Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for privacy-sweep
Inspiration
Shared NAS folders fill up with scans and downloads. Sooner or later a passport scan or a tax return is sitting where the whole household, or the whole office, can open it. Nobody checks for that by hand, and uploading those files to a cloud scanner is the opposite of what you want.
What it does
privacy-sweep asks you which folders to check, then has the NAS experts list the files, read the documents and images, and classify them: government ID or passport, tax form, medical record, bank statement, signed contract, other identity paperwork. You get a numbered report with evidence in words and a confidence level for each finding. Nothing personal is ever quoted or copied into the report. You confirm the real ones, and only then does the Skill tag them or move them into your private folder through the platform's approval flow. The report keeps the old paths so any move can be undone.
How we built it
One SKILL.md plus two reference files, written against the UGOS capability catalog in the Skill Developer Package. The parent Skill only delegates to the registered experts (file, document, photo and knowledge agents) in a fixed order; each step gets only the real result of the step before it. Reads need no approval; tagging, moving and saving the report go through platform approval where the catalog requires it. The package passes the bundled static checker with no diagnostics. Built with AI coding assistants under the entrant's direction; see references/disclosures.md in the package.
Challenges
The catalog has no way to tell whether an ordinary folder is shared, so the Skill never claims a folder is shared; it sweeps the folders you name and says so. There is no OCR action for documents, so scanned PDFs without a text layer are reported as not checked instead of guessed. Every instruction to an expert had to be self-contained and had to forbid returning document numbers, names or quotes.
What's next
Round 2 testing on a real UGREEN AI NAS with fabricated sample documents: classification accuracy, the approval prompts, and move verification.
Built With
- claude
- markdown
- ugos
- uliya-skill-kit
Log in or sign up for Devpost to join the conversation.