Files
obsidian-vault/personal/projects/personal-os/personal-os-zoom-import.md
T

3.1 KiB

Zoom My Notes manual browser fallback

The installed com.personalos.zoom-browser LaunchAgent runs every 30 minutes (StartInterval=1800); the versioned definition is scripts/com.personalos.zoom-browser.plist. The wrapper keeps the existing single-import lock and Chrome login behaviour. Only a successful source scrape refreshes sync-status.json.observed_at. Planning considers Zoom coverage stale after 90 minutes; Mattermost source fetches and discussion packet/work captures expire after 45 minutes. Completing model processing never refreshes these source clocks. Same-input reconciliation may advance a work's capture time only after rebuilding its source packet.

The scheduled importer is scripts/zoom-browser-sync.js. Use this manual workflow only when checking a page in the authorized Chrome session. The starting point is https://us01mynotes.zoom.us/home; use All notes → Owned by me. Import the authorized available notes, continuing all index pages and skipping unchanged IDs already represented in the local index. A former one-off date cutoff is not a completeness rule.

Use an authorized browser session for read-only browser actions. Open each eligible row, then read the visible page DOM. The transcript consists of a speaker/time header and one or more text blocks. This read-only extraction function has been checked against the 2026-09-29 Apple Weekly page:

let browserNote = await tab.playwright.evaluate(() => {
  const title = document.querySelector('#docs-title')?.textContent?.trim() || document.title;
  const seen = new Set();
  const transcript = [...document.querySelectorAll('span')]
    .filter(el => /^\d{2}:\d{2}:\d{2}$/.test(el.textContent?.trim() || ''))
    .map(el => {
      const block = el.parentElement?.parentElement;
      if (!block || seen.has(block)) return null;
      seen.add(block);
      const head = el.parentElement;
      const speaker = [...head.querySelectorAll('span')]
        .map(x => x.textContent?.trim() || '')
        .find(x => x && !/^\d{2}:\d{2}:\d{2}$/.test(x));
      const text = [...block.children].filter(x => x !== head)
        .map(x => x.textContent?.trim()).filter(Boolean).join(' ');
      return { time: el.textContent.trim(), speaker: speaker || '', text };
    }).filter(x => x?.text);
  return { note_id: location.pathname.slice(1), title,
    source_url: location.href, transcript };
});

Check that the URL is a single note under us01mynotes.zoom.us, the first and last transcript times look complete, and every cue has a speaker and text. Write the JSON returned by this function to a private temporary file using the normal file tools. For large notes, output and write it in numbered chunks, then validate the assembled JSON before import. Run:

node scripts/zoom-browser-ingest.js < /private/tmp/zoom-note.json

The importer writes transcript.md, transcript.json, and index.json to /Users/admin/obsidian/work/zoom-transcripts. It updates an existing note when its content changes. Keep the raw transcript local; do not post it to chat, Mattermost, or another service. Browser TXT download redirects to file-paa.zoom.us and is not part of this workflow.