Profile and memory files
CLAUDE.md tells the assistant how to operate. USER.md tells it who you are, your goals, and the people who matter. /setup writes both to the project root.
Technical DIY path
Use this checklist if you want the file-based Claude Code version of the assistant. It is the more technical path: you install local tools, unzip a resource pack, and keep the assistant's memory in plain files you can inspect.
About 30 to 60 minutes from a clean machine. One /setup command writes your profile and schedules the morning run. Optional phone notifications and a two-way phone channel are extras you can add later.
What you are building
The Claude Code version lives in a folder on your computer. That folder holds the assistant's operating instructions, your profile, local agenda files, saved skills, and the reports it writes. You can open every file and see what changed.
CLAUDE.md tells the assistant how to operate. USER.md tells it who you are, your goals, and the people who matter. /setup writes both to the project root.
now.md is the running cockpit and decision-log.md the durable record, both in the project root (no nested folders). /setup creates them on first run.
Setup, morning brief, meeting prep, email triage, capture, and skill-builder ship under .claude/skills/ and load automatically.
Overlap and difference
Cowork and Claude Code share the assistant pattern: profile, memory, connectors, repeatable skills, and a strict read/draft/approve boundary. The difference is where it lives and how much you touch the files.
| What matters | Cowork DIY | Claude Code (this guide) |
|---|---|---|
| Setup | Install a plugin, run the setup skill, let Cowork create the live Daily Brief artifact. | Unzip the pack, open it as a project, run /setup (the same skill, adapted), then start a fresh chat so the profile loads. |
| Memory | Project files plus a live artifact for the day-to-day view. | Plain local files in the project root: CLAUDE.md, USER.md, now.md, and decision-log.md. |
| Daily Brief | A live Cowork artifact that refreshes on open. | An HTML + text file in outputs/ the scheduled task rewrites each weekday. You can open, diff, and back it up. |
| Daily run | A scheduled task runs inside Cowork. | /setup creates a weekday scheduled task (Routine) for you. It runs while Claude Desktop is open and your computer is awake. |
| Best fit | The easier daily driver for non-technical DIY users. | People who want file-level control, a terminal, and optional phone channels Cowork does not offer. |
Setup steps
Steps 1 through 12 get you a working assistant with an automatic weekday brief, including optional one-way phone notifications. The optional two-way phone channel at the end is an extra. Each step expands.
Claude Desktop supports macOS 11 or newer and Windows 10 or newer. If you are on a locked-down work laptop, install early in case your IT policy blocks new apps.
Use the same Claude account on mobile so your assistant is reachable from your phone.
Claude Mobile is optional for the technical build, but useful when you want to check in from your phone.
Have a Google account ready, but do not connect sensitive mail or calendar data until you understand the permissions.
If your primary email and calendar are Google, then you are done. Keep that login available for the assistant setup.
If you use iCloud, Outlook, Yahoo, or another provider, create a free Google account at accounts.google.com/signup. Use it for starter exercises or for safe forwarding tests. You do not need to migrate your life into Google.
If your email or calendar lives in iCloud, Outlook, or Yahoo, keep the provider setup reference handy for provider-specific forwarding or calendar steps.
Safety note: Do not forward a sensitive inbox or publish a sensitive calendar just to test. Use sample events or a dedicated Google account if you are unsure.
A small tool Claude Code uses to work with local files and project history. Quick on both Mac and Windows.
Cmd+Space, type Terminal, hit Enter).xcode-select --installNot sure if you already have Git? That is fine. Running the step again does no harm.
Grab the resource pack and unzip it into the folder where you want your assistant to live.
.zip file. On Windows, right-click it and choose Extract All.The unzipped folder is named Claude-Code-Assistant-Pack. It ships with the six skills under .claude/skills/ (including setup), optional Telegram helpers in a telegram/ folder, a sample brief, and a README. Your profile and now.md start empty: /setup creates them in the project root on first run.
Optional: show hidden files. The pack includes a .claude folder, and folders whose names start with a dot are hidden by default. You do not need to see it for the assistant to work, but if you want to browse the skills yourself, turn on hidden files:
Claude-Code-Assistant-Pack folder and press Cmd + Shift + . (period). Hidden files appear, dimmed. Press the same keys again to hide them.Claude Code needs to work inside the folder so the packaged skills and local files stay together.
Claude-Code-Assistant-Pack folder. If Claude asks whether you trust this folder, approve it. You downloaded it from this page and it needs local file access to work./ in the prompt box; you should see setup and the other skills listed. They load automatically from .claude/skills/ because you opened this folder.Folder boundary: keep the assistant's work inside this project. The pack is designed so .claude/skills/ and your profile files in the project root stay together.
These connectors let meeting prep and morning brief work on your real calendar and inbox.
If your main mail or calendar is iCloud, Outlook, Yahoo, or another provider, use the provider setup reference to forward mail or share calendar context into Google where needed.
One-way only: your own brief, texted to you. Set this up before /setup so the morning task wires it in. Skip it and the brief still lands in outputs/.
This is notifications, not control: it sends your own brief to you and nothing else. It does not read your texts or take commands. Want to text your assistant and have it act? That is the two-way channel in the last step.
/setup asks for your own number or Apple ID email and wires the morning task to text you the brief.telegram/setup-telegram.sh once. It stores your bot token and finds your chat id. See telegram/telegram-setup.md for the two-minute walkthrough.telegram/send-telegram.sh.Boundary: both are one-way and outbound to you only. They send your own brief to you. They do not read inbound texts, take commands, or let the assistant message anyone else.
/setupOne command writes your profile, confirms your connectors, and creates the weekday morning scheduled task.
CLAUDE.md, USER.md, now.md, and decision-log.md in the project root.daily-brief-refresh that runs meeting prep, email triage (drafts only), and your morning brief, and writes the brief to outputs/.If your Claude Code version cannot create the task automatically, setup prints the exact Routines steps and a ready-to-paste task prompt instead, so you are never stuck.
Start a fresh chat after setup. Your new CLAUDE.md loads at the start of the next chat in this folder, not halfway through the current one.
This is the one click that keeps the morning brief automatic.
daily-brief-refresh and click Run now.If you pick "allow once," the morning run will quietly stall the next day waiting for you. The pack ships a .claude/settings.json that pre-authorizes the brief-delivery tools, so only the connector reads need this one-time approval.
Constraint: a local scheduled task only runs while Claude Code Desktop is open and your computer is awake. Enable Keep computer awake in Settings if you want it to fire reliably at 8:00 AM.
Read the file the task just wrote, then exercise each skill once.
The run from step 11 wrote your first brief to outputs/YYYY-MM-DD-daily-brief.html and .txt. Open the HTML file to read your day. From here on, the task refreshes it every weekday morning, and you can run any skill by hand whenever you want:
These workflows write local files and draft outputs inside the project. The Daily Brief is the file in outputs/, not a live artifact. At this point you have a working assistant.
Beyond notifications. Message your assistant from your phone and it works on your real files, then replies in the same chat. Uses Claude Code Channels (research preview).
This is different from Step 9. Step 9 sends your brief to your phone one way. Channels is two-way: you text your assistant a request from iMessage or Telegram, it runs in your open Claude Code session against your real files, and the reply comes back in the same chat. Messages only arrive while a Claude Code session is open, so keep one running (or a background session) when you want to reach it.
bun --version). On Team or Enterprise plans an admin must enable channels first.
/newbot, and copy the token. (If you already made one for notifications in Step 9, you can reuse that token.)Two-way means your assistant can receive requests from your phone, not act without you. It still reads, drafts, and reports, and still waits for your approval before it sends email, posts, deletes, buys, or changes a calendar event. Only allowlisted senders can reach your session (texting yourself on iMessage is auto-trusted), and anyone who can message you can approve tool use, so keep the allowlist to you. Channels is a research preview, so the exact commands may change; the official steps live in the Claude Code Channels docs.
Daily loop
The workshop deck's core loop still applies here: morning brief opens the day, capture keeps the state current, meeting prep adds context before important conversations, and email triage drafts replies for your review.
The weekday scheduled task runs morning-brief for you and writes the brief to outputs/. Run it by hand anytime too. It reads your profile, agenda, calendar, and inbox, then updates now.md.
Use capture whenever something changes: a decision, a follow-up, a person update, or a lasting preference.
Use meeting-prep before meetings and email-triage when you want Gmail drafts. You still approve the final send.
Saved skills
These are the same skills as the Cowork version, adapted for Claude Code. They live in .claude/skills/ inside the folder you open as a project and load automatically.
The single setup path: interviews you, writes your profile to the project root, confirms connectors, and creates the weekday scheduled task. Re-run it to refresh your profile.
/setup
Builds a mobile-readable brief covering meetings, threads needing attention, first move, goal progress, waiting items, and watch-outs.
Give me my morning brief for today.
Finds upcoming meetings, researches attendees, and writes linked meeting, contact, and company/org briefs for the people and orgs in the room.
Prep me for tomorrow's meetings.
Records decisions, follow-ups, reminders, meeting notes, contact changes, durable facts, and lasting assistant preferences.
Capture this update: [what changed].
Reviews recent Gmail, classifies threads, creates Gmail drafts when available, and prints the draft text in chat. Nothing is sent.
Triage my last 48 hours of Gmail.
Turns a repeatable workflow into a new skill, or edits an existing skill, while keeping external actions approval-gated.
Make this a reusable skill.
Safety boundary
The Claude Code resource pack keeps the workshop rule: the assistant can read connected sources and draft local files, reports, and Gmail replies, but anything that leaves your control waits for you.
| Read | Inbox, calendar, profile files, agenda files, meeting notes, contact notes, and connected context you approve. |
|---|---|
| Draft | Morning briefs, meeting prep docs, contact briefs, local agenda updates, and Gmail drafts for you to review. |
| Approve | You send emails, confirm calendar changes, approve external messages, and decide before anything irreversible happens. |
If the local setup is more than you want to manage, the no-terminal Cowork path covers the same ground. If you would rather build it live with help in the room, join the guided class waitlist.