-
UI
Inspiration
Finding one useful file can be difficult when documents are spread across many folders. A laptop warranty might be saved beside old receipts, manuals and several copies of the same document.
I wanted to make finding and organizing these files easier, while keeping the user in control of any changes.
What it does
VaultMind AI is a Skill designed for UGREEN AI NAS. Users choose which folders it can check and which folders it should leave alone.
They can ask questions about their documents, compare versions, review duplicate files and check for possible personal information. Answers should point back to the files they came from.
The main feature is the SafeAction Planner. Before moving anything, it shows the proposed changes and explains why they were suggested. The user chooses what to approve. It does not automatically delete duplicates.
How I built it
I used the official UGREEN Skill Developer Package to create the Skill and check its structure. The Skill uses Markdown instructions to coordinate UGREEN's file, document and knowledge agents.
I also built a separate local demo using Python, SQLite, HTML, CSS and JavaScript. It uses made-up sample documents to demonstrate search, duplicate review, privacy checks and approval steps. File moves in this demo are simulated; they do not change actual files.
I used Codex to help with research, implementation, testing and drafting the project materials.
Challenges I faced
The biggest challenge was matching the idea to what the platform actually supports. My original plan included a custom Python service, but the competition requires a Skill that works through UGREEN's supported agents. I adjusted the design around those requirements.
Another challenge was handling changes between preview and approval. If a source file changes or the destination already contains a file, the old plan should not be used.
What I learned
I learned that finding an answer and safely acting on it are two different problems. A useful assistant needs to explain its sources, respect folder boundaries and clearly report when something fails.
I also learned to separate what has been tested locally from what still needs testing on the actual device.
Progress so far
The Skill passes the official static checker with no errors or warnings. The local demo has 26 passing automated tests.
I have not tested the Skill on a real UGREEN NAS yet. Installation, AI search quality and the platform's handling of approved file changes still need device testing.
What's next
If selected for the next round, I will test the Skill on the provided NAS using sample files. I will focus on finding the correct documents, respecting excluded folders, checking approvals and recovering from interrupted operations.
