Inspiration

As an international student, I lived in many apartments across five schools and had several deposit disputes. In Texas, one landlord kept my deposit until I mentioned hiring a lawyer. The photos I needed were buried among pictures of other homes.

What it does

Deposift (deposit + sift) is a Skill for Uliya, the AI assistant on UGREEN NAS. It rebuilds a renter's security deposit evidence from the records that phone backup already put on the NAS. Move in inspection apps help only the people who photographed every room inside the app when they moved in, and evidence organizers start after the renter has already found and uploaded the right files. Finding them is the hard part.

The user names the home and the charge. The Skill reads the lease for the dates and finds the photos taken around move in and move out in a library that mixes several homes. It checks each one for the disputed spot and adds the landlord's messages, calls, and the deposit clause. After the user confirms, it copies the files into an evidence folder, compares each copy with its original by SHA-256, and saves a dated summary in which every line names its source file. Originals are never changed.

How I built it

I built it with the official UGREEN Skill Developer Package. The Skill is one SKILL.md file with nine delegated instructions to four expert agents, and every step uses an Action listed in the kit's capability catalog. The AI tools used during development are disclosed inside the Skill package.

Challenges I ran into

A date search returns photos from any place the renter visited in that period. The Skill treats each photo as a candidate until the renter confirms it.

Each instruction needs dates and file paths that exist only at run time. Each one ends with a label, and the real values from earlier steps are added after it when the Skill runs.

The catalog has no email search, so email counts only as saved files or screenshots.

Accomplishments that I'm proud of

The SKILL.md is complete and passes the kit checker with 0 errors and 0 warnings. All nine instructions also pass the checker after they are filled with synthetic dates and file paths.

What I learned

Most of the design work was deciding what to leave to the renter. Whether two photos show the same damage and who is responsible are the renter's decisions, so the Skill reports dates, sources, and what is visible.

What's next for Deposift

The Skill has not run on a NAS yet. The next step is to run two core tests and one edge case on a UGREEN AI NAS with synthetic data, then set the image batch size and the photo date window from what those runs show.

Built With

  • ai
  • markdown
  • nas
  • ugos
  • ugreen
  • uliya
  • uliya-skill-kit
Share this project:

Updates

Submission history