Inspiration

Every photographer who shoots to a brief knows the feeling: the client sends a shot list with 30 or 40 "must-haves", you shoot 2,000 frames, and then you spend an hour scrolling to make sure the ring close-up and the photo with the grandparents actually exist. Worse, sometimes you only find out a shot is missing when the client asks for it — after delivery, when the moment is gone.

Photo tools are great at answering "find photos of X". None of them answer the question that actually matters before delivery: "what is NOT in my shoot?" A UGREEN AI NAS already holds both the client's brief and the photos, and it can read documents and understand images locally — so it felt like the right place to answer that question.

What it does

Shot-List Coverage is a UGREEN AI NAS Skill that:

  1. Reads the client brief (PDF, Word, text) stored on the NAS and extracts every required shot — people, objects, setting, lighting, and priority.
  2. Lets you approve or edit the extracted shot list before anything is searched.
  3. Searches only your shoot (a folder, album, or date range) using face groups, recognised objects, location, OCR text, and capture time.
  4. Visually verifies every candidate against the shot description, so a shot is never marked covered without visual evidence.
  5. Produces a Coverage Report — Covered, Weak (needs your eye), and Missing — with must-have gaps first and a reason for each match.
  6. Optionally, after your confirmation: builds one selects album per shot, tags verified files, and saves the report as Markdown on the NAS.

A "missing only" mode re-checks only the remaining gaps from an earlier report — so you can run it at a lunch break during an event and still have time to get the shot.

How we built it

The Skill follows the official Uliya Skill Development Standard and was built with the UGOS Skill Creation Tool (uliya-skill-kit 0.7.3):

  • It is pure Markdown orchestration: the parent Skill delegates to the NAS's registered experts — document-agent, photo-agent, file-agent, and knowledge-agent — and never calls the CLI directly.
  • Every step was mapped to a real capability in the kit's capability catalog: ug-document (qa, extract), ug-photo (search-picture, meta, vision-vlm, create-album), ug-filemgr (tags), and ug-knowledge (save-text).
  • Each delegation is self-contained, uses a fixed scope, and passes only real results from the previous step — no guessed paths or IDs.
  • Read steps run freely; anything that changes data (albums, tags, saving) waits for explicit confirmation and platform approval.
  • All five delegation states (completed, failed, cancelled, waiting_user, waiting_approval), empty results, and partial results are handled explicitly.
  • The package passes the official static checker with 0 errors and 0 warnings.

Everything runs on the NAS. No external services, no credentials, and original files are never moved, renamed, or deleted.

Challenges we ran into

  • Designing without hardware. Round 1 has no NAS access, so the workflow had to be designed carefully against the capability catalog alone, with testing planned for the finalist day.
  • Avoiding false confidence. Search results alone aren't proof a shot exists. Adding a visual verification step and a separate Weak status made the report honest instead of optimistic.
  • Never inventing data. Each step had to pass only real upstream results, and every "nothing found" path had to stop cleanly instead of substituting similar photos or silently widening the search.
  • Keeping it tight. Fitting a complete, safe workflow into a clear Skill, and the proposal into 500 words, forced a lot of useful simplification.

What we learned

  • How UGOS Skills orchestrate registered experts through delegation, and why approval boundaries matter for anything that changes user data.
  • That the most useful AI feature for creatives is often not generating content, but checking work against intent.
  • How much a clear empty-result and partial-result design improves trust in an AI workflow.

What's next

  • Test and tune search and verification accuracy on a real UGREEN AI NAS iDX6011 Pro.
  • Support video-specific shot lists (b-roll, interview angles) and multi-day event briefs.
  • Publish to UGREEN Skills Hub.

Built With

  • claude
Share this project:

Updates

Submission history