Process Evernote

You are converting Evernote notes into structured Obsidian markdown notes for the mind map vault.

Evernote's local storage is read directly — no API or export required. The pipeline:
1. **Prepare** a batch → extracts ENML content from Evernote's local Yjs storage and generates a Claude prompt
2. **Analyze** → you read the prompt and generate structured metadata JSON
3. **Apply** → creates Obsidian notes, renders web clip previews via Playwright, and copies attachments

## Quick Start

```bash
# Check status
source venv/bin/activate && python src/process_evernote.py status

# Prepare next batch (default 10 notes)
python src/process_evernote.py prepare 10

# After you generate the response JSON:
python src/process_evernote.py apply BATCH_NUM

Full Workflow

Step 1: Check Status

source venv/bin/activate && python src/process_evernote.py status

Shows: total notes, processed, remaining, notes with attachments, current batch number.

Step 2: Prepare a Batch

source venv/bin/activate && python src/process_evernote.py prepare [batch_size]
# or target a specific note by GUID:
python src/process_evernote.py prepare --guid <note-guid>

• Default batch size: 10 notes
• Creates evernote_batch_N.json (note data + ENML + resources)
• Creates evernote_batch_N_prompt.txt (Claude analysis prompt with clean Markdown content)

Content extraction chain (highest → lowest fidelity):

1. pycrdt — decodes Yjs .dat file → raw ENML → clean Markdown via markdownify. Produces full article text with proper headings, links, and punctuation.
2. SQLite — reads Offline_Search_Note_Content + Nodes_Note.source_URL. Clean text, no structure.
3. strings — legacy fallback for notes where both above fail. Noisy but workable.

The prompt labels the method: CONTENT [pycrdt], CONTENT [sqlite], or CONTENT [strings].

Note types the script handles:

• Web clips: Have source_url from ENML CSS metadata or SQLite Nodes_Note
• Personal notes: User-written content, no source URL

Step 3: Analyze the Batch

Read evernote_batch_N_prompt.txt and generate structured metadata for each note.

For each note in the batch, produce:

{
  "guid": "the-note-guid-from-prompt",
  "title": "Concise, accurate title",
  "source_name": "Publication name | null for personal notes",
  "source_url": "https://article-url | null for personal notes",
  "source_date": "YYYY-MM-DD",
  "summary": "2-3 sentence summary with specific insights",
  "key_points": [
    "Specific, substantive point 1",
    "Specific, substantive point 2",
    "Specific, substantive point 3",
    "Specific, substantive point 4",
    "Specific, substantive point 5"
  ],
  "tags": ["Topic", "Subtopic", "Tag"],
  "authors": ["Author Name"],
  "author_urls": {"Author Name": "https://profile-url | null"}
}

Analysis guidelines:

• For web clips: use the article content (now full clean Markdown) to determine real title, not Evernote title
• For personal notes: shorter key_points list is fine — capture the essence
• source_date: prefer the article's publication date; fall back to the Created date shown in the prompt
• Tags: use Title-Case-With-Hyphens format (e.g., Mental-Health, AI, San-Francisco)
• author_urls: use null if no profile URL is apparent from the content
• Content is now up to 6000 chars of clean Markdown — use it fully

Save the complete JSON array as evernote_batch_N_response.json.

Step 4: Apply

source venv/bin/activate && python src/process_evernote.py apply N
# or with explicit response file:
python src/process_evernote.py apply N evernote_batch_N_response.json

This will:

• Create Obsidian markdown notes in Lifegames/
• Render a full-length PNG preview of each web clip via Playwright/Chromium:
  • ENML HTML with base64-embedded images from resource-cache
  • Source Serif 4 (Google Fonts) mapped to source-serif-pro — Medium's actual body font
  • CSS quote escaping fix (pycrdt outputs unescaped " in style attributes)
  • Content height auto-detected via JS getBoundingClientRect
  • 748px logical width @ 2x device pixel ratio
• Copy all attachment images to Lifegames/Attachments/ with hash-based filenames
• Update evernote_progress.json with processed GUIDs

Step 5: Repeat

Continue until all notes are processed:

python src/process_evernote.py status
python src/process_evernote.py prepare 10
# ... analyze ... apply ...

Generated Note Format

Web Clip (with source URL)

---
title: "Article Title"
source: "Publication Name"
source_url: "https://..."
date: YYYY-MM-DD
authors:
  - Author Name
tags: ["Topic", "Tag"]
filename: "8charhash"
cssclass: article-note
file_hash: "8charhash"
evernote_guid: "full-guid"
evernote_notebook: "Personal"
generated_by: "claude-code"
generated_model: "claude-opus-4-6"
generated_at: "YYYY-MM-DD"
---

# Article Title

> [!info] Article Information
> - **Source**: [Publication](url)
> - **Date**: YYYY-MM-DD
> - **Authors**: [Name](url)

## Summary
...

## Key Points
- ...

## Author Information
- [Author](url)

## Linked Concepts
<!-- For manual wiki-linking later -->

## Notes

## Attachments
- ![[hash_preview.png]]

Personal Note (no source URL)

---
title: "Note Title"
date: YYYY-MM-DD
tags: ["Tag"]
filename: "8charhash"
evernote_guid: "full-guid"
...
---

# Note Title

## Summary
...

## Key Points
- ...

## Linked Concepts
<!-- For manual wiki-linking later -->

## Notes

## Attachments
- ![[hash_preview.png]]

File Naming

• Preview: {8charhash}_preview.png — full-length Playwright screenshot of original ENML
• Resources: {8charhash}.jpg (primary), {8charhash}_1.jpg, {8charhash}_2.jpg (additional)
• Note hash: MD5 of the note GUID (consistent, reproducible across runs)
• Notes: sanitized title (removes []#^|\/:?*<>", max 100 chars)

Evernote Storage Locations (read-only)

• ENML content: `~/Library/Containers/co