Track: Unleash Your Creativity. Round 1 entry: the project proposal PDF and the pre-test Skill package (ZIP) are submitted on the hackathon page. The Skill has not been run on a NAS yet.
What does your Skill do?
It turns one event's photos into a review set: photos without anyone on a do-not-publish list go into a new album, and every other photo is set aside with a reason.
Who is your target user, and what is their problem?
The person who posts photos for a student club, class, or small team. Some members ask not to appear online. Before posting, this person checks hundreds of photos by eye for those faces, name tags, and screens. Missing one photo can expose someone who said no.
What is your solution?
The Skill uses the NAS's own face and text recognition to sort photos into three groups: Cleared, Excluded (a listed person was recognized), and Needs a human look (an unnamed face, readable personal text, or a failed check). Only Cleared photos go into the album.
What is the main user flow?
The user gives the event dates, the names to leave out, an album name, and a report folder. The Skill matches names to People albums, asks about any name it cannot match, sorts the photos, asks approval, creates the album, and saves a report listing every photo once with its reason.
How is your idea different from existing tools or common AI solutions?
Photo apps find photos that include a person. This Skill does the reverse and is built to fail safe: an unnamed face, a failed check, or an unmatched name never counts as cleared, and a failed person search blocks the album. Photos never leave the NAS.
Which UGREEN AI NAS features does your Skill use, and at which step?
People albums (step 1), photo search by date, folder, and figure (steps 2–4), OCR and the vision model (step 5), album creation (step 6), and knowledge save-text for the report (step 7).
What is your technical approach?
The parent Skill only delegates: ug-photo → photo-agent for steps 1–6 and ug-knowledge → knowledge-agent for step 7. Each step uses only results returned as completed. Album creation and saving need user approval. Video files are listed, not checked. No external service, script, or credential is used.
What have you built so far? What is not built yet? What still needs testing on a NAS?
Built: SKILL.md, report format, test cases, and disclosures; the package passes the bundled static checker with no errors or warnings. Not built: tuning of the text patterns. Needs NAS testing: figure search for unnamed faces, OCR on small text, and approval prompts.
How will you test your Skill if you become a Finalist?
Core 1: 40 photos, one listed test person in 6 of them; all 6 are excluded. Core 2: 20 photos, 4 with fake name tags or a fake address card; those 4 need a human look. Edge: a name with no People album returns waiting_user, and the report warns that face filtering was not possible for that person.
What are the known limits, risks, or outside dependencies of your Skill?
Face recognition and OCR can miss small, turned, or blurred faces and text, so the Skill never says a photo is safe; a person still decides. It depends on People albums being named. Test data will be synthetic, with consenting test subjects.
AI tools used
Claude Code (Anthropic) was used to propose ideas, draft the English text of this proposal and the Skill, and run the bundled static checker. The entrant approved the idea and submitted the text as drafted, without editing it.
Third-party components
None. The Skill contains no code or libraries.
Built With
- agent-skill
- claude-code
- nas
- photo-privacy
- ugos
Log in or sign up for Devpost to join the conversation.