FTP from Strava rides

Estimates FTP from rides the user has already done. No ramp test, no new workout.

Rules

• Read-only. Use only Strava's read tools: list_activities, get_activity_performance, get_athlete_profile, get_athlete_zones. Never change FTP or anything else in Strava or any other app. Give the user the number; they enter it where they want it.
• Never use Strava's own FTP estimate as an input. Show the FTP stored in Strava for comparison only.
• Don't guess numbers. Every watt in the input file comes from a tool result. If a value is missing, leave it out.
• The rides are the user's personal data. Keep the input file in a temporary or scratch folder, and don't share it anywhere.

Step 0: is the Strava connector there?

Look for Strava's tools (list_activities, get_activity_performance). Depending on the app they may carry a prefix, such as <something>__list_activities. If they aren't available, stop and tell the user: "Connect Strava's official connector in Claude (Settings, Connectors, Strava), turn it on for this chat, then run /ftp again."

Step 1: list the rides

Call list_activities with range_start = today minus 84 days (YYYY-MM-DDT00:00:00), range_end = today (YYYY-MM-DDT23:59:59), first: 100, include_tags: true. Follow has_next_page until done.

Keep sport_type Ride, GravelRide, MountainBikeRide and VirtualRide. Drop e-bikes, runs, walks, strength and generic "Workout" entries. Drop rides with moving_time under 180 seconds. is_trainer: true means indoor.

If more than about 40 rides are left, read only the last 42 days and set "windows": [42] in the input file. Each performance answer is 2 to 10 KB, and the 42-day window is the one that sets the number.

Step 2: read each ride's best efforts

Call get_activity_performance for each kept ride. For each one record:

• has_device_watts. If false, the ride has no power meter: add it with "has_power": false and no values.
• From best_efforts, the value of GreatestPower3min, GreatestPower5min, GreatestPower10min, GreatestPower20min, GreatestPower30min and GreatestPower1hr, as "3", "5", "10", "20", "30", "60". Skip any the ride doesn't have.
• From segment_efforts, any effort with moving_time of 3000 seconds (50 minutes) or more, as {"name", "minutes": moving_time / 60, "watts": avg_watts}. A long, steady climb is the best floor for FTP there is.

Strava's per-ride best efforts agree with a second-by-second recomputation of the power stream to within about 1 W, so there's no need to pull streams.

Optional extras: get_athlete_profile gives weight (kg) for W/kg. get_athlete_zones gives the FTP stored in Strava and ftp_is_estimated.

Step 3: write the input file

Follow examples/example-input.json in this skill's folder:

{
  "as_of": "YYYY-MM-DD",
  "windows": [42, 84],
  "weight_kg": 70,
  "current_ftp": {"watts": 240, "estimated": false},
  "rides": [
    {"date": "YYYY-MM-DD", "name": "ride name", "indoor": false, "has_power": true,
     "best": {"3": 330, "5": 318, "10": 296, "20": 272, "30": 262, "60": 248},
     "long_efforts": [{"name": "segment name", "minutes": 56, "watts": 250}]}
  ]
}

as_of is today. Dates are the ride's local start date.

Step 4: run the math

python3 <this skill's folder>/scripts/ftp_calc.py <input.json>

Add --out <file.md> to save the report, or --json for the raw numbers. Standard library only; it runs anywhere Python 3.9+ does.

If it exits with "These best efforts can't all be right", a value was copied wrong (a longer duration came out higher than a shorter one it contains). Re-read that ride's get_activity_performance and fix the file. Don't edit numbers to make the check pass.

What the script does

1. Best power per duration across each window.
2. Two primary estimates: 95% of the best 20 minutes, and a 2-parameter critical power fit (work against time, least squares) on the 3, 5, 10 and 20 minute bests.
3. One sensitivity estimate: the same critical power model fit as power against 1/time. It leans on short efforts and usually reads higher, so it only sets the top of the range.
4. Floors: the best 60 minutes, and the hardest steady effort of 50 minutes or more.
5. FTP = median of the 42-day primary estimates, rounded to 5 W. Range = the higher of the floor and the lowest estimate, up to the highest estimate.
6. Sanity ratios: 20 min / 5 min is typically 0.85 to 0.88, and 60 min / 20 min 0.90 to 0.95. Outside those bands, one of the efforts probably wasn't all-out, and the report says which estimate to trust.

Step 5: tell the user

Lead with the FTP and range. Name the rides behind the 5, 20 and 60 minute bests and the long effort floor. If Strava's stored FTP is 5 W or more away, say so; changing it is the user's call. Mention any indoor bests (trainer power can read differently from a power meter).

If they want a sharper number without a ramp test: one evenly paced, hard 20 to 30 minute effort, ideally on a steady climb. 95% of a true 20-minute effort, or the full 30-minute average, beats any model.

Limits

• Needs a power meter or smart trainer. Strava's estimated watts are not used.
• Bests set late in long rides understate fresh power, so the estimate leans conservative.
• It's an estimate from riding data, not a lab test.