process

Process all unprocessed notes in inbox/raw/, then scan thinking/ for any notes ready to promote.

Part 1: inbox/raw/

For each .md file in inbox/raw/:

1. Read the content and analyze what it contains.

2. Detect domain — Check if content matches any registered domain in CLAUDE.md:

   • High confidence (obvious match): suggest the domain
   • Low confidence (unclear): ask the user to choose or create a new domain
   • If no domains exist yet, ask user to create one first via the create-domain skill

3. Classify note type — Determine the type field:

   • concept, decision, reference, meeting, idea, note, belief, frame, lesson, model, goal, plan, research

4. Check for thinking content — If the content is primarily reasoning, a scratchpad, an open question, or ongoing research, offer to route to thinking/ instead of inbox/ready/:

   • "This looks like a thinking note (reasoning/research/scratchpad). Move to thinking/ with status: thinking?"
   • If yes: git mv inbox/raw/[file] thinking/[YYYY-MM-DD-topic].md, set status: thinking, skip remaining steps for this file
   • If no: continue to step 5

5. Check for multi-topic content — Scan for structural signals that the note covers 3+ independently page-worthy topics (each self-contained and useful on its own without the others):

   • 3+ H2 (##) sections with clearly different subjects
   • [fact]/[idea] blocks that are clearly unrelated and independently meaningful

   Only flag if topics are genuinely separable — not if sections are tightly related and better understood together. If detected, add a visible notice:

   "This note covers [N] independently page-worthy topics. When you run /distribute, you'll be offered to split it into separate pages."

6. Generate frontmatter — Add full YAML frontmatter:

   ---
   name: "[descriptive_kebab-case_name]"
   domain: "[detected-domain]"
   origin: braindump
   type: "[detected-type]"
   status: unprocessed
   description: "[~150 char summary]"
   tags: []
   date: "[original date from raw note, quoted string]"
   created: "[original date from raw note, quoted string]"
   last_updated: "[today, quoted string]"
   ---
   

7. Preserve original source — Before any transformation, capture the entire raw content in a collapsible callout at the bottom of the note:

   > [!quote]- Original Source
   > [verbatim raw text, exactly as captured, no formatting changes]
   

   The processed/transformed content lives above this callout as the main body. The callout is collapsed by default so it doesn't clutter reading, but the original is always available for reference.

8. Preserve observation categories — Keep any [fact], [idea], [decision], etc. syntax from the raw capture.

9. Move to inbox/ready/ — Use a two-step process to move and update status: 9a. Move the file: bash mv inbox/raw/[filename] inbox/ready/[filename] (Use plain mv for untracked files, git mv if tracked) 9b. Edit inbox/ready/[filename]: change status: unprocessed → status: ready

10. Update brain/QUEUE.md — Record or update the note's processing state:

    • If the note path is not already in the table: append a new row with Stage ready, today's date, and any relevant notes.
    • If the note path already exists (e.g., it was previously in the queue as raw): update the Stage field to ready and update Last touched.

    | inbox/ready/[filename] | ready | YYYY-MM-DD | — |
    

11. Report for each note:

    • Original filename
    • Detected domain + type
    • Generated description
    • New location in inbox/ready/ (or thinking/ if rerouted)