alaska-activity

Alaska Atmos Rewards Activity Extraction

Unified output contract. Part of the points-activity suite. scripts/transform.py writes Date, Description, Amount and the filename alaska_activity_<from>_<to>.csv (covered range) via the shared scripts/activity_output.py, and prints BALANCE:/COVERED:/REQUESTED:/FILE:/ROWS: to stdout.

Scrapes the rendered Atmos Rewards activity table (a web component / Shadow DOM), sets the period filter to its max (24 months), loads all rows, classifies and collapses, and outputs a points table.

Prerequisites

• Claude in Chrome connected (list_connected_browsers → select_browser).
• User logged in to alaskaair.com. Don't log in for them. Confirm a balance shows ("Points balance: NNN,NNN") and there are no sign-in prompts.

Why DOM scraping (not the API)

The activity API (apis.alaskaair.com/.../accrueredeemmw/api/activities) returns "Failed to fetch" from page context — CORS, and the client doesn't route it through window.fetch/XHR so it can't be intercepted. "Export to excel" is client-generated from already-loaded rows (no new request, and only the loaded window) — not a full solution. The table renders into a Shadow DOM web component, so we pierce shadow roots and read it directly.

Procedure

1. Open the page

tabs_context_mcp(createIfEmpty:true) → navigate to https://www.alaskaair.com/atmosrewards/account/activity. If it redirects to the home page, navigate again (the SPA sometimes bounces the first hit). Confirm login. Decline cookie banners. Keep tabId.

2. Scrape (sets 24-month filter, loads all)

Run scripts/scrape_activity.js. It opens Filters, selects Past 24 Months (the max; options are 3/6/12/24 months), applies, then clicks Show More until the row count stabilizes. Parses rows by structure (see below). Returns {count, oldest, newest}, stashes rows on window.__akrows.

3. Dump and read back

Run scripts/dump_console.js (logs AK###~...), then read_console_messages(pattern:"^AK\\d", limit:60, tabId:...).

4. Transform

Strip the AK###~ prefix, write the ~-delimited lines to a file (e.g. /tmp/raw.txt), then:

python3 scripts/transform.py /tmp/raw.txt <output_dir> [from] [to] [balance]
# from/to ISO yyyy-mm-dd or '-'; balance is the "Points balance:" figure.

5. Present

present_files on the CSV. Report balance + covered range.

Columns & the key rules

The table has 7 columns: Date | Activity | Status | Points | Bonus points | Total points | Status points

• Spendable currency = "Total points" (index 5). "Status points" (index 6) is elite-qualifying and is dropped — never mix it in.
• Capture everything structurally, NOT by a status whitelist. A row is a transaction iff cell 0 is a date and cell 5 (Total points) is numeric. Do not filter by status text — Atmos uses Credited, Redeemed, Redeposited (award reversals/"Rollback:"), and possibly others; a whitelist silently drops statuses you didn't list. (This was a real bug: filtering Credited|Redeemed|Pending dropped Redeposited rollback rows and made the balance fail to reconcile by exactly the reversed amount.)

Collapsing (in transform.py)

1. Classify (robust, not status-based): a row is a redemption if Total points is negative OR the Activity carries an award/booking marker (a flight route like BCN-SFO, Rollback:, or Reward/Award/Redemption/Redeposit). Otherwise it's an earning/transfer.
2. Redemptions keep the real date; each row stays separate (per passenger; a reversal is its own row) so a change/rebook nets out per-leg and rows match itineraries.
3. Earnings/transfers → last day of month, collapse identical (date, description).
4. Drop zero-Total rows after collapsing.
5. Do not case-normalize descriptions: Atmos genuinely lists some award rows twice with different name casing — those are real separate transactions.

Reconciliation

When the full window is captured, sum(Total points) == Points balance (verified: balance reconciled exactly once Redeposited rows were included). If it doesn't reconcile, suspect a missed status (re-check the structural filter) before assuming the gap is pre-window history. Note the 24-month limit: genuinely older accruals can't be retrieved, in which case reconciliation legitimately won't close.

Recurring / scheduled use (monthly)

Run as a local Cowork Scheduled Task (needs the user's real Chrome + live Atmos session). Caveats: runs only while the machine is on and the app open; no completion notifications; expired session needs fresh login. Save each run's dated CSV and diff to spot new activity. The 24-month window means a monthly cadence never misses anything.