Inspiration
The client signs a shot list. The card holds thousands of frames. The gap shows up weeks later, when the client asks for a shot that was never taken.
What it does
- Reads the shot list into numbered items.
- Searches the import by content and named faces.
- Asks a vision model two questions per frame: does it show this item? What part of it is absent?
- Searches again by parts (people, objects, moment) before any item is called missing.
- Reports missing items first, then weak, then covered.
- Tags verified frames in place, copies confirmed picks to a proofing folder, builds an album.
How we built it
UGOS Skill Creation Tool. Delegation only: document-agent, photo-agent, file-agent, knowledge-agent. Real paths and resultRefs move between steps. Writes wait for approval. Static check: zero errors, zero warnings.
Challenges we ran into
Mood items such as "candid feel" cannot be checked by a camera. They are listed, never scored.
What we learned
Culling tools rank frames. A shot list is a promise. Auditing the promise is a different job.
What's next
On a finalist NAS: recall of search-screen for shot wording, yes or no accuracy, RAW handling, condition-album timing.
Built With
- claude
- markdown
- ugos
- uliya-skill-kit
