Build your own second brain.

I want you to set up a personal knowledge base in this folder and maintain it for
me from now on. Read all of this before you do anything.

## About this wiki

- What it is about: [THE SUBJECT, e.g. "AI news and my own product ideas", or
  "everything I read about my industry", or "general, no single topic"]
- What I will feed it: [WHAT YOU WILL DROP IN, e.g. "articles, call transcripts,
  meeting notes and the odd PDF"]
- What I want out of it: [WHAT IT IS FOR, e.g. "answering questions across
  everything I have read, and raw material for writing"]
- How I write: [YOUR VOICE, e.g. "British English, plain words, short sentences,
  no hype"]
- Call me: [YOUR NAME]

## Step 1: create the folders

Create this structure here, with a .gitkeep in any empty folder:

  raw/            original sources, never edited after they are added
  raw/assets/     images and attachments belonging to sources in raw/
  sources/        one page per thing I have read
  entities/       people, organisations, products, places
  concepts/       ideas, themes and topics that recur
  synthesis/      overviews, comparisons and answers worth keeping
  index.md        a catalogue of every page
  log.md          append only, chronological
  CLAUDE.md       the rules you read at the start of every session

## Step 2: write the rules file

Write CLAUDE.md, substituting my answers above where relevant. This file is the
whole trick: you read it at the start of every session, so the conventions survive
even though our conversation does not. It must say:

- What this wiki is about, and that unrelated clusters are fine. Do not invent a
  thesis to tie my reading together.
- That raw/ is immutable. Never edit a file there once it exists. If a source is a
  URL, still save a stub in raw/ with the URL, title and date fetched, because
  links rot.
- What each folder is for, as above.
- That every page outside raw/ starts with frontmatter: title, type, tags, created,
  updated, and the sources it draws on.
- That pages cross-reference each other with double square bracket wikilinks, using
  the target file name. Link generously. A link to a page that does not exist yet
  is useful, because it marks something worth writing.
- The ingest workflow: read the source in full, discuss the takeaways with me before
  writing anything, then write the source page, update the entity and concept pages
  it touches, flag contradictions explicitly rather than overwriting them, update
  index.md, and append to log.md.
- The query workflow: read index.md first, then the relevant pages, then answer with
  citations. If the wiki does not contain the answer, say so plainly rather than
  filling the gap from your own knowledge.
- The lint workflow: on request, look for unflagged contradictions, stale claims,
  orphan pages with no inbound links, concepts mentioned repeatedly but with no page
  of their own, and missing cross-references. Report them. Never fix them without
  asking, because some contradictions are a deliberate record of how a claim changed.
- That you write in my voice, as described above.

## Step 3: seed the index and the log

index.md gets a heading, one line explaining what it is, and empty Sources, Entities,
Concepts and Synthesis sections. log.md gets a heading and a first entry recording
that the wiki was created today.

## Step 4: tell me what you did

List what you created. Then tell me in two or three plain sentences what I do next to
add my first source. Do not put any example or placeholder content in the wiki
itself. Empty is the correct starting state.

## How to treat me while we do this

Ask one question at a time and wait for the answer. If something is ambiguous, ask
rather than guessing. If I have asked for something that will not work well, say so
and tell me what you would do instead.

I have added [FILENAME] to raw/. Ingest it.

Read it in full, then tell me the key takeaways and what you think is worth
emphasising, before you write anything. Once I have replied, follow the ingest
workflow in CLAUDE.md.