Inspiration
Updating a product specification takes minutes. Finding every place that repeats the old specification can take much longer.
A business might change a product’s advertised battery life from twelve hours to eight hours, update the main specification, and still have the old promise sitting in a brochure, support guide or campaign image. Those files may be stored together on a NAS, yet the relationships between their claims remain hidden.
I wanted to make that follow-up work easier. That led to ChangeLens, a UGREEN AI NAS Skill built around one question:
One fact changed. Which files now need updating?
What it does
ChangeLens is designed to review a user-selected collection of documents and still images against one approved product change.
The user confirms the authoritative change source, product, revision and effective date. ChangeLens then coordinates UGREEN’s experts to discover files, read document claims and inspect text embedded in images.
Its output is an evidence-backed review queue showing:
- Claims that need updating.
- Possible impacts that require human review.
- Claims already aligned with the change.
- Historical records or different product variants.
- Unreadable material and unresolved questions.
Each finding should include the observed claim, its source, why it matters and a recommended next action.
For example, an eight-hour battery specification for revision B should flag a revision B brochure still advertising twelve hours. It should also recognize that a revision A manual may legitimately retain the older value.
How I built it
I developed the early Skill using the supplied Uliya Skill Kit 0.7.3 and its public capability catalog.
The build is a Markdown orchestration workflow with supporting references for evidence rules, report structure and validation. It delegates NAS operations to UGREEN’s registered experts:
- File expert: inventories the authorized collection.
- Document expert: extracts and compares product claims.
- Photo expert: reads image text through supported OCR and visual analysis.
- Knowledge expert: optionally saves the report after user confirmation and platform approval.
The initial scope is deliberately bounded: one product change and up to 25 selected documents or images. Dependent tasks must receive actual results from earlier steps, and every candidate must appear in the coverage report.
I used ChatGPT/Codex to assist with research, drafting and independent structural inspection.
Challenges I faced
The hardest design challenge was defining when a mismatch actually requires an update.
An older number might belong to a different revision, region or historical record. File modification dates do not establish when a specification became effective. ChangeLens therefore requires applicability evidence before labeling a claim as outdated.
Another challenge was avoiding false reassurance. A search miss, unreadable image or partial document preview cannot establish that a file is correct. I created explicit rules for incomplete coverage and uncertain findings.
I also adapted the idea to the toolkit’s supported workflow. The early build coordinates registered experts rather than depending on custom executable code or an external backend.
What I learned
Useful change review requires more than finding matching words. It requires understanding which product a statement describes, when it applies and what evidence supports the conclusion.
I also learned to distinguish a completed expert task from a complete business result. A task may finish while still returning unreadable files or partial coverage. Those gaps must remain visible.
These lessons shaped ChangeLens around inspectable evidence and specific review actions.
Current progress
The early build includes SKILL.md, review rules, a report contract and planned validation scenarios. File structure and ZIP integrity have been independently checked.
ChangeLens has not yet been installed or tested on a UGREEN NAS. The uploaded toolkit code was not executed, and no official-checker pass or measured accuracy is claimed.
What’s next
If selected for the final round, I will test ChangeLens on the assigned NAS using fictional product materials.
Two core tests will check whether it detects outdated document and image claims, preserves legitimate historical claims, and recognizes manually corrected files on a rerun. An edge case will verify that an unreadable authoritative change source stops dependent analysis.
I will also validate source attribution, OCR quality, coverage reporting and approved report saving. My goal is a focused workflow that helps users act on a product change with clear evidence of what needs attention.
Built With
- ai
- kit
- markdown
- nas
- ocr
- skill
- ugos
- ugreen
- uliya
