before you close Claude. It takes about two minutes and it's what makes the system compound over time.

Type at the end of every session:
/wrap-up

What it saves

The wrap-up skill writes three things:

memory/sessions/[date].md — today's session log. What you worked on, what decisions were made, what was produced.

memory/MEMORY.md — the persistent facts index. Key decisions, preferences, and context Claude should remember across all future sessions.

context/tomorrow.md — your open threads list. Anything left unfinished, any decisions still pending, any tasks carried forward.

What happens without it

Claude's built-in memory doesn't persist between sessions. Without wrap-up, the next session starts cold — no memory of what you decided, no open threads, no captured feedback. The system loses what it learned.

With wrap-up, every session builds on the last. This is the habit that makes the whole thing work.

The skills built specifically for how you work. Start with these.

Turn one recording into a full week of content — Facebook, LinkedIn, Instagram, and email.

This is your primary content workflow. You record a voice note or video — talking about something you're thinking about, a client situation, a pattern you've noticed — and Claude turns it into a full batch of platform-ready posts.

The whole process takes about five minutes of your time. Claude does the rest.

What you get from one recording

• 2 Facebook posts (different angles and frameworks)
• 1 short Facebook post + first comment (for engagement)
• 2 LinkedIn posts
• 1 Instagram caption
• 1 Instagram carousel (slide copy)
• 1 nurture email with subject line options

Everything lands in Airtable as a draft for you to review. Nothing gets posted without your sign-off.

Save your audio or video file to this folder in Dropbox:

Folder:
brand-context/Videos and Audios to repurpose/

Any format works — .m4a, .mp3, .mp4, .mov. Name it something recognisable, like Client questions.m4a or Monday voice note.m4a. You don't need to tidy it up.

If you already have a written transcript or an .srt subtitle file, you can drop that in the same folder instead — Claude can read those directly and skip the transcription step.

Open a Claude Code session and say:

Type:
Transcribe [filename] — it's in brand-context/Videos and Audios to repurpose/

Claude will run the transcription tool and save a .md file in the same folder. The file will be named with the date and topic, like 2026-08-27-energetic-triggers.md.

Once it's saved, Claude will tell you it's done. You can read it back to check, but you don't need to — it goes straight into the next step.

If you already dropped in a transcript or SRT file, skip this step entirely.

Once the transcript file exists, say:

Type:
Repurpose this voice note — [brief description of the topic]

For example: "Repurpose this voice note — it's about how people get energetically triggered by others who are spiritually open."

The description helps Claude pick the right angles and offer connection. Keep it to one sentence.

What Claude does next

• Reads the transcript and extracts the core insight and angles
• Writes each piece using a different framework and angle — no repetition
• Runs everything through the humaniser to remove AI tells
• Saves files to projects/social-posts/[date]_[topic]/
• Saves all 7 pieces to Airtable as You to check

Claude will ask you one question at most — usually about the CTA if it's not obvious from the recording. Otherwise it runs through without stopping.

When the skill finishes, your content is in Airtable. Open the You to check view and you'll see the new posts listed.

Airtable:
your content base → SM Content → You to check

Click into each record to read the post. If you're happy with it, move it to Finalised. If you want changes, drop a note in the record or come back to Claude and say what needs adjusting.

You don't need to review everything at once. The drafts sit there until you're ready.

Run the setup skills that load your voice, strategy, and audience into Claude.

Claude Code is installed and running. Now you build out your profiles so Claude shows up already knowing your business.

This module is where you run the setup skills that generate your strategy layer, your brand voice, your audience language, and your offer files.

Run them in this order:

• setup-user-profile — asks questions about your business and how you like to work, then writes USER.md
• setup-north-star — builds your strategy layer: mission, 90-day priorities, income target
• setup-brand-voice — reads your writing samples and brand stories, extracts your voice profile
• setup-icp — builds your ideal client profile and first audience language file
• setup-brand-stories — compiles brand stories and testimonials into a story bank

If you didn't finish gathering material beforehand, don't worry. Most setup skills can walk you through it on the spot.

Your User Profile

Type run setup-user-profile in the chat panel. It will pull in any material you already uploaded to your brand-context/about-me folder. Anything missing, and it will ask you questions one at a time: your name, your brand, what you sell, who you sell it to, how you like to communicate.

Your North Star

Type run setup-north-star. This skill writes two files: ceo-north-star.md (your mission, offers, and 90-day priorities) and revenue-tracker.md (your income target and current pipeline).

SOUL.md

SOUL.md comes pre-configured in the Business Brain OS. It controls how Claude behaves in your sessions. You don't need to edit it now — over time it will update as you develop preferences. The wrap-up skill proposes updates when it detects repeated patterns of corrections. You approve or reject each one.

Verify it's working

Once you've run both setup skills, close Claude Code completely and open it again. Type: Who am I and what am I working on?

Claude should respond with your name, your brand, and your current priorities.

Setup Brand Voice

Type run setup-brand-voice in the chat panel. This skill reads everything in your writing-samples and brand-stories folders and generates a voice profile automatically — the patterns, rhythm, tone, and things to avoid that make your writing sound like you.

It produces two files: brand-context/voice-profile.md (the voice guide every content skill reads before writing) and brand-context/writing-examples.md (a curated set of gold-standard sentences from your own writing).

Test it

Once the skill has finished, ask Claude to write a short Facebook post on any topic you know well. Read it back. Does it sound like you?

If it's way off, add more examples to brand-context/writing-samples/ and regenerate. You can run setup-brand-voice as many times as you need.

Setup Brand Stories

Run setup-brand-stories after setup-brand-voice. This skill reads everything in your brand-stories and testimonials folders, compiles them into a story bank, and updates your voice profile with real material for copy skills to draw on.

Your voice profile captures how you sound. Language rules add a separate layer: what Claude is explicitly never allowed to write.

Without language rules, Claude defaults to patterns that are technically fine but feel off: em dashes, phrases like "here's the thing", structures like "it's not X, it's Y". These are so common in AI-generated content that they've become dead giveaways.

Run setup-language-rules

Type run setup-language-rules in the chat panel. The skill starts with a curated list of the most common AI writing offenders, asks you four quick questions to personalise it, then saves your file.

Your language rules save to brand-context/language-rules.md. Every content skill loads this file alongside your voice profile before writing anything. The rules are treated as hard constraints, not suggestions.

The ongoing habit

Every time Claude produces a phrase that jars — something that makes you think "I'd never say that" — tell it in the session and then say: "Add that to my language rules."

Claude opens the file, adds the phrase, confirms it's done. That phrase will never appear in your content again.

Over time, this file becomes one of the most valuable parts of your system. The more you refine it, the less editing you do.

Set Up Your ICP

Type run setup-icp. This skill builds two files: your ideal client profile and your first audience language file. These are what every copy skill reads before writing anything promotional.

If you've saved material to your audience-language folder, it picks everything up automatically. If not, it walks you through questions one at a time: who your ideal client is, what they want, how they describe their problem, the exact words they use.

How to keep adding language over time

After every discovery call, testimonial, survey response, or comment — paste the quote directly. Say to Claude: "Add this to my audience language file" and paste what you have. It goes in as-is, verbatim.

The more you add over time, the more accurately the system can write for your specific audience rather than a generic version of them.

Going deeper with mkt-icp

When you're ready to go further, run mkt-icp goes out and finds language you don't have yet — researching what your audience says in public (Amazon reviews, Reddit threads, YouTube comments) and surfacing the exact words they use when nobody's watching.

Your brand guidelines file is the single source of truth for how everything visual looks — website pages, carousels, presentations, any design Claude builds for you.

Where it lives

Your visual brand guidelines are stored in:

Folder:
brand-context/Visual Brand Guidelines/

The key file is brand-guidelines.md. It contains your full colour palette (with HEX values), your typography system (heading, body, subheading, accent fonts and their Google Fonts fallbacks), CSS variables ready to drop into any build, your logo variants and when to use each one, and usage rules for colour and type.

How it gets used

When you ask Claude to build a page or a visual, it reads brand-guidelines.md first. That means every page starts from your own colour palette, your heading font, and your body copy font — not something generic. You don't describe your brand each time; it's already loaded.

Building a page

The typical workflow is:

1. Write the copy — using service-page-writer, sales-page-writer, or your own draft
2. Say: "Design a page for [offer] using this copy" — Claude reads your brand guidelines and builds the layout using viz-interface-design
3. The output is an HTML file you paste into a Divi code block on your WordPress site

If your brand evolves

Update brand-guidelines.md — change a colour value, swap a font, update a rule. Every future build picks up the change automatically. The HTML version (brand-guidelines.html) is a visual reference you can open in a browser to see everything at a glance.

If you ever need to rebuild it

The guidelines were extracted from your brand PDF (Brand Guide.pdf in the same folder). If you ever need to rebuild the file from scratch — say, after a full rebrand — say to Claude: "Rebuild my brand guidelines from the PDF in brand-context/Visual Brand Guidelines/" and it will extract the colours, fonts, and rules and rewrite the file.

Where all your content drafts live between being written and being published.

Your Airtable base is called your content base. The main table you'll work in is SM Content.

The three views

• You to check — drafts that Claude has written and saved. These are ready for your review.
• You's feedback — posts you've flagged for rework. Add a note in the record and come back to Claude to fix them.
• Finalised — approved posts ready to schedule or publish.

You move a post between views by changing its Status field. That's all there is to it.

Other tables

• Videos and Audios — a record of source recordings and what's been repurposed from them
• Freebies — lead magnet tracking
• Testimonials & Case Studies — client results and quotes

When new content appears in You to check, open each record and read it through.

If you're happy with it

Change the Status to Finalised. It moves out of You to check and into the Finalised view.

If something needs changing

Change the Status to You's feedback and add a note in the record — what's off, what direction you want instead. Then come back to Claude and say:

Example:
"The Facebook Post 1 from the [batch name] batch needs reworking — the hook is too abstract. Can you try a more personal story angle instead?"

Claude will rewrite it and save the updated version back to the same record.

What to look for when reviewing

• Does it sound like you?
• Is the opening strong enough to stop the scroll?
• Is the offer or CTA the right one for where you are right now?

Once you're happy with a post, change its Status to Finalised. It moves into the Finalised view, where you can schedule or copy it across to your platform.

The Date and Published fields are there if you want to track when something went out — these are optional but useful if you want to look back at what's been posted.

Scheduling

Airtable doesn't post to social platforms directly. You'll copy the content into Meta Business Suite, LinkedIn, or your scheduling tool of choice. The Airtable record is your review and tracking layer — not a publishing tool.

Work through one recording's batch at a time. Review, approve, schedule — then move on to the next. It's faster than dipping in and out.

Write the copy first, then ask Claude to build the HTML. The skill reads your brand guidelines automatically so every page matches your design system.

This is a two-step workflow. The copy and the design are separate jobs — getting them in the right order means you're never retrofitting words into a layout, or redesigning around copy that's going to change.

The two steps

1. Write the copy — use a writer skill to draft it, review and approve it
2. Build the HTML — run web-page-builder, which reads your brand guidelines and existing pages to produce consistent HTML ready for Divi

What web-page-builder does

• Reads your brand guidelines (colours, fonts, CSS variables, usage rules)
• Studies the existing pages to match the established patterns
• Plans the section structure and waits for your approval before building
• Produces a single self-contained HTML file saved to brand-context/Website/pages/

Use the right writer skill for the page type:

• Service page (individual services) → service-page-writer
• Sales / programme page (signature programme, group offer) → sales-page-writer
• Short page (opt-in, thank you) → write it yourself or ask Claude to draft it directly

Say:
Run service-page-writer for my [offer name] page

The skill will ask you questions about the offer, the audience, and the keyword you want to target. Answer them one at a time. It will produce a full copy draft.

Before moving to Step 2

Read through the copy. Make any changes — add specifics, adjust the tone, swap out any lines that don't sound like you. Once you're happy with it, say: "The copy is approved — build the HTML."

With the copy approved, say:

Say:
Build the HTML for this page using web-page-builder

The skill will:

1. Read your brand guidelines and study the existing pages
2. Plan the section layout and show you the structure — hero, content sections, CTA
3. Wait for you to confirm the plan before building
4. Produce the full HTML and save it to brand-context/Website/pages/[Page Name]/

Reviewing the output

Open the saved HTML file in your browser to preview it. You can ask Claude to adjust anything — spacing, section order, colours, copy tweaks — before it goes into WordPress.

Once you're happy with the preview:

1. Open the HTML file and copy the entire contents
2. In WordPress, open the page in Divi Builder
3. Add a full-width section → insert a Code module
4. Paste the HTML into the code module and save
5. Preview the page in WordPress to check it renders correctly

Divi's own header and footer wrap the page automatically — the HTML Claude produces is just the page body content, so there's no duplication.

If something looks off

Come back to Claude and describe what's wrong: "The hero text is too small", "The CTA button isn't showing", "The spacing between sections is too tight." Claude will fix the HTML and you re-paste it.

Two settings that turn every Zoom call into usable material for Claude.

1. Switch to local recording

Zoom defaults to saving recordings to the cloud, but cloud recordings don't include a VTT transcript file. To get the transcript, you need to switch to local recording and enable the caption option.

In Zoom: go to Settings → Recording. Under Other settings, tick Save closed caption as a VTT file. Then change the save location from cloud to local, and set the folder to your Dropbox so recordings are easy to find.

At the end of each recorded call, your Dropbox folder will contain both the video and a .vtt transcript file.

Please be aware these video files can be extremely large — once you have the transcript, you may want to delete the original video file to free up space.

2. Move the transcript to your Business Brain OS folder

Copy the .vtt file across to your Business Brain OS Dropbox folder. A good place for it is source-material/calls/ so your call transcripts stay in one place and are easy to find.

You can do this yourself by dragging the file across in Finder or File Explorer, or you can ask Claude to move it for you: "Move the VTT file from [location] into my source-material/calls folder."

Once it's there, drop it into Claude and say what you want to do with it: content, session notes, a summary, action items. It can work from that directly, without you needing to retype anything.

Before you can use Business Brain OS, you need to get the files onto your computer. This module walks you through getting your copy via Dropbox — and how to pick up new skills when they're released.

• Set Up Dropbox — make sure it's installed and syncing
• Accept Your Shared Folder — the master copy of the system
• Make Your Working Copy — where you'll actually work
• Getting New Skills — how to add updates to your copy

Business Brain OS is delivered via Dropbox. If you don't already have it installed on your computer, do that first — it needs to be the desktop app, not just the website, so the files actually sync to your machine.

What to do

1. Go to dropbox.com/install and download the desktop app
2. Install it and sign in (or create a free account if you don't have one)
3. Make sure it's running — you should see the Dropbox icon in your taskbar or menu bar

If you already have Dropbox installed and syncing, you're good to move on.

You'll receive an email from Dropbox inviting you to a shared folder called Business Brain OS — Client Install. This is the master copy — it contains all the skills and shared resources. You have view access only, so nothing you do can affect it.

What to do

1. Open the invite email from Dropbox
2. Click Accept invite
3. The folder will appear in your Dropbox — check it's there before moving on

If you haven't received the invite, check your spam folder or let Cathy know.

The shared folder is the master — you don't work in it directly. Instead, you duplicate it to create your own personal copy. This is where your CLAUDE.md, context files, and personalised settings will live.

What to do

1. Open Dropbox on your computer and find the Business Brain OS — Client Install folder
2. Right-click it and choose Copy
3. Paste it somewhere sensible — your main Dropbox folder or Documents works well
4. Rename it to something clear, like Business Brain OS or My AI OS

This copy is yours. You'll customise it, add your context files to it, and work from it every day.

New skills are added to the master folder regularly. When one is released, you'll get a message letting you know — then it's a simple copy across to your working folder.

When a new skill is available

1. Open the Business Brain OS — Client Install shared folder in Dropbox
2. Find the new skill folder inside .claude/skills/
3. Copy that folder
4. Paste it into the same location in your working copy

Your context files (CLAUDE.md, SOUL.md, USER.md, etc.) are never affected — only the skills folder gets new additions.

Your Business Brain OS is built from a specific set of folders, each with a clear job. Understanding where things live means you know where to look, what to edit, and what to leave alone.

• context/ — who you are and how Claude behaves
• brand-context/ — your voice, audience language, and offer files
• memory/ — what Claude remembers between sessions (don't edit this)
• offers/ — detail files for each of your offers
• source/ — raw material you bring in (transcripts, notes)
• CLAUDE.md — the instruction file Claude reads at the start of every session

Almost everything in the Business Brain OS is stored as a markdown file — plain text with a .md extension. This module covers what that means in practice and how to work with those files confidently.

• What is Markdown? — why it's used and what it looks like
• Opening and Editing Files — how to read and update your context files
• Getting Documents In — how to bring external documents into your system

▶

This is where Claude learns who you are and how you work.

It contains files like:

• USER.md — your profile: name, business, role, how you like to communicate
• ceo-north-star.md — your strategic priorities and income targets
• soul.md — how Claude behaves: its tone, working rules, and personality. This is what makes it feel consistent every time you open a session.

Every time you open a session, Claude reads these files and that's why it can show up already knowing your context.

There's also tomorrow.md, a running list of open threads and tasks carried forward from your last session. Think of it as the handover note you leave for yourself.

This is where everything that accumulates over time lives.

Claude saves persistent facts about your business (decisions made, projects in progress, preferences Claude should remember) in one folder.

There is also a sessions subfolder with one file per day logging what happened in each session.

You don't need to manage either. They build automatically.

What you can be confident of is that Claude is remembering what you tell it, and building knowledge over time of you, your business, and your preferences. This is how your system learns and gets better over time.

▶

Everything that makes your content sound like you and speak to your audience.

Your voice profile lives here, your audience language files (the exact words your clients use, not your paraphrased version), your writing samples, and any website assets.

This is the folder the system reads before writing anything. The richer it is, the better every output will be.

▶

One file per active offer. Claude reads these when it's writing promotional content, so it knows what you sell, who it's for, what's included, and what makes it different. The more specific these files are, the better the copy.

If you sell three things, you'll have three files here. If you add a new offer, you add a new file. Simple.

Your raw material goes here. Transcripts of calls, voice recordings, workshop replays, discovery call notes — anything you want to feed into the system to generate content or capture insight.

This folder is entirely yours and is never touched when you pull updates from GitHub.

Think of it as your inbox: things come in here before they get processed into content or context.

Where the intelligence of the Business Brain OS lives.

Inside your Business Brain OS folder there's a hidden folder called .claude. You'll never need to open it directly, but knowing what's inside it helps you understand how everything connects.

skills/

The library of 50+ pre-built workflows. Each one handles a specific task — writing a Facebook post, running a keyword search, drafting an email sequence. When you run a skill, Claude finds the instructions here, loads your context files, and does the work. When you pull an update from GitHub, this is the folder that gets refreshed. Your other folders are never touched.

commands/

The slash command shortcuts for the most-used skills. When you type /content-router or /wrap-up, Claude is reading the corresponding file here. Commands are just a faster way of calling skills you use every day. There's nothing to set up — they're already configured.

hooks/

Small scripts that run invisibly in the background. These fire automatically at certain moments — for example, loading your memory snapshot at the start of every session so Claude already has your context before you've typed a word. You never call these directly and don't need to think about them. They just run.

The short version: skills do the work, commands are shortcuts to the most-used skills, hooks handle the behind-the-scenes setup.

▶

The master instruction file. This is the first thing Claude reads at the start of every session. It tells Claude what context files to load, how to behave, what skills are available, and how this system works.

You don't need to edit it. It comes pre-configured for the Business Brain OS.

Everything slots into place once you start running the setup skills in the Configure Your System module. For now, just know where things live.

▶

How to open, read, and edit the files your system runs on.

Almost everything in the Business Brain OS is stored as a markdown file — a plain text file with a .md extension. Your brand voice, your audience language, your offer details, your daily memory, your context files: all markdown.

If you've never heard of markdown before, don't worry. You don't need to learn it properly. But knowing what it is and how to handle it will save you confusion later.

Why Markdown?

Claude reads plain text files natively and quickly. Markdown adds basic structure (headings, bullet points, bold text) without needing a word processor. The files are small, portable, and easy to version-control — which is how GitHub tracks updates to your skills without touching your personal files.

Most importantly: Claude can read, write, and edit markdown files directly inside your session. You don't need another app.

What it Looks Like

Markdown is just text with a few simple symbols. A # at the start of a line makes a heading. **text** makes it bold. A - at the start of a line makes a bullet point. Everything else is a normal paragraph.

That's 90% of what you'll ever see in your files. The rest Claude handles for you.

▶

Inside Claude Code, click any .md file in the sidebar. It opens in a panel on the right. You can read it there, and you can ask Claude to make changes: "Update my north star file to add X" or "Add this to my writing samples."

You can also edit files directly. Click the pencil icon in the panel to switch from preview to edit mode, make your change, and save.

Outside Claude Code, any text editor will open a markdown file: Notepad on Windows, TextEdit on Mac, or VS Code if you have it. It's just text.

Never open a markdown file in Microsoft Word. Word adds invisible formatting that can break how Claude reads the file.

You'll sometimes want to bring in content from outside: a Google Doc, a PDF, an old email, notes from a call. Here's how.

• If you can copy the text — paste it into the chat. Then ask Claude to save it: "Save this as a file in brand-context/writing-samples/ and call it email-example-1.md." Claude creates the file, formats it cleanly, and puts it in the right place.
• For Google Docs — copy the text and paste it into the chat. The formatting won't come through cleanly, but a quick "tidy this up and save it" prompt fixes it in a few seconds.
• For PDFs — either paste the text (if it's selectable) or drag the file into the Claude Code chat window. Claude can read PDFs directly and pull out what you need.

Before you start configuring your system, you need to be comfortable in the tool that runs it. This module walks you through getting Claude Code installed, understanding the interface, and knowing what it can and can't do.

• Install and Open — get Claude Code running and open your folder
• The Interface — the chat panel, the file sidebar, and the modes
• Permissions — what Claude is and isn't allowed to do
• Connecting External Tools — how Airtable and other MCPs connect
• Your Memory System — how Claude remembers things between sessions

▶

Get comfortable in the interface before you start configuring anything.

• Install the Claude Code desktop app from claude.com/download and sign in
• Open Claude Code, click File → Open folder, navigate to the folder you cloned and open it
• Confirm it's working: Claude should greet you and reference the Business Brain OS

You've downloaded the Business Brain OS and know what's inside it. Now you need the engine that runs it. By the end of this module you'll be comfortable in the interface, you'll understand what Claude can and can't do, and you'll be ready to start configuring the system.

When you open Claude Code, you'll see three main areas.

1. The Chat Panel

The middle of the screen — where your conversations with Claude take place. You type at the bottom prompt, press Enter, and Claude responds in the space above.

The context bar shows how much of the chat panel is being filled as the conversation grows. When it gets close to full, Claude starts to lose track of things said earlier in the session. Around 120K tokens is a good time to wrap up rather than pushing on.

Modes: Leave it on Auto for day-to-day use. Switch to Plan if you're about to ask Claude to do something big and want to review the steps first. Don't use Bypass permissions.

Models: Use Sonnet 4.6 as your default. Switch to Opus only if you're doing something that genuinely needs more reasoning power.

2. The Left Sidebar

Your conversation history. Every session you've had in Claude Code is listed here, most recent at the top. You can pin sessions, rename them, and group related ones into folders.

3. The Right-hand File Panel

Your window into the folder your system lives in. Click a file to open it in preview. Click the pencil icon to switch to edit mode.

Use it for: checking what's in a context file, making a quick edit, confirming Claude saved something correctly. For rewriting, reorganising, or updating skills — ask Claude to do that instead.

▶

The mode selector sits at the bottom of the chat panel. There are four options — here's what they mean and which one to use.

Auto (use this one)

Claude acts on your instructions without asking permission for every step. It still respects the hard limits built into the system — it can't delete files, log into anything, or do anything outside its configured permissions. But within those limits, it just gets on with it. Use Auto for everything.

Plan

Before doing anything, Claude writes out a step-by-step plan and shows it to you. You can read it, revise it, ask questions, and approve it before anything actually happens. Good for big or complicated jobs where you want to see the approach first. It also uses fewer tokens at the planning stage, so it's worth using for large multi-step tasks.

Manual

Claude asks for permission at every individual step. This slows everything down significantly. You can use it if you're curious to see exactly what Claude is doing under the hood — but it's not practical for day-to-day use.

Bypass Permissions

Don't use this. It removes the safety guardrails built into your system and lets Claude override the configured limits. There's no reason to ever switch to this mode.

▶

The context bar (click the icon next to the mode selector) shows how full your current chat is. Every time you send a message, Claude reads the entire conversation from the top before replying — so the longer the chat, the more it has to hold in memory, and the more likely it is to start making mistakes.

The limit is 200,000 tokens. When you're at 50–70%, it's time to think about wrapping up. Don't wait until Claude auto-compacts the chat — you want to control what gets saved.

The right approach: /wrap-up

When your context is getting full, run /wrap-up. This skill saves anything outstanding to tomorrow.md, captures any learnings, and compacts the conversation cleanly. Then start a fresh chat — Claude will load your context files and pick up exactly where you left off.

You can also run /wrap-up at the end of any session, even if the context isn't full. It's good practice to close every session this way.

Let's talk about the context bar. You open it from the icon next to the mode selector, and it tells you how full your current chat is.

Within a chat, every time you ask Claude to do something, it goes to the top of the conversation and reads through everything, every single time. So the longer the chat gets, the more it has to read — and the more tokens it uses each time, because it's reading all of that and then doing the next step. As a result, the fuller the context window gets, the more likely Claude is to start getting things wrong. It's got too much to hold in memory.

And then what also happens is that once it hits the 200,000 token limit, it will start compacting the chat itself — keeping what it thinks is relevant and compressing the rest to give itself more room. You don't want that to happen on its own, because you're not in control of what gets kept.

So a good rule of thumb is: when your context window gets to around 140,000 of 200,000 — you know when it's really getting up there — that's when you want to tell Claude to wrap up the conversation. Run /wrap-up and it will go through the memory, pull out anything relevant from the conversation, capture any learnings, and save anything outstanding to your tomorrow.md. Then you can start a fresh chat with confidence that nothing's been lost.

There's also a /compact command you can use at any point if you want Claude to compress the chat without fully wrapping up. But the best practice is always /wrap-up — it saves what you need, tells you what it's doing, and you stay in control.

▶

Your chat history lives here. Every session you've had in Claude Code is listed in order, most recent at the top. You can rename sessions and pin the ones you're still working on.

Groups

You can organise chats into groups, but be aware that Claude periodically asks you to log out and back in — and when that happens, groups reset and everything goes back into one flat list. It's not worth relying on groups for organisation.

Pinning and archiving

Pin any chat you haven't fully wrapped up yet. Once you've run /wrap-up and confirmed nothing's been lost, archive it. Treat archived chats as gone — you may not be able to retrieve them. That's fine, because the memory system has captured everything worth keeping.

Artifacts and Routines

Artifacts are visual outputs Claude generates (HTML pages, diagrams). They're temporary scratch space — once you're happy with something, get it saved properly into your folder. Routines let you schedule recurring tasks (e.g. daily content creation). They currently only run when your laptop is open.

The left sidebar is the other part of the Claude Code interface. Every time you have a chat or conversation with Claude, it saves it here on the left, in order of what you've been working on. You can rename any chat if you want to, and you can create groups.

I used to do this — I was very organised about it. If it was for a client, I'd create a group for the client. If it was anything to do with updating my website, I'd create a group for that. But what was happening was that Claude periodically asks you to log out of the desktop app and log back in. Every single time I did that, all my groups were ungrouped and everything went back into one big list. After the third or fourth time that happened I thought, there's no point. If it's not going to stay grouped, I'm not going to bother in the first place.

So what I do instead is pin conversations. I pin anything that isn't quite 100% finished — anything I'm still actively working on. I try to be disciplined and run the wrap-up at the end of most chats, and as long as I'm happy that I'm continuing in the way I wanted and nothing's been lost, I'll archive them. They're kind of gone once they're archived — a few months ago I was able to pull them back up, I tried again recently and I wasn't able to. So just have the mentality that if a chat's archived, the actual conversation in it is gone. But that's why we've got the memory system — all the relevant stuff that came out of that conversation has been saved. You're not actually losing anything.

Artifacts are the visual outputs Claude generates — HTML pages, diagrams, that kind of thing. I use these as a kind of scratchpad, a way to test ideas. Once I'm happy with something, I get it saved properly as the HTML file or whatever it needs to be, back into my folder so it's in the system.

Routines are for setting Claude up to run things on a schedule — once a day to create content, to go through a folder and tidy things up, whatever you want. They tend to only run when your laptop is open rather than just running at all times. That may change in the future, but that's where we're at.

Dispatch lets you send tasks to Claude from your phone. I've played with it and couldn't really get it working reliably — it's been a bit glitchy. I've actually decided that's probably fine. I don't necessarily want to be working all the time, every minute of the day, so for me that boundary actually works.

▶

The $20/month Claude Pro plan gives you a usage allowance split across two levels: a weekly total and a 5-hour session limit. Most day-to-day marketing work stays well within both.

Weekly allowance

You're given a set number of tokens per week, which resets on Saturday. Keep an eye on it if you're doing heavy work — building systems, processing large files, or running multiple complex sessions in a day.

5-hour session limit

Within any 5-hour window, there's a cap on how many tokens you can use. If you hit it, you have two options: wait for it to reset, or add usage credits to your account which top you up automatically when you hit the limit.

If you hit your limit

Use the time for work that doesn't need Claude: recording videos, reviewing content, admin tasks. Most users only bump into the limit occasionally and during heavier-than-usual sessions.

Something to be aware of is your usage limits. I'm on the $20 a month plan. I use Claude pretty much all day every day, and most of the time I get what I need within my five-hour limit.

The way it works is that your allowance is split into two levels. You get a certain amount each week — mine resets on Saturday. I very rarely come up against this as an issue. I've been doing quite heavy work lately — building dashboards, building course content — so I'm using it more heavily than I usually would and I keep an eye on it. But for normal marketing work, the weekly allowance is rarely a problem.

Within each session you also get a limit on how many tokens you're allowed in a five-hour window. Again, for most of the marketing work I do, that's loads — I rarely bump up against it. But when I do, I've got a couple of options. I can just sit it out and wait till it resets. I tend to use that time to think: okay, what can I work on that isn't AI-driven? There's usually stuff that just needs to be done manually. Recording videos, for example — I've recorded videos while waiting for the limit to reset. Or you can add credits to your account, and then once you hit your limit it will tick over and use your usage credits automatically.

A note on usage credits: sometimes they burn through faster than you'd expect, especially during heavy sessions where you're building systems or doing a lot of back and forth. The last couple of weeks I've found it's been pretty good, but it can vary. So it's always worth keeping an eye on where you're at rather than assuming credits will stretch indefinitely.

Claude Code is not running in the background. It doesn't do anything unless you open a session and give it an instruction. When you close the app, nothing runs.

The Business Brain OS comes pre-configured with a permissions file that controls exactly what Claude is and isn't allowed to do.

What it's allowed to do

• Read, edit, and write files inside your Business Brain OS folder
• Run specific scripts that power the skills
• Basic file navigation: listing folders, checking where things are

What it's blocked from doing

• Installing software — blocked by default
• Deleting files or folders permanently — blocked by default
• Reading your credentials or .env files

Even if Claude misunderstood an instruction and tried to delete a folder, the system would stop it.

The mode selector controls how much Claude checks in with you before acting — not what it's capable of doing. The guardrails above apply regardless of which mode you're in.

Left to itself, Claude can only work with files on your computer. It can't see your email, your Google Drive, your Airtable, or any other external tool — unless you connect it.

That's what MCP connections do. MCP stands for Model Context Protocol — the standard that lets Claude plug into external apps. Each connection is something you set up deliberately, and it only gives Claude access to that specific tool.

What becomes possible with MCP connections

• Airtable — Claude can read your content queue, save posts, and update records directly
• Google Drive — Claude can read documents from a link, without you copying and pasting
• WordPress — Claude can publish blog posts as drafts straight from a session

If a connection isn't set up, Claude simply can't access that tool. And the connections only work within the scope of what you've configured.

Airtable uses a Personal Access Token to allow Claude to read and write to your content system. You generate it once in Airtable, then share it with Cathy so she can connect it to your system.

How to generate your token

• Go to airtable.com/create/tokens (or: sign in → click your avatar top right → Developer hub → Personal access tokens)
• Click Create new token
• Give it a name (e.g. "Claude Code")
• Set the scopes: data.records:read, data.records:write, schema.bases:read
• Under Access, choose your workspace so the token has access to your bases
• Click Create token
• Copy the token immediately — Airtable only shows the full token once, at this step. After that it's hidden for security.

Once you have the token, don't paste it into the chat — chat messages can be logged or visible in session history. Instead, open Claude and say: "I have my Airtable Personal Access Token — where should I save it securely so you can use it?" Claude will tell you exactly where to put it on your computer so it stays private.

Every time you open Claude Code in your Business Brain OS folder, the first thing it does is read your CLAUDE.md file. From there, it reads the relevant files depending on what you've asked it to do.

Your open threads list

Your open threads list is saved in a file called tomorrow.md that carries forward between sessions. Anything left open at the end of a session gets saved here by the wrap-up skill. When you open a new session, Claude reads this file and picks up where you left off.

The context window

You'll notice a progress bar at the top of the chat panel. This is the context window — the total amount of conversation Claude can hold in memory at once. Around 120K tokens is a good point to either wrap up or use /compact.

/compact

When you're in the middle of something and don't want to stop, type /compact and save to working memory what we're working on. Claude compresses the conversation history and saves a summary so it doesn't lose the thread.

The wrap-up skill

At the end of every session, type /wrap-up. It writes to three files:

• memory/sessions/[date].md — today's session log
• memory/MEMORY.md — the persistent facts index
• context/tomorrow.md — your open threads list

Without running wrap-up, you risk losing all the learnings and context from that session. This is the habit that makes the whole system compound over time.