Consent Change Review

Inspiration

A small production team has its videos scheduled for release. Then a consent-change notice arrives. Which of those planned uses need another look? To find out, the team must compare the notice with its release plan, checking the relevant assets, dates and channels rather than assuming every use is affected.

What it does

Consent Change Review is a proposed read-only Skill for UGREEN AI NAS. It is designed to compare two files selected by the user: one consent-change notice and one planned-use manifest (a list of scheduled media uses).

For each assessed plan row, the intended output gives one of three results: review required, no match within the selected notice's scope, or indeterminate. Where evidence is sufficient, it includes the reason, quotations from both files and a question for human review.

A non-match does not clear a use for publication. Missing evidence, uncertain asset relationships and incomplete coverage stay visible. The Skill neither decides legal permission nor edits or blocks content.

How it was built

The v03 Markdown workflow was authored in Codex following UGREEN's Skill kit 0.7.3 guidance and capability catalog 2.0.0.

It is designed to use file-agent (ug-filesystem) to identify and fingerprint the two selected files, document-agent (ug-document) to compare their contents, and file-agent again to check whether the files have changed. If a source cannot be tied back to the original selection, dependent steps stop.

The initial scope is intentionally small: two synthetic, non-sensitive UTF-8 text files, each up to 128 KiB, and no more than 50 planned-use rows. The package adds no executable scripts, third-party API integrations or file-writing actions.

ChatGPT and Codex assisted with research, drafting, test planning and local checks.

Challenges

Three details shaped the rules.

A notice's issue date is not necessarily its effective date. A renamed or cropped asset is not automatically covered by the same terms. And a partial document read can look deceptively complete.

The workflow is designed to leave these cases unresolved unless the selected sources provide enough evidence.

What makes it different

An ordinary AI chat can compare two documents. This approach is designed to bind the review to two explicitly selected NAS files, check their fingerprints before and after reading, and require traceable evidence for each finding.

The intended advantage is a more controlled review process, not a claim that the AI understands consent law. Whether it improves review quality or saves time compared with an equally equipped ordinary AI prompt remains to be tested.

What I learned

The result that needs the most care is "no match." Without a clear reason and supporting evidence, it could easily be mistaken for permission to publish.

That shaped the decision to show the relevant source quotations, identify what was checked and leave uncertain cases unresolved.

Current status and next steps

The initial Skill package, proposal and validation plan are prepared. The recorded static-check result was "static-pass-needs-target-validation": zero errors and three dynamic-binding warnings.

Installation and operation on UGREEN AI NAS, the processing route and real user benefit have not been verified.

If selected as a finalist, the plan is to test a use after the effective date, a use after the issue date but before effectiveness, and an unsupported derivative link. The results and review effort will be compared with manual checking and an equally equipped ordinary AI prompt.