Plan award trips from Claude Code. getaway sweeps seats.aero across 28 mileage programs, composes whole journeys — flights out and home, hotel award nights via rooms.aero — and judges every option like you would: seat quality, layovers, status earning, cash-fare anomalies.

/plugin marketplace add yasyf/getaway
/plugin install getaway@getaway

getaway's hooks ride the captain-hook plugin and its live boards ride cc-present. Both install themselves — the first session registers their marketplaces and Claude Code resolves the dependencies — so you add only getaway's own marketplace, on install and on upgrade alike.

You need uv on PATH — the planning engine is a bundled Python CLI — and a seats.aero Pro API key, generated on the seats.aero Settings page under the API tab:

export SEATS_AERO_API_KEY=pro_YOUR_KEY

Run /getaway:onboard once. The form arrives pre-filled from Gmail and the airline and bank sites you're already logged into, and nothing is written until you hit Submit. Skipping it is fine: planning runs on a neutral profile, and balances bias ranking, never gate it. Then hand over the trip in one sentence, as messy as you like:

I want to go away for roughly a week, leaving in the next couple days, and want something outside north america, warm, beachy, and has cheap points tickets for business class. want to avoid the common places we always go like seoul or tokyo

Claude pins the ask into a trip, derives a judgment profile from every clause — business class puts seat quality and affordability first, "warm, beachy" activates destination context, "always go" writes Seoul and Tokyo to your durable avoid list — and dispatches the planning pipeline. Finalists land on a live board, one evidence line per active factor:

Driving with an agent? Paste this:

/plugin marketplace add yasyf/getaway
/plugin install getaway@getaway

Checking United, Aeroplan, and Alaska one site at a time, for the same seat, is an evening gone. Ask once instead:

Any business award space from SFO to NRT or HND, September 1–14?

One sweep covers all 28 programs, and every row carries per-cabin mileage, remaining seats, and a freshness stamp. Rows land in a local cache, so follow-ups — other dates, other cabins — answer without spending another API call.

A six-figure Aeroplan balance with no destination in mind is money rotting. Flip the search around:

Where can Aeroplan take me in business from North America in October?

Bulk availability scans one program across whole regions, so the answer is a list of real routes with real award space — island space most people never think to search included:

2026-07-11  CMB  MLE  aeroplan  12500  9  FZ, GF
2026-07-11  TNR  MRU  aeroplan  12500  9  MK
2026-07-11  BLR  MLE  aeroplan  22500  1  AI, GF
2026-07-11  FRA  FUE  aeroplan  22500  1  2L, LX, WK

A direct award to a second-tier city rarely exists, and when it does it costs double. Lie-flat space to the big gateways is everywhere. Split the trip instead:

Fly me lie-flat to Asia in November — happy to land at a hub like NRT
and hop the last leg on a cash ticket.

The planner composes hybrid routings — gateway awards with cash hops, open jaws, two-award stitches — beside direct awards and compares them on total cost. Each cash leg's cabin is judged per leg from its length: a short hop books economy, a long haul books business:

SFO  NRT  aeroplan business  88000 miles + $118 taxes  lie-flat
NRT  TPE  cash economy       $96                       3h45m hop

A trip plan that dies with the session is a trip planned twice. Every constraint, sweep, and finalist persists per trip under ~/.getaway/trips/<slug>/, so resuming is one ask:

Pick up the warm-beachy trip where we left off.

The resume brief carries the pinned constraints, per-phase freshness, finalists so far, and expiring credits. Phases whose inputs haven't changed skip wholesale — a resumed plan spends zero API quota until you change something.

Every real ask becomes a trip: the verbatim ask, pinned constraints and preference lanes, and a judgment profile derived from both. The bundled CLI compiles each trip into an execution graph — sweeps, shortlists, journey composition, evidence, assess, rank — and the shipped workflow walks it. Preferences order and annotate but never gate, cost stays dominant with judgment reordering only within a cost tier, and a trip with lodging in scope picks up hotel award nights from rooms.aero along the way. The full doctrine lives in skills/getaway/references/.

The plugin ships three skills, each triggered by its own asks:

• A seats.aero Pro subscription for the API key. Pro keys get 1,000 calls per day; the planner tracks quota and answers follow-ups from cache.
• uv — the planning engine is a bundled Python CLI run through uv run; the first call builds its environment.
• The 1Password op CLI, when the key comes from an op_ref in ~/.getaway/preferences.json instead of the environment — a secret reference like op://Vault/item/field that the CLI resolves with op read whenever the env var is unset.
• The cc-present plugin renders the live boards — a declared dependency, auto-installed once its marketplace is added; getaway ships its block pack at .claude/components.

The raw Partner API surface — endpoints, params, data shapes, quota, and per-program coverage — lives in docs/seats-aero-api.md. The CLI documents itself: uv run --project <plugin-root>/cli getaway --help.

Licensed under PolyForm-Noncommercial-1.0.0.