-
This image is currently processing. Please be patient...
What does the Skill do?
It builds a curated album of the best badminton match or training photos from a date range, after the user confirms the shortlist.
Target user and problem
The author runs a badminton concept store in Taiwan and operates a badminton AI coaching app; match photos from the store and its players pile up constantly, and this Skill solves a problem the author faces every day. Typical users are players, parents and coaches with many badminton photos on the NAS. Finding the usable ones among blurry or off-topic shots by hand is slow, so they are rarely organized.
Proposed solution
The user states a date range and album name. The Skill searches the photo library, checks each candidate visually, presents a shortlist, and creates an album only after approval.
Core user flow
Say the request, provide dates and album name, review the shortlist with rejected count, confirm, receive the album name and photo count. Originals are never modified.
Difference from existing tools
Generic album tools group by time or faces and cannot judge whether a photo is a usable badminton moment. This Skill adds a sport-specific visual check, a mandatory human approval point, and strict non-destructive behavior, all on the NAS.
NAS capabilities and steps
ug-photo search-picture finds candidates by time or location (step 1). ug-photo vision-vlm checks each candidate on the local model (step 2). ug-photo create-album writes the album after confirmation (step 3).
Technical approach
A single photo-agent is orchestrated through three sequential delegations; each step receives only the verified result of the previous step. Inputs are a date range or folder, album name, and an optional person or location. The only state change is album creation, which requires explicit user approval. Analysis uses the NAS local model, so no photo leaves the device. The package contains only SKILL.md and two linked reference files, with no scripts or credentials.
Current development status
Implemented: the initial Skill package, which passes the official static checker. Not implemented: tuning of the visual-check prompt and batch sizes. Pending NAS validation if selected as a Finalist: actual command parameters, expert responses, and approval behavior.
Final validation plan
Test 1: a date range containing mixed badminton and non-badminton photos; verify the album contains only accepted photos and originals are unchanged. Test 2: user rejects part of the shortlist; verify only the confirmed photos are added. Edge case: a range with zero photos; verify the Skill reports no results and stops without creating an album.
Known limitations
The visual check may misjudge unusual angles or lighting. Behavior depends on the NAS photo index being complete and the local vision model being available. Videos are out of scope. Nothing has been validated on a real NAS yet. No third-party components or external services are used.
Built With
- ai-nas
- badminton
- skill
- ugos
- ugreen
