Source Code

Air France (AFKL Open Data) flight tracker

Quick start (one-off status)

Create an API key (and optional secret)

Register on: https://developer.airfranceklm.com
Subscribe to the Open Data product(s) you need (at least Flight Status API)
Generate credentials (API key; some accounts also provide an API secret)

Provide API credentials (do not print them):

Preferred: env vars AFKL_API_KEY (and optional AFKL_API_SECRET)
Or files in your state dir (CLAWDBOT_STATE_DIR or ./state):
- afkl_api_key.txt (chmod 600)
- afkl_api_secret.txt (chmod 600, optional)

Query flight status:

Run: node skills/airfrance-afkl/scripts/afkl_flightstatus_query.mjs --carrier AF --flight 7 --origin JFK --dep-date 2026-01-29

Notes:

Send Accept: */* (API returns application/hal+json).
Keep within limits: <= 1 request/sec. When making multiple calls, sleep ~1100ms between them.

Start monitoring (watcher)

Use when the user wants proactive updates.

Run: node skills/airfrance-afkl/scripts/afkl_watch_flight.mjs --carrier AF --flight 7 --origin JFK --dep-date 2026-01-29

What it does:

Fetches the operational flight(s) for the date window.
Emits a single message only when something meaningful changes.
Also follows the previous-flight chain (flightRelations.previousFlightData.id) up to a configurable depth and alerts if a previous segment is delayed/cancelled.

Polling strategy (default):

36h before departure: at most every 60 min
36h→12h: every 30 min
12h→3h: every 15 min
3h→departure: every 5–10 min (stay under daily quota)
After departure: every 30 min until arrival

Implementation detail: run cron every 5–15 min, but the script self-throttles using a state file so it won’t hit the API when it’s not time. The watcher prints no output when nothing changed (so cron jobs can send only when stdout is non-empty).

Input shorthand

Preferred user-facing format:

AF7 demain / AF7 jeudi

Interpretation rule:

The day always refers to the departure date (not arrival).

Implementation notes:

Convert relative day words to a departure date in the user’s timezone unless the origin timezone is explicitly known.
When ambiguous (long-haul crossing midnight), prefer the departure local date at the origin if origin is known.

(For scripts, still pass --origin + --dep-date YYYY-MM-DD.)

Interpret “interesting” fields

See references/fields.md for:

flightRelations (prev/next)
places.* (terminal/gate/check-in zone)
times.* (scheduled/estimated/latest/actual)
aircraft (type, registration)
“parking position” / stand-type hints (when present)
Wi‑Fi hints and how to reason about cabin recency

Cabin recency / upgrade heuristics

When aircraft registration is available:

Use tail number to infer sub-fleet and likely cabin generation.
If data suggests older config (or no Wi‑Fi), upgrading can be more/less worth it.

Be conservative:

Open Data often doesn’t expose exact seat model; treat this as best-effort.