Race Recap

1. One sentence: What does the Skill do?

Race Recap turns one race weekend into a private recap on the NAS: photos matched to the kilometer where they were taken, splits, a short written recap and an album.

2. Who is the target user and problem?

Amateur triathletes and runners, and the family who photographed them. They hold hundreds of photos and a GPX or FIT file, unconnected. Uploading faces and home-start routes to a public cloud is a poor trade.

3. What is your proposed solution?

The Skill finds the activity file and race-day photos, places each photo on the course by capture time, picks about a dozen, computes splits, drafts a 120-word recap from real numbers and builds an album, all on the NAS.

4. What is the Core User Flow?

1. The user asks: "Make a recap of Sunday's half marathon."
2. The Skill finds the track and the photos from the race window.
3. It shows proposed photos with kilometers, splits and the draft recap.
4. The user approves or edits.
5. The Skill creates the "Race Recap" album and reports what it made.

Sharing needs a separate yes.

5. What substantive difference does your idea offer?

Photo apps sort by date and face, sports apps show the track, and neither joins them. Cloud recap tools need uploads. Race Recap joins both by time and position, quotes only real numbers and never leaves the device.

6. Which UGREEN AI NAS capabilities will be used, and at which step?

• Files: locate the activity file and photo folder (step 2).
• Documents: read track points, distance, pace, heart rate (step 2).
• Albums: search photos by time, location and named person (step 2); create the album (step 5).
• Local AI model: shortlist photos and write the recap (step 3).

7. Technical approach

SKILL.md only, no scripts or downloads. Inputs: the request, optionally race date, activity file and photo folder. Workflow: read the track, map photo times to the nearest track point, derive kilometers and splits, shortlist with the local model, write the recap from computed values. Permissions: read chosen folders, create one album, confirm before creating or sharing. GPX is plain XML, so we start there; if Documents cannot read FIT, the Skill asks for a GPX export. No network calls.

8. Current development status

Implemented: SKILL.md and two reference files drafted. Not implemented: any NAS run. Pending NAS validation if selected: FIT reading, photo time and location search, album creation, local model quality.

9. Final validation plan

Core test 1: a half marathon with a GPX file and 150 geotagged photos. Photos sit on plausible kilometers, splits match the file, the album appears. Core test 2: a triathlon with a FIT file. The Skill reads it or asks for a GPX export, and the recap quotes only real numbers. Edge case: photos in the race window without timestamps, then a request with no activity file. The Skill names what is missing and stops, with no substitute photos.

10. Known limitations

No NAS run yet. FIT support depends on Documents. Photos without time data cannot be placed. Person search needs named people in Albums. Clock offsets between camera and watch can shift kilometers, so the Skill asks the user to confirm one landmark photo.

Built With

• gpx
• markdown
• ugreen