hyatt-activity

Unified output contract. This sub-skill is part of points-activity. Its scripts/transform.py writes Date, Description, Amount and the filename hyatt_activity_<from>_<to>.csv (covered range) via the shared scripts/activity_output.py, and prints BALANCE:/COVERED:/REQUESTED:/FILE:/ROWS: to stdout. The orchestrator passes the requested period and reads those lines. References to collapsed.csv / Points / Miles below are the older single-program wording; the actual columns are Date, Description, Amount.

Hyatt Account Activity Extraction

Pulls full World of Hyatt account-activity history via Hyatt's internal JSON API (no DOM scraping, no "Load More" clicking, no chevron expanding), normalizes it, and outputs a collapsed points table whose total equals the account balance.

Prerequisites

• Claude in Chrome extension connected (list_connected_browsers → select_browser).
• User logged in to hyatt.com. The API relies on the live session cookie; the skill cannot and must not log in on the user's behalf.
• Sessions expire fairly quickly. On a scheduled run, if the fetch returns an error or zero rows, stop and ask the user to log in — do not retry blindly.

Procedure

1. Open the page

tabs_context_mcp(createIfEmpty:true) → navigate the tab to https://www.hyatt.com/profile/en-US/account-activity. Ask the user to confirm they're logged in (skip if already confirmed this session). Keep the tabId.

2. Fetch all transactions

Run scripts/fetch_activity.js via javascript_tool (action javascript_exec) in that tab. It returns {count, oldest, newest}. Sanity-check newest is recent — if it's months stale, something dropped the first page (see API quirks below).

If it returns {error: ...}, the user is almost certainly not logged in. Stop and ask.

3. Dump rows to console and read them back

Returning the raw array directly is often blocked by the chat ("Cookie/query string data"). Instead:

• Run scripts/dump_console.js (logs one HX###~... line per transaction).
• Read them with read_console_messages(pattern:"^HX\\d", limit:200, tabId:...).

4. Transform

Strip the HX###~ prefix from each line, write the remaining tilde-delimited lines to a file (e.g. /tmp/raw.txt), then run:

python3 scripts/transform.py /tmp/raw.txt <output_dir> [from] [to] [balance]
# from/to are ISO yyyy-mm-dd or '-' for unbounded; balance is the current balance or omit.

It writes normalized.csv (full, all components, all types — keep for auditing) and collapsed.csv (the deliverable), and prints the points sum.

5. Verify and present

The printed points sum must equal the account's current balance (visible on the page, top-left "Current Point Balance"). If it matches, present collapsed.csv with present_files. If it does NOT match, do not silently ship it — report the mismatch; something in the API shape likely changed (see below).

The collapsed table (the deliverable)

collapsed.csv: columns Date, Component, Points. Produced by:

1. Normalize every transaction into one row per point-bearing component.
2. Keep only real points rows (PointsType P); drop night credits (N).
3. For every category except AWARD, move the date to the last day of its month; AWARD keeps its real date.
4. Group by (Date, Component), sum Points.

All the reasoning lives in scripts/transform.py — read its docstring before editing.

API quirks (do not rediscover these the hard way)

• Endpoint: GET https://www.hyatt.com/profile/api/stay/pastactivity?pageSize=15&pageIndex=N&transactionType=&locale=en-US&startDate=&endDate=
• Pagination is 0-BASED. pageIndex=0 is the NEWEST page. Starting at 1 silently drops the most recent ~15 transactions — the single most likely bug.
• Large pageSize breaks: pageSize>=200 returns []. Use 15.
• Stop condition: response field showLoadMoreButton === false.
• Field semantics: points live in totalAmount for BONUS/AWARD/PARTNER_EARN; for STAY use baseAmount + each bonusDetail item (the STAY totalAmount can be wrong).
• bonusDetail holds the nested point-bearing components (Globalist bonus, Double Points, promos). These must each become their own row.
• AWARD and PARTNER_EARN can have a blank pointsType yet still move points — count them when totalAmount != 0.

Recurring / scheduled use (monthly)

Best run as a scheduled task on the user's own machine — it needs the user's real Chrome with a live Hyatt session, which only exists locally, so a cloud routine won't work. Use whichever scheduling primitive the host provides — Cowork: mcp__scheduled-tasks__create_scheduled_task; Claude Code: cron + headless invocation. Note for the user: local scheduled tasks run only while the computer is on and the app is open, there are no completion/failure notifications yet, and a month-old session will usually need a fresh login before the run. Each run is a clean session, so to report only new activity, save each run's collapsed.csv (e.g. dated filename) and diff against the previous one rather than re-reporting all rows.