chlkey_

Learning Out Loud

chl0key's Complete Playbook

Build Your AI-Powered Second Brain

A comprehensive, step-by-step guide to setting up Claude CoWork as a personal operating system. Based on 25+ iterations, 20+ research sessions, and months of real-world use. Want the short version first? Read What a Tuesday Looks Like When AI Runs Your Life.

1 Session to Start

~45 Minutes to Set Up

∞ Hours Saved

Before You Start

Make sure you have these three things in place:

A Claude Pro or Max subscription. CoWork requires a paid plan. Sign up at claude.ai if you haven't already.

The Claude desktop app. Download it from claude.ai/download. CoWork runs inside the desktop app, not the browser.

Open CoWork. Launch the desktop app and click "Cowork" in the top navigation bar (next to Chat and Code). That's the interface you'll use for everything in this guide.

The CoWork home screen — where you start new sessions and access your workspace.

A CoWork session in progress — your conversation, progress tracker, workspace files, and context panel.

Overview

How the System Works

Your CoWork second brain has three core ingredients: context (who you are), instructions (how Claude should behave), and automations (work that happens while you're away). Everything else builds on these.

Global Instructions

How Claude should behave

CLAUDE.md

Who you are

Memory Files

Deep context on specific topics

Folder Structure

Where everything lives (PARA)

Custom Skills

Pre-built workflows

Scheduled Tasks

Automations that run while you're away

Session Continuity

So Claude picks up where it left off

Connected Tools

Gmail, Calendar, Notion, Drive

Quick Capture

Send tasks from your phone

Before you start: here's how this works

Read the whole guide once to understand the layers. Then open a CoWork session and work through it section by section WITH Claude.

Every section below has a ready-to-paste prompt you can copy straight into CoWork. Claude will do the heavy lifting — interviewing you, drafting files, building your system. You don't have to think of what to say. Just paste the prompt, answer Claude's questions, and move to the next section.

You don't need to do everything in one sitting. Start with Discovery, build from there, and refine as you go.

Part 0

The Discovery Process

Before you fill in a single template, you need to have the right conversations with Claude. The quality of your system depends entirely on the quality of information you put into it.

Here's what to do: Open Claude Desktop, click Cowork at the top, and start a new session. Then copy the first prompt below and paste it into the text box. Claude will start asking you questions — just answer them naturally. When you finish one prompt, paste the next one into the same chat and keep going.

Keep all 4 prompts in one session

Do all 4 Discovery prompts in the same CoWork chat. Don't start a new session between them — Claude needs the context from each conversation to build on the next. Paste one prompt, answer the questions, then paste the next when you're ready. You can take breaks, just come back to the same session.

Who You Are (15-20 min)

I'm setting up my CoWork second brain from scratch. I want you to interview me so you can build a system that actually knows me. Start with who I am. Ask me these questions ONE AT A TIME (wait for my answer before moving to the next). Go conversational, not robotic: 1. What do I do for work? Not just my title, but my actual day-to-day. 2. What am I working toward right now? Career goals, personal goals, active projects. 3. Walk me through a typical weekday. Then a typical weekend. 4. What tools do I already use? (Note-taking, calendars, email, project management) 5. What's broken in my current system? What falls through the cracks? 6. What have I tried before that didn't work? (And why did it fail?) After I answer all 6, summarize what you learned and ask if I'd add or correct anything. Then save the notes to a file called "discovery-notes.md" in my workspace so we can reference them when building the system. Don't limit yourself to just these questions — if something I say sparks a follow-up or you think of other questions that would help you understand me better, ask those too.

Claude will ask each question one at a time and wait for your answer.

Questions covered in the prompt above: your work, goals, daily routine, current tools, what's broken, and what you've tried before.

How You Work (10-15 min)

Now let's talk about how I work. Ask me these one at a time: 1. How do I like information presented? Bullets, prose, visual, or something else? 2. Am I a morning person or a night person? When am I most productive? 3. When I ask you to do something complex, do I want you to ask me questions first, or just go and show me the result? 4. What makes me feel productive vs. overwhelmed? What's the difference? 5. What's my biggest time sink right now? 6. Do I work in sprints or steady-state? How long can I focus before I need a break? After all 6, add these notes to the discovery-notes.md file under a "How I Work" section. These are starting points — ask me any other questions that come to mind about how I work best. The more you know, the better the system will be.

The Personal Stuff (10-15 min)

Now the personal stuff. This is what makes the system actually feel like mine instead of generic. Ask me one at a time: 1. Who are the most important people in my life? Start with work (boss, direct reports, key collaborators), then personal (partner, family, close friends). Get names and enough context that you'd never have to ask "who is [name]?" again. 2. What are my hobbies and interests? Include the guilty pleasures and the embarrassing ones. 3. What do I eat? Am I picky? Any allergies or strong preferences? Favorite cuisines? (This matters for restaurant recs, travel planning, meal ideas.) 4. What words, phrases, or writing styles make me cringe? What makes AI output feel fake or "off" to me? 5. Are there any personality patterns that affect how I work? (Perfectionism, procrastination, avoidance, decision fatigue?) Add to discovery-notes.md under "Personal Context." Go beyond these questions if you want — ask me anything else that would help you actually know me as a person, not just a user profile.

Your AI Relationship (5-10 min)

Last discovery section — my relationship with AI. Ask one at a time: 1. What AI tools have I used before? What worked, what didn't? 2. How technical am I? Do I code? Do I want to? 3. If this CoWork system could do ANYTHING for me, what would it do? Dream big. 4. What am I NOT comfortable with Claude doing? Any boundaries? 5. What would make me stop using this system? (What's my friction threshold?) Add to discovery-notes.md under "AI Relationship." Then give me a full summary of everything you learned across all 4 sections. We'll use this to build the actual system files next. If there's anything else you're curious about that would help you build a better system for me, ask. These questions are just the starting framework.

Why this matters

Without this step, every time you open a new session you'd have to re-explain who you are, what you're working on, and how you like things done. The Discovery conversation gives Claude enough context about your life that it can actually help you — not just answer generic questions. Think of it like onboarding a new assistant: the more they know upfront, the less you have to repeat yourself later. You don't need to get everything perfect right now. The system is designed to learn and improve as you use it.

Part 1

Global Instructions

In Discovery, you told Claude about yourself. Now you're going to turn that into a set of rules Claude follows every time you open a new session. These are called Global Instructions — they live in Claude Desktop > Settings > Instructions and load automatically.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will read your Discovery notes from the previous session and use them to draft your instructions.

First, read the file called discovery-notes.md in my workspace — it has everything from my Discovery conversation. Now let's build my Global Instructions. These go into Claude Desktop > Settings > Instructions. Using everything from that file, draft my Global Instructions. It should include these sections: - "Who I Am" (1-2 sentences) - "How to Work With Me" (behavioral rules for every session) - "Output Style" (tone, format, length preferences based on what I told you) - "Standing Rules" (non-negotiable rules that apply every time) - "Quality Standard" (how to check your own work) - "How to Work Smarter" (parallel research, extended thinking, verification) Show me the draft. I'll tell you what to change. After I approve it, save it as GLOBAL-INSTRUCTIONS.md in my workspace. Then tell me exactly how to paste it into Settings. Feel free to ask me additional questions beyond this list if there's anything else that would help you write better instructions for yourself.

Claude will draft the file based on your discovery conversation, then ask for your approval before saving.

CoWork will walk you through the process, but once you have your Global Instructions drafted, here's a visual reference for getting them saved to your settings.

What to look for in Claude's draft

When Claude shows you its draft, these are the sections that make up a strong set of Global Instructions. Use this as a checklist to make sure nothing important is missing.

Required Who I Am — name, role, where your files live

Required How to Work With Me — read CLAUDE.md first, ask clarifying questions, reference folders

Required Output Style — tone, format, length preferences

Required Standing Rules — rules for every session, no exceptions

Important Quality Standard — re-read output, verify facts, flag uncertainty

Optional How to Work Smarter — you can skip this for now and add it later (click to learn more)

DIY template — if you want to build it yourself

If you'd rather write your Global Instructions manually instead of using the prompt above, copy this template into a new .md file in your workspace folder and fill in each section. (In CoWork, you can ask Claude: "Create a new file called GLOBAL-INSTRUCTIONS.md with this template.")

You'll see references to folders and files that don't exist yet (like memory/ and 04-archive/). That's fine — you'll build those in later sections. Everything connects.

## Who I Am
[1-2 sentences. Your name, what you do, where your files live.]

## How to Work With Me
- ALWAYS read CLAUDE.md first before starting any task
- For any non-trivial task: ask me clarifying questions BEFORE doing work
- Reference my folder structure — know where things go
- Check the memory/ folder for deep context on specific topics
- Use date prefixes (YYYY-MM-DD) for any time-sensitive files

## Output Style
- [Your preferred tone]
- [Your preferred format — markdown? bullets? prose?]
- Always end substantive work with a "Next Steps" section

## Standing Rules
- If you make a mistake: (1) fix it, (2) log it in CLAUDE.md,
  (3) update relevant memory files
- Never delete files — move to 04-archive/
- Check 03-resources/skills/ for custom skills before starting

## Quality Standard
- Re-read output before delivering. Check for accuracy, tone,
  completeness, and whether it answers what was asked.
- If unsure about a fact, say so.

Real-life example — see what a finished version looks like

This is from a real system. Yours will look different — the point is to see the level of specificity that makes Global Instructions useful.

## Who I Am
I'm Chloe Keywell. I'm leading AI upskilling across non-product
functions at Bloomberg LP and I'm actively job searching for roles
in the tech/AI industry. My second brain lives in the folder
you're connected to.

## How to Work With Me
- ALWAYS read CLAUDE.md first before starting any task
- For any non-trivial task: ask me clarifying questions BEFORE
  doing work
- Reference my folder structure (PARA method) — know where things go
- Check the `memory/` folder for deep context on specific topics
- Use date prefixes (YYYY-MM-DD) for any time-sensitive files

## Output Style
- Concise and direct — no corporate fluff
- Markdown by default
- Always end substantive work with a "Next Steps" section
- When writing for my job search: use tech/AI industry language,
  emphasize transferable skills from Bloomberg

## Standing Rules
- If I drop something in `00-inbox/`, process it: summarize, tag,
  and suggest where it belongs
- If you make a mistake or I correct you: (1) fix the mistake,
  (2) add the learning to "Corrections & Learnings" in CLAUDE.md
- When doing research, always cite sources and include dates
- For job-related tasks, keep a running log in
  `01-projects/job-search/`
- Never delete files — move to `04-archive/` instead

Global Instructions

How Claude should behave. The rules it follows in every session — ask questions first, use your voice, check its work. This is what you just built.

CLAUDE.md (up next)

Who you are. Your context, your people, your projects, your preferences. This is the knowledge base you'll build in the next section.

Part 2

CLAUDE.md — The Master Memory File

Global Instructions tell Claude how to behave. CLAUDE.md tells Claude who you are. It's the single most important file in your system — it lives at the root of your workspace and Claude reads it automatically at the start of every session.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will read your Discovery notes and Global Instructions from the previous sessions and use them to draft your master memory file.

First, read these files in my workspace: discovery-notes.md and GLOBAL-INSTRUCTIONS.md. Those have everything from my previous sessions. Now let's build my CLAUDE.md — the master memory file that you'll read at the start of every session. Using those files, draft a CLAUDE.md with these sections. Ask me any follow-up questions you need to fill gaps — don't guess: 1. "Who Is [My Name]" — A paragraph about me. Not a resume. Who I am as a person. 2. "Quick Reference" — My email, voice description, tools I use, calendar setup. 3. "Key Constraint" — The single biggest limitation on when/how I use Claude. 4. "How [My Name] Works" — Communication style, what makes me productive, personality patterns. 5. "Key Pain Points" — The problems I want this system to solve. 6. "Key Memory Files" — A routing table (we'll create the memory files next, but set up the links now). 7. "Folder Structure" — The PARA structure we'll create. 8. "Output Preferences" — Tone, format, length by task type, date format. 9. "Things to Avoid" — Anti-patterns, banned words, behaviors I hate. 10. "Corrections & Learnings" — Empty section to start (this grows over time). Also add a "How to Work With Me" section that includes this rule: Always use AskUserQuestion to ask clarifying questions before starting any complex task. Don't guess — check first. This applies to research, email drafts, trip planning, anything with multiple possible approaches. Keep it under 150 lines. Link to memory files for detail instead of putting everything here. Show me the draft, get my approval, then save as CLAUDE.md. Ask me any other questions that would give you helpful context — anything about my life, preferences, or habits that would make you a better assistant.

Make Claude ask before it assumes

CoWork has a built-in feature called AskUserQuestion — it lets Claude show you a set of options to pick from before it starts working. Instead of guessing what you want and getting it half-right, Claude checks first. It looks like a little card with choices you can click.

This is especially useful for complex tasks like "plan a trip" or "draft an email" where there are a dozen ways Claude could go. One quick question up front saves you from a mediocre output you have to redo.

The paste prompt above already tells Claude to add this to your CLAUDE.md. Once it's in there, Claude will do this automatically on every task that needs clarification — you don't have to ask for it.

What to look for in Claude's draft

When Claude shows you the draft, check that it includes these sections. Some reference things you'll build in later steps (like memory files and folder structure) — that's expected. Claude will set up the links now and you'll fill them in as you go.

Required Who Is [You] — paragraph about who you are, not a resume

Required Quick Reference — email, voice, tools, calendar

Required Key Constraint — the single biggest limitation on how/when you use Claude

Required How to Work With Me — ask before assuming (AskUserQuestion), clarify before complex tasks

Required How [You] Works — communication style, decision-making, personality patterns

Required Key Pain Points — what's broken (the problems you want solved)

Required Memory File Routing — which files to read for which tasks

Important Folder Structure — map of your PARA system

Important Custom Skills — list with trigger phrases

Important Key Automations — what runs and when

Important Output Preferences — tone, format, length, date format

Important Things to Avoid / Anti-Patterns — banned words, behaviors

Required Corrections & Learnings — grows over time, logged mistakes

Optional Session Continuity — session-log.md instructions

Optional Composable Skill Chains — multi-skill workflows

Optional Visual Design Preferences — for HTML outputs

Optional Recurring Dates — monthly/yearly events Claude should know

Pro tip

Keep it under ~150 lines. Detailed info goes in memory files (the next section). CLAUDE.md is the index, not the encyclopedia.

Pro tip

Be specific about what you hate. The "Things to Avoid" section is just as important as the rest. Telling Claude what NOT to do saves you from correcting the same mistakes over and over.

Pro tip

Update it over time. Paste this into any session: "Re-read CLAUDE.md and update it with anything new you've learned about me in this conversation."

Part 3

Memory Files

CLAUDE.md is the overview — it tells Claude the big picture. Memory files go deeper. Each one is a detailed reference on a single topic (how you write, who's in your life, your career context). Claude only reads the ones it needs for the task at hand.

Without memory files, Claude forgets your preferences between sessions. With them, it already knows who the key people in your life are, how you actually sound in emails, and what your daily routine looks like — without you having to explain it every time.

Start a new CoWork session for this section. The first prompt builds your most important memory file — your writing voice. This one takes the most time because you'll paste in real writing samples.

Pro tip

Real writing samples beat descriptions. One real email teaches Claude more about your voice than a paragraph of adjectives. Have a few recent emails or messages ready to paste in before you start.

Core Memory Files

Everyone should have these five. They cover the 80% of context Claude needs most often. Click any card to load its prompt below, then copy and paste it into CoWork.

Start here

writing-voice.md

How you actually sound — samples, quirks, banned words

Core

people.md

Key people in your work and personal life

Core

life-areas.md

Routine, hobbies, goals, personality

Core

career-profile.md

Career history, target roles, job search

Core

glossary.md

Acronyms, nicknames, jargon, shorthand

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. Let's build my memory files. These are the deep-context files that make you actually useful. We'll start with the most important one: writing-voice.md. For writing-voice.md, I need you to ask me: 1. How would a friend describe the way I write? What's my vibe? 2. Ask me to paste 3-5 real emails or messages I've sent recently — to different people in different contexts (a colleague, a friend, a professional contact). I'll paste them in. 3. How do I sign off emails? Does it change by person? 4. Do I use exclamation points, emojis, lowercase, any typing quirks? 5. What words or phrases would I NEVER use in writing? 6. How does my tone shift between casual and professional? After I answer, build the full writing-voice.md file: voice characteristics, professional writing style, casual style, "what my voice is NOT," banned words, and paste in my real writing samples with annotations about what to notice in each one. Show me the draft before saving to memory/writing-voice.md. Ask me anything else about how I communicate that would help you nail my voice — these questions are a starting point, not the whole picture.

This is the most important memory file. Real writing samples are the magic ingredient.

You can stay in the same session and keep going through each file, or come back later — just paste the next prompt when you're ready.

Part 4

Folder Structure (PARA)

You've built your instruction files and memory files. Now Claude needs to know where to put things — and where to find them.

Without a folder structure, Claude dumps everything into one place. Your job search notes mix with your grocery lists. Files pile up with no logic. And when Claude needs to find something you made three weeks ago, it can't. A folder structure is like giving Claude a filing cabinet with labeled drawers — it always knows where things go and where to look.

PARA is just a way to organize folders by purpose: active Projects, ongoing Areas of your life, Resources you reference, and an Archive for anything you're done with.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will read your existing files and build the folder structure around them.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. Let's set up my PARA folder structure. Before you create anything, ask me: 1. What active PROJECTS do I have right now? (Things with a clear goal and end date — job search, a side project, an event, etc.) 2. What ongoing AREAS of my life do I want you to help manage? (These never "end" — career development, health, finances, learning, life admin, etc.) 3. What REFERENCE material do I come back to regularly? (Templates, guides, saved research) 4. Is there anything I've been meaning to organize but haven't? (That goes in the inbox.) Based on my answers, create the full folder structure: - 00-inbox/ - 01-projects/ (with a subfolder for each project I mentioned) - 02-areas/ (with a subfolder for each area) - 03-resources/ (including skills/, templates/, skill-feedback/) - 04-archive/ - 05-automations/logs/ - memory/ Create all folders and add a brief README.md in each project folder explaining what it's for. Then update my CLAUDE.md with the folder map. Ask me if there's anything else I'd want organized — side projects, collections, reference material — anything that should have a home in this system.

00-inbox/          → Drop zone. Unsorted captures, raw ideas.
01-projects/       → Active projects with goals and end dates.
   [project-name]/ → One folder per project.
02-areas/          → Ongoing life areas. No end date. (Career, health, etc.)
   [area-name]/    → One folder per area.
03-resources/      → Reference material. Templates, skills, research.
   skills/         → Custom Claude skills (see Part 7).
   templates/      → Reusable templates for common outputs.
   skill-feedback/ → Logs of what worked per skill.
04-archive/        → Done or dormant. Nothing gets deleted.
05-automations/    → Automation configs and logs.
   logs/           → Output logs from scheduled tasks.
memory/            → Deep context files Claude reads.

Intermission

Go touch grass. Seriously.

You just built a working second brain. Claude knows who you are, how you think, and where everything goes. That's the hard part — and it's done.

Discovery Notes ✓ Global Instructions ✓ CLAUDE.md ✓ Memory Files ✓ Folder Structure ✓

Everything from here on is optional — but it's where things get really fun. These sections build on the foundation you just created. Come back whenever you're ready.

Part 5

Session Continuity

You've got a second brain. But right now, every time you start a new session, Claude doesn't know what happened in the last one. These two simple files fix that — they're the handoff notes (like leaving a note for the next person who sits down at your desk) that keep everything connected.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will set up the files that let it pick up where it left off between sessions.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. Set up my session continuity system. Create these two files: 1. session-log.md — A shift handover file. The template should have sections for: session date, session type, what happened, completed items, key decisions made, pending items, files modified, and "next session could" suggestions. Start with an entry for today's session summarizing everything we've set up so far. 2. TASKS.md — A task tracker with sections: Active, Waiting On, Someday, Done. Populate it based on anything I've mentioned wanting to do during our setup conversations. Then update CLAUDE.md to include Session Continuity instructions: - "At the END of every session: Write a shift handover to session-log.md" - "At the START of every session: Read session-log.md before beginning any work" Show me both files before saving. If you have other ideas for how to maintain continuity between our sessions — beyond what's listed here — suggest them.

What it creates

session-log.md

The note Claude leaves for itself between sessions. At the end of every session, Claude writes what happened, what's pending, and any decisions made. At the start of the next session, Claude reads it and picks up where it left off.

Sections: Date, session type, what happened, completed items, key decisions, pending items, files modified, what next session could do.

What it creates

TASKS.md

A simple, always-current task list. Claude reads and writes to it. Sections: Active, Waiting On, Someday, Done. Check anytime by asking "what's on my plate?"

What Claude will add to your CLAUDE.md:

## Session Continuity
- At the END of every session: Write a shift handover to session-log.md
- At the START of every session: Read session-log.md before beginning work

Part 6

Connectors

You just set up session continuity — Claude can pick up where it left off. Now let's give it access to the tools you already use every day. When Claude can read your email, check your calendar, and search your files, it stops being a smart assistant and starts being a real teammate.

What's a connector?

A connector is how Claude gets permission to access an outside service — like Gmail, Google Calendar, or Notion. Think of it like giving a new coworker login access to your tools on their first day. Once connected, Claude can read your emails, check your schedule, search your files, and more — without you having to copy-paste anything into the chat.

You connect these through CoWork's settings, not through a chat prompt. Claude can't connect them for you — you have to grant access yourself. But once you do, every session after that has access automatically.

How to Connect a Tool

This happens in CoWork's settings, outside of any chat session. Here's the flow:

Step 1 Open Claude Desktop and click CoWork at the top

Step 2 Click the gear icon (settings) — it's at the bottom left of the CoWork panel

Step 3 Find the Connectors section — you'll see a list of available services

Step 4 Click Connect next to the service you want (e.g., Gmail)

Step 5 A browser window opens — sign in and grant permission like you would for any app

Step 6 Done! The service now shows as connected. Claude can access it in every future session.

Steps 1–2

Steps 3–4

Which Connectors to Start With

There are a lot of connectors available, but don't try to connect everything at once. Start with the ones that give you the most value immediately. You can always add more later.

Service	What It Unlocks	Priority
Gmail	Inbox triage, email drafting, morning briefings, thread search	Start here
Google Calendar	Meeting prep, daily agenda, scheduling awareness	Start here
Google Drive	Document access, file search, shared doc references	Start here
Notion	Database access, quick capture from phone, cross-platform sync	Great if you use it
Slack	Channel summaries, message search, catch up on threads	Great if you use it
GitHub / Linear	Issue tracking, PR reviews, project management	If you're technical

More connectors are being added all the time

This isn't the full list — CoWork regularly adds new connectors. Check the settings page to see what's currently available. If you use a tool and it's not listed, it might show up soon.

After Connecting: Verify With Claude

Once you've connected your tools in settings, open a new CoWork session and paste the prompt below. Claude will verify everything is working and update your system files.

Customize step 2 for your connectors

The prompt below includes example verification tasks for Gmail, Calendar, Drive, and Notion. Edit that part to match whatever you actually connected. The idea is simple: give Claude a small task that would only work if the connector is live. Connected Slack? Ask it to pull recent messages from a channel. Connected Spotify? Ask it what's currently playing. Any quick task that proves the connection works.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. I just connected some tools through CoWork settings. Let's verify everything is working and update my system. 1. Check which connectors are active — can you access my Gmail? Google Calendar? Google Drive? Notion? Any others? 2. For each connected tool, do a quick test: - Gmail: Pull my 3 most recent emails and show me the subjects - Calendar: Show me what's on my calendar this week - Drive: Search for a recent file I've worked on - Notion: List my workspaces or a recent page - Any others: Do a similar quick check 3. Tell me which ones worked and which ones didn't 4. Update the "Connected Tools" section in CLAUDE.md to list what's connected and what each one is used for 5. If any connections failed, walk me through fixing them Also ask me: are there any other tools or services I use daily that might be worth connecting? I might not think of everything on my own.

Bonus Level Up

Quick Capture Bridge click to expand

This is for people who want to send tasks to Claude when they're away from their computer — like from your phone during the day. It's totally optional, and your system works great without it. But if you want that "fire off a task on the subway and come home to finished work" workflow, this is how.

Important: your laptop needs to stay open

For quick capture to work, Claude Desktop just needs to be open on your laptop — you don't have to be sitting there or watching it. Leave it running in the background, go about your day, and send tasks from your phone whenever. Claude handles them automatically. The only thing that breaks it is if your laptop is closed or Claude Desktop is quit entirely.

There are a few ways to set up quick capture. Pick the one that fits your setup best:

How this works: Click one of the three options below. A ready-to-paste prompt will appear underneath it. Copy that prompt, open a new CoWork session, and paste it in. Claude will walk you through the rest.

Recommended

Dispatch

Send tasks straight from your phone to Claude. Built into CoWork — everyone gets it.

If you use Notion

Notion Hub

Drop notes in a Notion page from your phone. Claude checks it each session.

iPhone users

Apple Notes

Jot notes on your phone. Claude reads them via the Apple Notes connector.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. I want to set up Dispatch so I can send tasks to you from my phone. Walk me through the setup: 1. How do I find and pair Dispatch? (I need step-by-step instructions — I haven't done this before.) 2. Once paired, what kinds of things can I send? (Tasks, research requests, quick notes, links?) 3. What are the limitations I should know about? (Does my laptop need to be open? How fast do tasks get picked up?) After we get it working: - Add a "Dispatch" section to CLAUDE.md explaining that I use it to send tasks from my phone - Include any tips for writing good Dispatch messages (so Claude understands what I want) - Update CLAUDE.md with the setup Also ask me: what kinds of things would I most likely send from my phone during the day? That will help you give me better advice on how to use it.

Dispatch is built into CoWork — no extra apps needed. Just pair your phone with a QR code and start sending tasks.

These aren't the only options — any app that Claude has a connector for can work as a capture tool. The key idea is: drop something in from your phone, and Claude processes it when you're back at your computer.

Part 7

Custom Skills

You've connected your tools — Claude can now read your email, check your calendar, and search your files. Now let's teach it HOW to do specific tasks the way you want them done.

What's a skill?

A skill is a set of instructions that lives in a file in your workspace. Think of it like a recipe card — it tells Claude exactly how to do a specific task: what steps to follow, what tone to use, which of your memory files to read, what format the output should be in, and what mistakes to avoid.

Skills take what Claude already knows about you — your voice, your preferences, your routine — and turn it into a repeatable workflow. Like the difference between telling a new hire "figure out how to write emails" vs. handing them a style guide and three examples.

You don't write the skill yourself — Claude builds it for you based on a conversation. You just answer questions about how you want things done, review the draft, and approve it.

Skills are powerful even without connectors

You don't need Gmail, Calendar, or any connected tools for skills to work. Claude already knows your writing voice, your preferences, your career context, and your daily routine from your memory files — that's enough to draft emails, do research, plan trips, and more. Connected tools make skills even more powerful, but a skill without connectors is still far better than no skill at all.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will interview you to figure out which skill to build first, then build it with you.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. I want to build my first custom skill. Before we start, interview me to figure out which skill to build first: 1. What tasks do I do repeatedly that follow a similar pattern each time? 2. What tasks do you currently do "okay" but not great for me? 3. What tasks do I avoid because they're tedious? 4. If I could snap my fingers and have one thing done perfectly every time, what would it be? Based on my answers, recommend which skill to build first (pick the one with the highest impact). Then build it with me: - Ask me clarifying questions about how I want the task done (format, tone, what to include/exclude, what "good" looks like) - Create the SKILL.md file with all required sections: frontmatter (name + description + triggers + "do NOT use for"), context, pre-flight questions, execution steps, output format template, key memory file references, self-improvement loop, common mistakes, and one example - Save it to 03-resources/skills/[skill-name]/SKILL.md - Update CLAUDE.md with the new skill listed under Custom Skills Show me the draft before saving. I want to review it. Ask me anything else about how I want this task done — the more detail you get from me now, the better the skill will perform every time it runs.

Start with ONE skill. Get it right, then add more. Don't try to build 5 at once.

Already know what you want to build? Hit "Copy" on any idea below — it copies a complete, ready-to-paste prompt with all the instructions Claude needs to build the skill with you.

Communication

Email Drafter

Writes emails in your actual voice — not generic AI voice. Reads your writing samples and knows who you're writing to (boss, colleague, friend, cold outreach) so it adjusts tone automatically.

I want to build an Email Drafter skill. It should write emails in my real voice (read my writing-voice.md for samples). It needs to handle different contexts: writing to my boss, colleagues, friends, and cold outreach — each with a different tone. It should read people.md to know who I'm writing to. Ask me who the email is for and the key points I want to hit, then draft it. Always let me review before sending.

Productivity

Inbox Triage

Scans your email and sorts it into tiers: needs a response, FYI only, and skip. Offers to draft quick replies for the urgent ones. Prioritizes speed over perfection — the goal is inbox zero, not inbox art.

I want to build an Inbox Triage skill. It should scan my recent emails (via Gmail connector) and sort them into three tiers: (1) needs a response from me, (2) FYI / read later, (3) skip / archive. For tier 1, offer to draft quick replies in my voice. Prioritize speed — I'd rather get through 20 emails in 10 minutes than craft 3 perfect responses. Show me the triage results as a simple list I can act on.

Research

Research Brief

Deep research on a topic, company, or person. Gives you a structured brief with key findings, what it means for you, and clear next steps — not a wall of text you have to parse yourself.

I want to build a Research Brief skill. When I give you a topic, company, or person to research, produce a structured brief: key facts, recent news, why it matters to me (check my career-profile.md and current projects for context), and specific next steps I can take. Cite your sources with dates. Keep it concise — I want to be informed in 5 minutes, not read a thesis.

Planning

Trip Planner

Plans trips that actually match how you travel — your airline status, dietary restrictions, fitness routine, and budget. No generic "top 10 restaurants" lists. Real recommendations based on what you like.

I want to build a Trip Planner skill. It should read my travel-preferences.md and food-preferences.md before planning anything. I want it to handle flights (using my airline status and preferences), hotels, restaurants (I'm picky — take my food preferences seriously), and daily itineraries. Ask me about the destination, dates, budget, and trip vibe before building the plan. Include a packing list and fitness options.

Review

Weekly Review

End-of-week checkpoint that actually keeps you on track. Summarizes what got done, flags what's stalled, refreshes your memory files, and sets up next week — so Monday-you doesn't start from scratch.

I want to build a Weekly Review skill. Every Friday, it should: (1) summarize what I accomplished this week, (2) flag stalled tasks or projects, (3) review and suggest updates to my memory files based on what changed, (4) do a system health check on my skills and automations, and (5) set priorities for next week. Read my TASKS.md, session-log.md, and project folders to do this. Keep the output scannable — I want to review my week in 5 minutes.

Career

Interview Prep

Company-specific cheat sheet with a 60-second pitch, STAR stories mapped to their values, smart questions to ask, and landmines to avoid. Then runs a mock interview with coaching feedback.

I want to build an Interview Prep skill. Given a company name, it should: (1) research the company (or use an existing research brief if I have one), (2) build a one-page cheat sheet — 60-second pitch, 3 STAR stories tailored to their values, talking points, questions to ask, and landmines to avoid, (3) read my career-profile.md to map my experience to what they care about. After the cheat sheet, offer to run a mock interview with realistic questions and coaching feedback.

Where do skills live?

Claude saves your skills to 03-resources/skills/ in your workspace folder. Each skill gets its own subfolder with a SKILL.md file inside. You don't need to create anything manually — Claude handles all of this when you run the prompt above.

Using your skills

Once a skill is built, you don't need to do anything special to activate it. Just talk to Claude like you normally would — the trigger phrases built into the skill file handle the rest. For example:

You: "Draft an email to Nancy about the Q2 budget update"

You: "Triage my inbox"

You: "Research Anthropic for me"

You: "Prep me for my interview at Notion on Thursday"

Claude recognizes what you're asking, finds the right skill, and follows the process you defined — your voice, your format, your preferences. No special commands or syntax required.

Anatomy of a Skill File click to expand

You don't need to memorize any of this — Claude builds the entire skill file for you. This is just so you know what's inside if you ever want to peek under the hood or tweak something.

Required Frontmatter — name + description with trigger phrases and "do NOT use for" conditions

Required Context — why this skill exists and what problem it solves

Important Pre-flight questions — what Claude should ask before diving in

Required Execution steps — the step-by-step process

Required Output format — exact template for the output

Important Key references — which memory files to read

Important Self-improvement loop — how the skill learns from corrections

Optional Common mistakes — specific anti-patterns to avoid

Optional Example — sample input/output showing the skill in action

Level Up

The Self-Improvement Loop click to collapse

Level Up

Composable Skill Chains click to expand

Every skill has a built-in feedback mechanism — it learns from your corrections and gets better over time.

How it works

Before running: Claude checks past corrections. After running: it collects your feedback. When you correct something: (a) fixes the output, (b) logs the feedback, (c) proposes a skill update if there's a pattern.

After a few weeks, mediocre skills become excellent ones. The paste prompt already includes this — just know it's working in the background.

Part 8

Scheduled Tasks

You've built the knowledge base and the skills. Now let's put them on autopilot. Scheduled tasks are Claude sessions that run automatically — they do work while you're away so things are ready when you sit down.

What's a scheduled task?

A scheduled task is a CoWork session that runs on a timer without you having to start it. Think of it like setting a coffee maker the night before — you set it once, and every morning the work is already done when you show up.

For example: a morning briefing that summarizes your email and calendar before you wake up, or a weekly review that recaps your progress every Friday. You tell Claude what to do, when to do it, and where to save the results — then it just runs.

Start a new CoWork session for this section. Copy the prompt below and paste it in. Claude will figure out which automations make sense for your routine.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. Let's set up my scheduled automations. Before creating anything, ask me: 1. What time do I usually wake up / start my day? 2. When do I typically sit down to use Claude? (Morning, evening, weekends?) 3. Do I have email connected? (Gmail / Google Workspace) 4. Do I have a calendar connected? 5. What would be most useful to have done FOR me automatically — email summary, calendar prep, weekly review, industry news, something else? 6. Are there any recurring tasks I do on a schedule that could be automated? Based on my answers, recommend 2-3 scheduled tasks to start with (don't overwhelm me). For each one, explain what it will do and when it will run. Get my approval, then set them up. Important: Remind me that my laptop needs to be awake and Claude Desktop needs to be open for these to run. Ask me about anything else in my routine or workflow that could benefit from automation — I might not think of everything on my own.

Keep your laptop awake

Scheduled tasks only run when your computer is awake and Claude Desktop is open. If your laptop is closed or Claude isn't running, the task misses its window. The easiest fix: just leave Claude Desktop open in the background and let your laptop sleep on its own schedule.

Where do scheduled tasks live?

Scheduled tasks are managed by CoWork itself — you can see and edit them in Settings → Scheduled Tasks. Each task has a name, a schedule (time + days), and a prompt that tells Claude what to do. The output gets saved wherever the prompt specifies.

Automations to Consider

Daily

Morning Briefing

Summarizes your email, walks through today's calendar with context on each meeting and who you're meeting with, flags priorities, and surfaces any relevant news. Runs every weekday, 30-60 minutes before you start work.

Weekly

Weekly Review

End-of-week checkpoint: what got done, what's stalled, project status updates, memory file refresh suggestions, system health check on your skills and automations, and priorities for next week.

Daily

Inbox Digest

Scans your email and produces a prioritized digest timed for when you usually sit down with Claude. Groups emails into tiers so you can batch-process responses instead of scrolling through everything.

Weekly

Industry Tracker

Tracks news, job postings, and company updates in your target industry or area of interest. Runs weekly so you stay informed without having to manually check dozens of sources.

A screenshot from my daily morning briefing

Real example: Newsletter automation

I built an automation that reads all my newsletters, summarizes them, and delivers a single digest so I never have to scroll through a cluttered inbox again. I wrote about how it works: How I Got an AI to Read My Newsletters For Me.

Level Up

Making Automations Reliable click to expand

Your first automations will probably need some tuning. Here's how to get them running smoothly:

Be specific in the prompt. Don't just say "check my email." Tell it exactly what to scan, how to prioritize, what format, and where to save the results.
Point to your skills. If you've built an Inbox Triage skill, tell the automation to use it — that way it follows the same rules every time.
Check the output for the first few days. Automations need tuning early on. Review what they produce, correct mistakes, and they'll improve quickly.

Part 9

Ongoing Maintenance

You've built everything. Now let's make sure it stays useful. A system is only as good as its maintenance — here's the rhythm that keeps it sharp.

What does maintenance actually mean?

This isn't busywork. Maintenance just means Claude gets better over time instead of staying the same. Most of it happens automatically — when you correct a mistake, Claude logs it and learns. When a scheduled task runs, it checks for updates.

Your only real job is to correct Claude when it gets something wrong, and do a quick weekly check-in to make sure nothing's stalled. That's it.

Your Maintenance Rhythm

Every Session

Happens Automatically

Claude reads session-log.md to pick up where you left off, checks TASKS.md for what's on your plate, and scans quick capture for anything you dropped in during the day. When you correct a mistake, Claude logs it and learns. You don't need to do anything special — just use the system.

Weekly

5-Minute Check-In

Once a week (or set up a scheduled task to do it for you): review what got done, flag anything that's stalled, refresh memory files if something changed in your life, check if your skills and automations are working well. This is your system tune-up.

Monthly

Quick Audit

Skim CLAUDE.md — is it still accurate? Archive any completed projects. Check if there's a manual workflow you keep repeating that should become a skill. This takes 10 minutes and keeps the system from getting stale.

The Corrections Loop

This is the single most important part of maintenance. Every time Claude gets something wrong, just tell it. Claude will: (1) fix the mistake, (2) log it in CLAUDE.md with the date so it doesn't happen again, (3) check if the mistake reveals something new about you and update the relevant memory file, and (4) check if a skill needs updating and propose changes.

You don't need to manage this process — just correct Claude when it's wrong and it handles the rest. Over time, mistakes become rare because Claude is learning from every correction.

Your Graduation Step

Once you've been using the system for a few sessions and things feel solid, create a personal reference doc that summarizes how everything works. This is the doc you'd reread if you ever forget how your system is set up.

Start a new CoWork session for this section. Copy the prompt below and paste it in.

First, read these files in my workspace: discovery-notes.md, GLOBAL-INSTRUCTIONS.md, and CLAUDE.md. Those have everything from my previous sessions. We've built the whole system. Now create my GETTING-STARTED.md — a personal reference doc that explains how everything works. It should include: 1. "How It Works" (30-second version) — Context + Instructions + Automations 2. "Setup Checklist" — What's connected, what's configured 3. "What's In Here" — My folder structure, memory files, skills, automations 4. "How to Use This Day-to-Day" — Quick commands, example prompts I can use 5. "When Claude Gets Something Wrong" — How the correction loop works 6. "Weekly Rhythm" — What happens each day of the week 7. "Pro Tips" — The power-user stuff I should know 8. "Things This System Can't Do (Yet)" — Honest limitations Write it in my voice (reference writing-voice.md). Save to GETTING-STARTED.md at the root of my workspace. This is the doc I'd reread if I forget how the system works. If there's anything else you think should be in this reference doc based on everything you've learned about me, add it — and ask me about anything you're unsure of.

Setup Checklist

Your Step-by-Step Setup Flow

Session 1
Discovery. Have the discovery conversation with Claude. Don't build anything yet. Just talk. (Part 0)
Session 2
Global Instructions + CLAUDE.md. Set up how Claude behaves, then create the master context file it reads every session. (Parts 1-2)
Session 3
Memory Files. Build writing-voice.md first (with real samples), then people.md, life-areas.md, glossary.md. (Part 3)
Session 3-4
Additional Memory + Folder Structure. Build career-profile.md and any optional memory files, then set up your PARA folder structure. (Parts 3-4)
Session 4
Session Continuity. Set up session-log.md and TASKS.md so Claude picks up where it left off. (Part 5)
Session 5
Connectors + Quick Capture. Connect your tools (Gmail, Calendar, etc.) and set up a way to send tasks from your phone. (Part 6)
Session 5-6
Custom Skills. Build your first 1-2 custom skills for your most common tasks. (Part 7)
Session 6-7
Scheduled Tasks. Set up automations that run while you're away — morning briefing, weekly review. (Part 8)
Session 7
Maintenance + Getting Started guide. Learn the maintenance rhythm, then write your personal GETTING-STARTED.md reference doc. (Part 9)
Ongoing
Refine. Use the system. Correct mistakes. Let it learn. It gets better every session.

You made it to the end. I'm impressed.

I'm Chloe — I built this guide while learning AI out loud. If you found it useful, here's where to find more of whatever this is.

@chl0key Substack

Get the AI Mega Daily Digest

Want my daily newsletter? Send an email to keywellchloe@gmail.com with the subject line "AI Mega Daily Digest" — no body text needed — and I'll get you on the list.