Export format#

What scrap export DIR writes, and what Settings, Export everything (GET /api/store/download) puts in its zip: the same files either way. How to make one is in Export your pages.

Layout#

Path	Holds
`<slug>.md`	One per page not in the trash: front matter, then the page’s text
`files/<sha256><extension>`	The originals of every page’s files, each once, named by content: the SHA-256 in hex and the usual extension for its type (none if there is none). Not the name it was uploaded under
`files/<sha256>-<capture id>.excalidraw`	A photo’s drawing (a drawing belongs to a capture, so one photo on two pages can have two)
`_claude/<slug>.md`	What a model wrote about the page, for a page that has any

The file name is the page’s slug, the last part of its address (/p/<slug>), so a [[slug]] or [[slug|label]] link in a page’s text names the file it points to, with nothing rewritten.

A page’s file#

The front matter is YAML between --- lines, and holds only what the page’s owner gave it. Every value is JSON (which YAML reads), except created and updated, which are bare so that tools such as Obsidian read them as dates.

Key	Value
`title`	The page’s title
`aliases`	A list holding the title, so Obsidian finds the page by it
`id`	The page’s id (a ULID); it stays the same when the title or slug changes
`kind`	`note`, `todo`, `consume` (a Later page), `whiteboard` or `document`
`tags`	The tags you gave it, as stored (lower case, `/` for nesting, spaces kept); left out if none. Tags Claude added are in `_claude/` instead
`created`, `updated`	ISO 8601 times in UTC: when the page was made and last edited
`due`, `priority`	A todo’s due date (`YYYY-MM-DD`) and priority (`high`, `medium`, `low`), if it has them
any field	Each of the page’s fields as a key of its own, such as `status`, `subtype` or `summary`
`fields`	Only a field whose name is one of the keys above, under its name
`made_by`	`claude` (written over MCP), `topic` (a Claude topic page) or `gc` (`scrap gc`’s review page); left out for a page you made
`attachments`	The files the page has but its text does not show or link to, each `"[[files/<name>]]"`

Then comes a blank line and the page’s text exactly as stored, but with each address of a file of scrap’s (/api/blobs/<sha256>...) made a relative files/<name>. Image alt text and link text are kept as typed. A todo’s task boxes stay in its text; their count is not in the front matter.

Nothing a model wrote is in a page’s file: not a summary, a tag Claude added, a write-up, a photo’s description, the text read from a photo or a drawing.

Claude’s notes: `_claude/<slug>.md`#

Front matter generated_by: "claude" and page: (the page’s slug), a line saying a model wrote it, and a link to the page. Then a section headed Claude’s notes (written by a model, may be wrong) with whichever the page has: the summary, the tags Claude added, a board’s write-up or a topic’s synthesis, related pages, and for each photo the description (a quote), the text read from it (a fenced block) and a link to its drawing. Its links to pages are relative ([label](../<slug>.md)), so they do not find another _claude/ file of the same name. A [[...]] whose target is not a slug, such as [[Some Title]], is kept as typed.

Examples#

A note with tags and links, notes-on-ibek.md:

---
title: "Notes on ibek"
aliases: ["Notes on ibek"]
id: "01M64CHHHWJE5BARWPH83NZW7T"
kind: "note"
tags: ["epics/ibek", "lab meeting"]
created: 2026-10-29T17:50:21.755Z
updated: 2026-10-29T17:50:21.755Z
---

Notes on ibek
See [[rtems-iocs]] and [[rtems-iocs|the RTEMS page]].

A todo with a photo, fix-the-broken-fence-panel.md:

---
title: "Fix the broken fence panel"
aliases: ["Fix the broken fence panel"]
id: "01M64CGYA7MZNKGWCYG8XCCA1K"
kind: "todo"
tags: ["home", "garden"]
created: 2026-10-29T17:50:02.054Z
updated: 2026-10-29T17:50:02.054Z
due: "2026-10-24"
priority: "medium"
status: "open"
---

Fix the broken fence panel
- [ ] measure the gap
- [ ] buy a panel

![](files/5bbacd0b…9d.jpg)

A whiteboard photo with a PDF its text does not show, sprint-board.md (file names shortened here: each is the full 64-digit SHA-256):

---
title: "Sprint board"
aliases: ["Sprint board"]
id: "01M64CGYGY8K291Q2C11G4EG3F"
kind: "whiteboard"
tags: ["scrap"]
created: 2026-10-29T17:50:02.269Z
updated: 2026-10-29T17:50:02.286Z
attachments: ["[[files/5a838678…f5.pdf]]"]
---

Sprint board

![](files/81fdcce0…15.jpg)

My typed notes from the meeting.

and its _claude/sprint-board.md:

---
generated_by: "claude"
page: "sprint-board"
---

*Written by a model, may be wrong. Not part of the page: an import never reads this folder.*

About [Sprint board](../sprint-board.md).

## Claude's notes (written by a model, may be wrong)

**Summary**

> The scrap architecture board.

**Tags Claude added:** architecture

**Write-up**

> The board shows the PWA talking to FastAPI, which keeps SQLite.

### Photo [whiteboard.jpg](../files/81fdcce0…15.jpg)

> A whiteboard with four boxes joined by arrows.

```text
PWA -> FastAPI -> SQLite
Ingest
```

[Drawing of whiteboard.jpg](../files/81fdcce0…15-01M63YSFGWQ1NXW1RNJ47E65GH.excalidraw)

What is left out#

Pages in the trash, revisions, the text scrap read from PDFs and text files, the pages that link to a page (each page’s own links say it), and Claude’s call records. A file whose embed was deleted from a page’s text, which the page no longer shows, is left out too.

Reading it back#

scrap import and Settings, Import Markdown… (POST /api/import) read this layout back: every page as it was (its id, slug, title, text, kind, fields, a todo’s due date and priority, tags, dates, made_by and files), with the tag import/YYYY-MM-DD added. They never read _claude/. Each files/<name> in a page’s text becomes the stored file’s address again, in one form: an image shown as /api/blobs/<sha256>/preview?size=2048, any other file linked as /api/blobs/<sha256>. A topic page is not read back (Claude makes topics again). See Import your pages.