Upgrade Your
AI Second Brain

You are upgrading my second brain (this folder) with a self-wiring knowledge graph.

Right now my notes are a pile of files. I want every person, company, and project I've ever mentioned to become its own page, automatically linked to every note that mentions it, so my brain knows how my whole world connects without me organizing anything.

Do this in four passes. Show me a short summary after each pass before moving on.

PASS 1 — Build the entity folders.
Create these if they don't already exist:
- entities/people/
- entities/companies/
- entities/projects/

PASS 2 — Extract every entity.
Read every note in this vault. Skip the entities/ folder itself, skip templates, skip anything in an archive folder. Pull out every:
- Person I work with, mention, or track (not public figures named once in passing)
- Company or organization
- Project or initiative
Build one list. Merge obvious duplicates ("Garry Tan" and "Gary Tan", "Acme" and "Acme Inc"). Show me the list grouped by type and WAIT for my OK before writing anything.

PASS 3 — Create the entity pages.
For each entity, create a page in the right folder, named after it (e.g. entities/people/garry-tan.md). Every page gets this exact structure:

---
type: person            # or company / project
aliases: []             # other names and spellings, so links still resolve
status: active
created: {today}
---

# [Name]

**What:** one line — who or what this is, in MY world specifically.
**Key facts:** 3 to 6 bullets, pulled only from my notes. No outside info, no guessing.
**Open threads:** anything unresolved — a promise, a next step, an unanswered question.

## Mentions
(leave this heading empty — Obsidian fills it automatically with backlinks)

Only use facts that appear in my notes. If a field has nothing, write "unknown." Never invent.

PASS 4 — Wire the links.
Go back through my notes and wrap the first mention of each entity in each note in a [[wikilink]] pointing to its page (e.g. [[garry-tan]]). Don't touch quotes or code blocks. Don't link the same entity twice in one note. When you're done, tell me how many notes you edited and how many entity pages you created.

Then I'll open Obsidian's Graph View (the graph icon, top-left) and see my whole world wired together.

LAST STEP — make it permanent.
Add this block to my CLAUDE.md (create the file if it doesn't exist) so every future note wires itself with zero effort from me:

## Self-wiring graph (standing rule)
Whenever I add or edit a note:
1. Scan it for people, companies, and projects.
2. For any without a page in entities/, create one using the template above.
3. Wrap their first mention in this note in a [[wikilink]].
4. If a note conflicts with an entity page, flag it — never silently overwrite.
Never invent facts. My notes are the only source of truth.

This is my brain's nightly "dream cycle." Run it once a day, on a schedule, with nobody watching.

Your job: make my second brain a little smarter and cleaner than it was yesterday, then leave me a short note about what you found. Work through these six jobs in order. Make the changes directly. Keep a running list of what you did.

1. MERGE DUPLICATES.
Look through entities/ for pages that are the same thing under different names or spellings ("garry-tan" and "gary-tan", "acme" and "acme-inc"). Merge each set into one page, move the extra names into the aliases: field so old links still resolve, and repoint any links aimed at the dead page.

2. FIX BROKEN LINKS.
Find every [[wikilink]] pointing to a page that doesn't exist. Either create the missing page (if it's clearly a real entity) or fix the link to point at the right one. List anything you couldn't resolve.

3. CONNECT NEW DOTS.
Read notes added or changed in the last day. Wire any new people, companies, or projects into the graph — new pages plus [[wikilinks]] — using the same rules as the rest of the vault.

4. FIND CONTRADICTIONS.
Scan for places where two notes disagree on a fact: a date, a number, a decision, a name. Do NOT fix them — you don't know which is right. Collect them for my morning brief so I can settle them.

5. SCORE WHAT MATTERS.
Look at what's active right now — recent notes, open threads on entity pages, projects marked active. Pick the 3 to 5 things that actually matter for tomorrow. Ignore the noise.

6. WRITE THE MORNING BRIEF.
Create a file at brief/{tomorrow's date}.md with:
- Top 3 to 5 for tomorrow — what matters and why, each linked to its note or entity.
- Open threads — promises and next steps going stale.
- Contradictions to settle — the conflicts you found in job 4.
- What I cleaned up — one line on the merges and links you fixed, so I trust the system is working.
Keep it scannable. I should read it in under two minutes with my coffee.

Never invent facts. If you're unsure, say so in the brief instead of guessing.

I don't have a second brain yet. Build me one from scratch in this folder, then we'll wire it up.

1. Create this structure:
   raw/                  — where I dump everything, unorganized
   wiki/                 — where you organize what I dump
   entities/people/
   entities/companies/
   entities/projects/
   brief/                — your daily morning briefs land here
   CLAUDE.md             — your operating instructions
   USER.md               — who I am and how I work

2. Write CLAUDE.md so it says: read everything before answering, organize raw/ into wiki/, never invent facts, and follow the self-wiring graph and nightly dream-cycle rules I'm going to paste in next.

3. Interview me. Ask me 5 to 7 questions about who I am, what I do, the main people, companies, and projects in my world, and what I want this brain to help me with. Write my answers into USER.md.

When that's done, tell me to drop my first 10 to 30 files into raw/ and run the self-wiring graph prompt.