1. What does the Skill do? (one sentence) It turns one photo shoot stored on UGREEN AI NAS into a shortlist of sharp, distinct portfolio candidates - never modifying, moving, or deleting the originals; it only adds a new album and a written report.
2. Who is the target user and problem? Photographers who return from a shoot with hundreds of frames and lose hours culling near-duplicates and soft shots before editing. Their photos already live on the NAS, but culling there is still manual.
3. What is your proposed solution? An AI Skill that scans a user-selected shoot folder, groups near-duplicate frames, assesses each frame's quality, and produces a ranked shortlist with a reason for every pick - delivered as a new album plus a written report, with the originals untouched.
4. What is the Core User Flow? User points the Skill at a shoot folder and sets a target count -> Skill catalogs the photos -> groups bursts and near-duplicates -> assesses frame quality -> selects a diverse shortlist -> asks for confirmation -> creates the picks album and a Markdown report. No deletion step exists.
5. What substantive difference does it offer? It culls where the photos already live, keeps the originals untouched by design, and explains each selection instead of returning a silent list.
6. Which UGREEN AI NAS capabilities are used, and at which step? Albums (ug-photo): find the shoot's photos and create the final shortlist album. Files (ug-filesystem): save the Markdown report in the shoot folder. On-device visual QA plus EXIF capture metadata (ug-photo): burst grouping and per-frame quality assessment. All are documented in the official creation tool's capability catalog, and the package passes its static check.
7. What is your technical approach? SKILL.md orchestrates the NAS's registered photo and file experts over documented UGOS capabilities: (1) find the shoot's photos; (2) group bursts and near-duplicates using capture-time metadata and visual comparison; (3) assess each frame with on-device visual QA for sharpness, exposure, and subject prominence; (4) a diversity pass so picks are distinct; (5) after user confirmation, create the picks album and save a Markdown report. Inputs: folder path and target count. All analysis runs on the NAS; no external services.
8. What is the current development status? Concept and Skill workflow are defined; the SKILL.md package is written to the Uliya standard and passes the official creation tool's static check. Nothing is hardware-validated yet - Round 1 provides no NAS access - so runtime behavior is pending NAS validation if selected.
9. What is your final validation plan? Core test 1: a mixed-quality shoot produces a shortlist matching a manual cull (sharp, distinct frames). Core test 2: a burst sequence contributes only its best frame. Edge case: a folder smaller than the target count, or frames that are all near-identical, returns an honest short list instead of padding.
10. What are the known limitations? Pick quality depends on the on-device visual assessment's judgment. RAW files may rely on embedded previews. The read-only design delivers picks as an album and a report, not reorganized folders. No cloud-service fallback is planned, by design.
Built With
- ugos
- ugreen-ai-nas
- uliya-skill
