Inspiration
Public forum posts describe the problem. One user found that many photos on a NAS were duplicates or app thumbnails and wanted them separated. Another noted the same picture existing as an original and as a messaging-app copy. Others said text recognition in photo search is what they miss most.
What it does
Photo Library Declutter is a Skill for UGREEN AI NAS. It scans a photo scope that the user names and produces a suggestion report. The report separates exact duplicates, possible same-photo pairs, screenshots, and photographed documents from real photos. Each suggested item has a path, a group, and the reason it was suggested. The Skill never deletes, moves, renames, or overwrites a photo. After the user confirms, it can save the report as a Markdown file, create suggestion albums, and copy files into suggestion folders.
How I built it
The package contains Markdown files only. The Skill delegates to three registered experts: file-agent (ug-filemgr, ug-filesystem, file-workflow), photo-agent (ug-photo), and knowledge-agent (ug-knowledge). Saving, albums, and copies need the user's confirmation and platform approval. I used the UGOS Skill Creation Tool 0.7.7 from the Hackathon Starter Kit as a format and capability reference only.
Challenges I ran into
Recompressed copies are not exact duplicates, so an identical-file scan does not find them. To avoid false pairs, a possible pair is listed only when both images record a capture time, the times match to the second, and the dimensions are close. Group labels come from the library's recognition, so the Skill states that they are suggestions.
Accomplishments
Built: the full SKILL.md workflow, group rules, and disclosures, checked by hand against the published limits.
What I learned
A Skill is Markdown only and acts by delegating to registered expert agents. It cannot contain script code blocks.
What's next
Not built: any run on a NAS. Needs NAS testing: screenshot and document recall, metadata on messaging-app copies, and album behavior.
Disclosure
Claude (Anthropic) was used to draft the Skill text and the proposal.
Built With
- claude
- markdown
- ugos-skill-creation-tool
- ugreen-ai-nas
