Safe Share

Safe Share checks a photo, album, folder, or document before it leaves your NAS, shows the private data it carries, and stages risky files for review so nothing is shared without your confirmation.

Comment

Safe Share UGREEN Hackathon 2026 on AI NAS, Round 1 Proposal

Skill name: safe-share Track: Open Innovation Entrant: Taiwo John (solo) Deadline: Sunday, November 15, 2026, 11:45 pm PT (Monday, November 16, 2026, 8:45 am WAT) Ten substantive answers: 456 words (limit 500).

  1. What does your Skill do? (one sentence)

Safe Share checks a photo, album, folder, or document before it leaves your NAS, shows the private data it carries, and stages risky files for review so nothing is shared without your confirmation.

  1. Who is your target user, and what is their problem?

Anyone who shares photos or documents from their NAS: parents posting family photos, freelancers sending contracts, small teams. Every photo carries hidden location and device data, and every document can carry personal details, so sharing often leaks more than the sender meant to share.

  1. What is your solution?

Safe Share reads what each file carries before it is shared, then gives the owner a plain-language report, copies risky files into a review folder, and asks for confirmation before anything leaves the NAS.

  1. What is the main user flow?

The owner picks the files and the destination. Safe Share finds them, reads their metadata and content, and reports each file, its private data, and a verdict: safe, review, or do not share. Risky files go to a review folder, and the owner confirms before anything is shared.

  1. How is your idea different from existing tools or common AI solutions?

Most sharing tools trust the file and send it. Safe Share inspects a file for what it would leak, in plain language, on the device where the file lives, before sharing happens. The check runs locally, so files are never uploaded to a cloud service to be checked.

  1. Which UGREEN AI NAS features does your Skill use, and at which step?

Find: ug-filesystem find and ug-photo search-picture. Read: ug-photo meta reads photo EXIF, ug-filemgr mediainfo reads media metadata, ug-document metadata and ug-document extract read document data and content. Report: ug-knowledge save-text. Stage: ug-filesystem mkdir and cpmv copy risky files to a review folder after confirmation. Share: only after confirmation.

  1. What is your technical approach?

SKILL.md is the entry point. The parent workflow collects the files and the destination, then delegates to file-agent, photo-agent, document-agent, and knowledge-agent for finding, metadata, extraction, and the report. Each step uses only the real result of the previous step, and every state-changing action needs user confirmation. The local Uliya model reads the metadata and extracted text for personal details. No scripts, executables, credentials, or external services.

  1. What have you built so far? What is not built yet? What still needs testing on a NAS?

Built: this proposal and the SKILL.md package, written against catalog 2.0.0 and checked with the kit checker, static pass with 0 errors and 0 warnings. Not built: no runtime testing, no NAS is available in Round 1. Needs NAS testing: ug-photo meta returning GPS for real photos, extract coverage, and the confirmation flow.

  1. How will you test your Skill if you become a Finalist? Name two core test cases and one edge case.

Core 1: twelve birthday photos, three with home GPS, named in the report. Core 2: a contract with bank details, named and staged. Edge: a photo with no metadata, the Skill says so and stops.

  1. What are the known limits, risks, or outside dependencies of your Skill?

Known limit: the Skill does not strip EXIF data or redact text, because the catalog has no metadata-write or redact action. It reports exactly what to remove, and the owner removes it before sharing. The local model is small, so the Skill checks short concrete fields.

Third-party components (excluded from the 500-word limit)

  • UGREEN Skill Developer Package (uliya-skill-kit 0.7.3, provided by UGREEN for this hackathon), used to build the Skill to the Uliya Skill Development Standard. Source: https://ai.ugreen.com/pages/hackathon-starter-kit
  • UGOS CLI capabilities and the Uliya local model (UGREEN), used as the Skill runtime. Source: the capability catalog in the developer package, version 2.0.0.
  • No open-source libraries are used in the Skill package itself, which is Markdown only.
  • AI tools used: Muse, Meta's personal AI assistant, helped draft the proposal text and the SKILL.md file. The Skill itself runs on the NAS local model.

Built With

  • developer
  • skill
  • ugos
  • ugreen
  • uliya
Share this project:

Updates

Submission history