Inspiration
After an event shoot, a photographer can be left with hundreds of nearly identical photos. Shooting is the fun part. Sitting for hours deciding which blurry, blinking or repeated frames to throw out is not. I wanted a Skill that does that tiring first pass, so photographers can get straight to editing the good shots.
What it does
Shoot Picker looks through a folder of event photos on your UGREEN AI NAS, sets aside the bad shots, sorts the rest into moments (ceremony, speeches, dancing), and gives you a short list of the best ones, with a reason for each. It never deletes an original, and nothing is moved until the photographer approves.
How we built it
Shoot Picker is one SKILL.md that coordinates four expert agents:
- Quality Agent: checks sharpness and closed eyes
- Duplicate Agent: finds repeats and burst shots
- Grouping Agent: sorts photos into moments using capture time and visual similarity
- Report Agent: writes a review report of top picks, with a short reason for each Everything runs locally on the NAS, so no photos leave it. I used Claude to help brainstorm and draft the proposal and Skill instructions. The idea and direction are mine.
Challenges we ran into
- Making the workflow clear enough that anyone can follow it from trigger to result.
- Using only capabilities the UGOS Skill scope supports, instead of inventing features.
- Planning for edge cases like RAW+JPEG pairs and files that can't be opened.
- Building without a NAS in Round 1, so real testing is still ahead.
Accomplishments that we're proud of
- A complete four-agent workflow, from picking a folder to a final report.
- A safe design: originals are never deleted, every rejection is only a suggestion, and nothing moves without approval.
- Clear failure handling, so one bad file never breaks the whole run.
- A full proposal and Skill package, written to fit the hackathon's limits.
What we learned
Designing a Skill is mostly about planning the steps and the failure cases, not just the happy path. Asking for approval before moving anything makes the tool feel safe to trust with someone's work.
What's next for Shoot Picker
Test on a real UGREEN AI NAS with a large event folder, tune the scoring rules, and polish the report layout.
Built With
- claude
- markdown
- ugos
- ugreen-ai-nas
- ugreen-skill-developer-package
