Inspiration
Creative professionals rarely deliver a single file. A typical client handoff bundles final exports, thumbnails, logo variants, captions, source assets, and approval records, scattered across folders with names like final_v3_revised2.mp4. Before every delivery, someone manually hunts these down, compares them against the brief, identifies gaps, and assembles a folder. It's repetitive, error-prone work that should never go wrong but regularly does — missing deliverables delay client reviews, trigger rework, and quietly damage a studio's reputation.
When we saw UGREEN's Track 03 — Unleash Your Creativity, helping creators search, tag, select, organize, and review media on AI NAS — we realized the NAS is the natural home for this workflow. The files already live there.
Plus, We ourselves have had a hands-on experience of media creativity and video creation and editting. So it's more of a personal thing than a "hackathon-only project".
What it does
StudioFlow AI is a UGREEN AI NAS Skill that turns creative-project handoff into a guided, evidence-linked, reviewable workflow:
Brief → Checklist → Discovery → Match → Readiness Report → Approved Handoff Folder
Each requirement from the brief is classified as one of four statuses:
- Found — a plausible candidate exists with sufficient evidence
- Needs review — candidates exist but version, approval, or content is uncertain
- Conflict — project sources disagree on what's required
- Not found in searched locations — no candidate within the authorized scope
The Skill never silently declares a project complete. Write actions (creating or populating a delivery folder) require explicit user approval, and originals are never deleted, moved, or overwritten.
How we built it
We designed StudioFlow AI as a single coordinated SKILL.md orchestrating five logical roles — Project Locator, Requirement Extractor, Asset Matcher, Handoff Reviewer, and Package Organizer — where each step consumes the actual output of the previous step. No vague references like "that directory from before."
The matching engine uses a conservative evidence hierarchy: exact filename reference in a checklist (strongest) → file type + folder context + metadata → format and path only → vague filename resemblance (weakest, never auto-promoted to "Found").
Challenges we ran into
Resisting scope creep — visual content verification, auto-share links, and batch audits all tempting to add. We deliberately scoped them out to keep the core workflow reliable.
Accomplishments that we're proud of
The four-status taxonomy (Found / Needs review / Conflict / Not found in searched locations). It forces honesty about what the Skill actually verified versus what it merely located — and it makes the difference between "I think this is done" and "here's exactly what's done, what's uncertain, and what's missing."
What we learned
Designing an AI Skill for a NAS forced us to think differently about uncertainty. A general chatbot can guess; a Skill that touches real client files cannot. We learned to surface conflicts instead of silently resolving them, to say "not found in searched locations" instead of "does not exist", and to treat approval as a hard gate rather than a polite suggestion.
What's next for StudioFlow AI
Well, the possible improvemenst ARE isted in the proposal. We will surely start scaling this project to new heights, and our selection as Finalists will serve as a booster of our productivity and interest towards this project.
