Identity · Outputs · Templates
The 3-folder setup
Three folders Claude reads every session so you stop re-explaining yourself
Browser Claude forgets every chat. Cowork reads files from disk every time you open it. You set up three folders (who you are, what you ship, what you reuse), drop in three ABOUT ME files, and point Global Instructions at them once. From then on every Cowork session starts with full context. No briefing prompt at the top of every conversation.
Worth knowing:Once your three files are in place, you stop pasting context into prompts. Most prompts get shorter. Claude's first reply is closer to what you actually wanted, every time.
Browser-Claude users tired of re-briefingAnyone who shares Claude with a teammateCreators with a consistent voice or POV
Start Block 1 →One worked workflow, end to end
Cowork Projects
Scoped memory that remembers within a project and forgets across boundaries
A Cowork Project is a workspace with its own files, its own context, and its own memory of what you discussed. Different from Browser Projects (which sync across devices), Cowork Projects live on your machine and execute work locally. This lesson walks you through one workflow end to end (a recurring newsletter), then names the other three you can adapt the same pattern to.
Worth knowing:Cowork Projects forget across boundaries on purpose. The newsletter project doesn't accidentally read your client research. Different work, different memory.
Anyone with recurring work (newsletters, reports, weekly reviews)Freelancers juggling 3+ clients in ClaudePeople burned by context bleed in long chats
Start Block 2 →Build once, run anywhere
Skills + Scheduled Tasks
Reusable artifacts that run when you call them or on a clock
A Skill is a packaged prompt + instructions Claude can call by name (or auto-fire from natural language). You build one with Claude itself (no code), test it, then it shows up across every Cowork session. Scheduled Tasks run a Skill on a clock. Your weekly report at 8am Monday, your competitor scan at midnight Sunday. This lesson builds one of each.
Worth knowing:The Skill Creator is itself a Skill. You describe what you want in plain language, it interviews you, then writes the Skill. Building Skills is the easiest part once the first one exists.
Anyone with a prompt they paste more than weeklyOperators with a repeatable workflowAnyone who wants Claude doing work overnight
Start Block 3 →Build your text-file twin
Voice extraction
A voice file that lets Claude write in your voice, not the default
Default Claude writes like Default Claude. A voice file fixes that. You run a long structured interview with yourself (taste, references, what you would never say, what you say all the time), Claude compiles it into a portable voice document, and that file goes into your identity folder. Every piece of writing Claude produces from that point reads like you wrote it. Includes the security note Ruben skipped: where NOT to store this file.
Worth knowing:The voice file is also a hiring document, a brand brief, and a positioning statement in disguise. Once it exists, you reuse it for everything that needs to sound like you.
Writers, founders, and anyone with a public voiceAnyone whose Claude output sounds genericSolo operators whose voice IS the product
Start Block 4 →Obsidian · plain text · no terminal
Your context-file editor
Read and edit your Cowork files without ever looking at raw markdown
Your Cowork folder runs on plain text files. The problem: they look terrible in a standard file viewer, so you stop updating them and they drift. Obsidian fixes this. Point it at your Cowork folder and every file renders as clean, formatted notes. Edit a line, save, and the change is live in the next session. This lesson also covers the SKILLS subfolder pattern (one note per Skill, searchable and maintainable) and the mid-session update pattern: spot something you want Claude to stop doing forever, update the file in Obsidian, open a fresh session, and the correction holds permanently.
Worth knowing:The mid-session update pattern turns a one-time correction into a permanent rule. You stop re-explaining the same preference.
Anyone whose Cowork context files are out of datePeople who built Skills in Lesson 3 and want to maintain themAnyone who avoids editing files because markdown looks messy
Start Block 5 →Gmail · Calendar · live artifact
Live admin tracker
A tracker that re-scans your inbox and updates itself each run
One-shot summaries disappear. A live artifact is a file in your OUTPUTS folder that Claude re-reads and updates each time you run it. This lesson covers the full loop: grant Gmail and Calendar permissions once (the step most people skip), build an invoice or travel tracker that reads your live email data, correct the output twice, then say 'now rewrite my original prompt with all corrections and save it as a Skill.' Three correction cycles become one polished, reusable Skill, without writing the Skill body yourself.
Worth knowing:The 'rewrite my prompt as a Skill' shortcut turns a session you already ran into a Skill you never have to rebuild.
Anyone who runs the same report or inbox scan weeklyFreelancers tracking invoices or client communicationsAnyone who wants Gmail data inside a Cowork Skill
Start Block 6 →Blotato · Canva · posts-log.md
Wire a connector
Let Claude draft and queue posts without opening any app
Without a connector, Claude is a consultant: you bring it information, it gives advice, and you execute elsewhere. With an MCP connector, Claude is staff: it reads from your tools, writes to them, and queues posts for your review before anything publishes. This lesson covers installing Blotato or Canva, running a discovery prompt to see what the connector actually exposes, building a write-content Skill through an interview until Claude is 95% confident about your brand voice, and setting up the posts-log.md safety gate so nothing publishes until you say go.
Worth knowing:The discovery prompt shows what a connector actually exposes in your session. It is almost always different from what the marketing page claims.
Creators who post to multiple platforms each weekAnyone building a content workflow in CoworkPeople who finished Lesson 3 and want to publish, not just draft
Start Block 7 →Folders · Inventory · Privacy · Output
Safe desktop file work
Start with one folder and prove what Claude can read
This block teaches the safe start for local files: choose one task folder, inventory sources before output, separate private or stale files, and choose the exact output shape before Claude writes.
Worth knowing:The first useful answer is not the final output. It is the source inventory that proves what Claude is about to use.
Anyone using Claude with local filesPeople with messy project foldersTeams handling sensitive source material
Start Block 8 →Summary · Rewrite · Comparison · Proof
Sourced file outputs
Turn local files into work you can check
You turn approved files into a sourced summary, rewrite, comparison, or handoff packet, then check claims against file names so the output can be trusted and corrected.
Worth knowing:A good file answer points back to the source. If it cannot name the file, it is not ready to trust.
Operators summarizing project foldersFreelancers preparing client handoffsAnyone comparing drafts or exports
Start Block 9 →Versioning · Memory · Handoffs · Cleanup
Local file habits
Keep the folder usable after the first run
You set version rules, decide what Claude should remember, package the work for another person, and create a folder maintenance checklist so local file work stays clean after the first success.
Worth knowing:The folder rule is living documentation. It keeps the next run from rediscovering the same mistakes.
People returning to the same folders weeklyTeams handing files to another personAnyone building repeatable desktop workflows
Start Block 10 →