Inspiration
This project comes from something that happened to me. I once uploaded files and accidentally included a scan of my CNIC. My personal details and location spread widely online, and I dealt with serious problems because of it.
The mistake was small: one file I didn't notice in a folder. The damage was not. Most of us only check what we are sharing after we share it, and by then it is too late. I wanted a tool that checks before you share, on your own device, so one missed file can't change your life.
What it does
Share Check is an AI Skill for UGREEN AI NAS. Before you share a folder or album, it scans the contents and flags:
- ID documents such as CNIC, passport and licence scans
- Financial papers such as bank statements and card images
- GPS location and device details hidden in photo metadata
- Faces that may belong to minors
It gives you a risk report and a safe-to-share plan, file by file: keep, strip metadata, exclude or blur. Nothing changes until you approve. Cleaning happens only on copies in a new "share-ready" folder, and your originals are never touched. A final re-scan confirms that no high-risk item remains.
How I built it
Share Check is written to the Uliya Skill Development Standard as a SKILL.md entry point, with Markdown reference files for the risk rules and the report format. The workflow is: validate the path, inventory files, read metadata, extract text from scans, detect faces, score risk, draft the plan, wait for approval, execute on copies, then re-scan.
Everything runs locally on the NAS. No external services are used and no data leaves the device. Failure handling covers unreadable scans, empty folders, missing permissions and very large folders.
Challenges I ran into
- Building without NAS access in Round 1, so the workflow is designed carefully but not yet tested on hardware.
- Deciding what counts as sensitive without being too strict or too loose. I added strictness levels and audience settings (family, friends, public link).
- Designing the approval gate so the Skill can never change files without the user's say.
- Staying inside the supported NAS capabilities instead of relying on custom scripts.
Accomplishments that I'm proud of
- Turning a painful personal experience into a practical safety tool.
- A complete workflow with an approval gate and a "copies only, never originals" rule.
- A clear risk-rule set and report format that make the Skill's decisions easy to understand.
- A privacy-first design: nothing leaves the NAS.
What I learned
Sensitive data hides in more places than the files themselves. A photo's metadata can reveal a home location, and a single scan in a big folder can expose an identity. I also learned that a safe tool is more about restraint than power: it should explain, ask for approval and work on copies.
What's next for Share Check
If selected as a Finalist, I will validate the Skill on the NAS using synthetic data only:
- Core test 1: a mixed folder with a fake ID scan and bank statement, checking that the risk report is correct.
- Core test 2: geotagged photos, checking that GPS is stripped on the copies only.
- Edge case: an empty folder or an unreadable scan, checking that it fails gracefully.
After that, I want to add more document types and languages, and support for scheduled checks before shared links go live. Accuracy depends on the NAS's OCR and face-detection capabilities, so Share Check reduces the risk of leaks but cannot guarantee it. The user always makes the final decision.
Built With
- claude
- markdown
- ugos
- ugreen-ai-nas
- uliya-skill
