Set It Up Today (Claude Code Walkthrough)

What You Need Before Starting

Prerequisites are minimal by design:

• Claude Code (v2.1.72 or later for scheduling features). This is the only required tool.
• A working directory — any folder on your machine where the knowledge base will live.
• Obsidian (optional but highly recommended). It lets you browse wiki pages in real time with a graph view as Claude writes them. Free at obsidian.md.

No databases, no plugins, no code. Just Claude Code and a folder.

Step 1: Create the Folder Structure

Open Claude Code in your working directory and ask Claude to set up the structure:

Set up my second brain knowledge base. Create these folders:

knowledge-base/raw/session-notes/
knowledge-base/raw/pages/
knowledge-base/wiki/
knowledge-base/outputs/

Then create these empty files:
knowledge-base/wiki/index.md (with placeholder category sections)
knowledge-base/wiki/log.md (append-only log, just the header)
knowledge-base/wiki/processed.md (empty registry)

Finally, create CLAUDE.md in this root folder as my operating manual
for a knowledge base about [YOUR TOPIC HERE].

Claude will create all the folders and seed the files. Review CLAUDE.md carefully — this is the most important step. Make sure your topic focus is specific, not generic.

Step 2: Review and Tailor Your CLAUDE.md

The CLAUDE.md Claude generates is a starting template. Edit it to add your specific topic focus and confirm these sections are present and correct:

• What this knowledge base is for (topic, purpose — be specific)
• The folder structure with the two hard rules called out explicitly
• All five operations with step-by-step instructions
• The log.md format with an example entry
• The processed.md format
• The self-improving rule stated explicitly: "Every ingest and every valuable query must update a page in raw/pages/ AND append to wiki/log.md"

Spend time on this. A well-written CLAUDE.md means Claude behaves correctly in future sessions without you having to re-explain the system every time.

Step 3: Schedule the Weekly Dream Sequence

The Dream Sequence is the health check that runs automatically: it ingests any new files in raw/, checks for contradictions and orphan pages, and logs a summary. You want it running on a schedule without you having to remember.

The quickest way during an active session:

/loop 7d read CLAUDE.md then run the dream sequence: ingest any new files in raw/ not yet in processed.md, check for contradictions and orphan pages in raw/pages/, log the results to wiki/log.md

This schedules the Dream Sequence to run every 7 days for the current session.

Important limitation: The /loop command is session-scoped — the task stops when you close Claude Code and expires after 7 days regardless. For a durable weekly routine that runs even when Claude Code is closed, use a Desktop Scheduled Task in Claude Code settings (Settings → Scheduled Tasks → New Task). Configure it to run weekly with the same dream sequence prompt. Desktop tasks run on your machine and persist across sessions.

Alternatively, create a .claude/loop.md file inside your knowledge-base directory with the dream sequence prompt as the content — this becomes the default behavior whenever you run a bare /loop command in that project.

Step 4: Your First Ingest

Find an article, paper, or resource you want to add. Paste the URL or text into Claude Code and say:

Add this to my second brain: [URL or paste text here]

Watch what Claude does. It should complete all six ingest steps:

1. Save the raw source untouched into raw/
2. Extract key points and identify relevant themes
3. Write or update the relevant page(s) in raw/pages/
4. Update wiki/index.md with links to any new or significantly updated pages
5. Append a line to wiki/log.md in the correct format
6. Record the source filename in wiki/processed.md

If Claude skips any step, point it out, ask it to complete the step, and then update CLAUDE.md to make that step more explicit. Early iteration on CLAUDE.md is normal and important — the first few ingests reveal what your operating manual is missing.

Step 5: Verify Everything Worked

After your first ingest, check four things:

• Does a new file exist in raw/pages/?
• Is wiki/index.md updated with a link to it?
• Does wiki/log.md have a new entry in the correct format?
• Is the source filename in wiki/processed.md?

Four for four means your second brain is working. Any miss means that step needs to be clarified in CLAUDE.md.