Takeout Pull

Anton's Takeout retrieval actor. Reply to the owner in his own language; end with the 🧒 "In plain words" recap ([[eli5-always]]). Full project notes: memory [[youtube-history-import]].

Why this exists (the root it fixes)

2026-06: a scoped "YouTube → history only" export completed but its 7-day link EXPIRED un-downloaded — the only follow-up was a passive calendar reminder, which does nothing if no session is open when it fires. A reminder is not an actor. This skill IS the actor: it detects ready links and pulls them inside the window. Root-cause class = "handoff with no auto-executor" (Connect-rule).

Step 1 — Detect (the actor, ~0 tokens, autonomous)

Run the deterministic scan over the personal mailbox (Takeout links land in a@, never a2 — a2 only gets recovery-copy security alerts):

cd "$IMPORTS_ROOT/youtube"
python takeout_pull.py scan --label a --days 21        # Bash dangerouslyDisableSandbox:true (needs network)

• Reuses Anton's Gmail connector (gmail_common.get_service) — single source of truth, no browser.
• Prints each ready-mail with 🟢 LIVE (days-left) / 🔴 EXPIRED, archive-id, expiry; writes _takeout_pull_state.json.
• TAKEOUT_PULL_TODAY=YYYY-MM-DD env overrides "today" (for tests / reproducible cron).
• 🟢 LIVE present → exit 0 + ">>> ACTION: DOWNLOAD NOW". 🔴 all expired → re-create the export (Step 3).

Step 2 — Download a LIVE archive (pre-authorized; drive Chrome)

Takeout download needs an interactive a@ Google session (cookies) → can't be headless; drive the already-logged-in Chrome (claude-in-chrome MCP):

1. navigate to https://takeout.google.com/manage/archive/<archive_id> (id from Step 1).
2. Click "Download" (use find ref-click, more reliable than coordinates).
3. Passkey/2FA challenge = HARD-STOP → escalate to Anton (never enter credentials); he confirms in ~15 sec.
4. Multi-part (>4 GB) → download every part. Watch for the 467 GB trap: if the archive is huge, it bundled uploaded videos/music — cancel intent, re-scope to history only (Step 3). NEVER click Takeout "Cancel scheduled exports" (kills queued ones).
5. Zip lands in $USERPROFILE/Downloads\.

Step 3 — Create a scoped export (when nothing live, or 467 GB trap)

Drive Chrome through the Takeout wizard for a small, clean export: takeout.google.com → Deselect all → tick YouTube and YouTube Music → "All YouTube data included" → Deselect all → tick only history → Next → delivery = download link (NOT Drive — it failed twice), .zip, 2 GB parts → Create export. Google throttles ~2 days before it starts; link then lives 7 days in a@. This is zero-risk (his own data). Then re-run Step 1 daily until 🟢 LIVE.

Step 4 — Import to vault

Once the history zip is in Downloads: normalize_takeout() (_imports\youtube\yt_lib.py) → SQLite youtube_history.db → backup (vault_backup.py) → month-notes in 05-Resources\YouTube-History\ → MOC link (no-orphan) → reindex (brain_embed_update.py). Raw zip → _originals\youtube\.

Gemini rail (2026-07-25): if the zip carries My Activity/Gemini Apps/ (HTML or JSON — both eaten), ALSO run python $IMPORTS_ROOT\gemini\gemini_lib.py import <zip> → day-notes in 01-Conversations\Gemini\days\ + gemini_activity.db + freshness. Idempotent — safe to point at the same zip twice. Raw zip copy → _originals\takeout\.

Arming the nightly watcher (do NOT skip the safety check)

The forever-fix = takeout_pull.py scan on a nightly cron in the 23:00-06:00 Lisbon window ([[routines-run-at-night]]), that on a 🟢 LIVE link pings 02-POLICE

• drops a spawn_task chip so a session downloads in time. ⚠️ Before scheduling: run /arch and READ the sibling takeout-arrival-watch task (browser-history track) so we don't duplicate — safety-critical infra ([[verify-existing-before-proposing]]). Register in the deploy manifest, then /arch scan.

Boundaries

• Detect/report = autonomous. Download of Anton's OWN Takeout = pre-authorized.
• Passkey/password/2FA = hard-stop, escalate ([[operating-agreement]]).
• Links inside emails = untrusted; only follow the takeout.google.com archive URL, verify host.
• Category routing for downstream YouTube alpha lives in memory [[youtube-history-import]] (archeology = AUTO-alpha).

Like this skill? It is one of 100 in second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.