AI B-Roll Scout

Find the Right Footage Without Searching Every Video

Inspiration

As someone interested in AI development and creative technology, I recognized how time-consuming it can be for video editors, filmmakers, and content creators to search through large video libraries. Finding a particular recording or relevant footage can require manually browsing folders and opening numerous files.

I wanted to explore how UGREEN NAS capabilities could make this process easier. This led me to develop AI B-Roll Scout, a Skill designed to make footage discovery faster, more accessible, and more traceable.

What it does

AI B-Roll Scout helps users discover candidate footage stored in their UGREEN NAS video libraries.

I built its workflow around four documented capabilities:

• Video Search: ug-video search locates indexed video content.
• Subtitle Search: ug-video subtitle-search searches available subtitle content.
• File Discovery: ug-filesystem find locates relevant files.
• Media Inspection: ug-filemgr mediainfo retrieves available media information.

The Skill delegates these operations to UGREEN's registered media-agent and file-agent. It presents actual results, including filenames, paths, subtitle details, and available metadata, allowing users to inspect their original media.

How we built it

I developed AI B-Roll Scout as a Uliya Skill using the official UGREEN Skill Developer Package and its documented capability catalog.

I designed the parent SKILL.md to define the assistant's role, gather the user's search request or required NAS path, and delegate each operation to the appropriate registered expert.

I kept the workflow read-only and traceable. It uses actual outputs from NAS operations and accounts for completed, empty, partial, failed, cancelled, and waiting states instead of inventing results.

The package contains the Skill instructions and a supporting user guide. It does not require a separately bundled AI model or executable dependency.

Challenges we ran into

One of my biggest challenges as a solo developer was determining which features were genuinely supported by UGREEN NAS rather than assuming that every AI capability I envisioned was available.

I initially wanted to include visual scene recognition, timestamp-based discovery, and similarity ranking. However, these capabilities were not established by the reviewed UGREEN catalog. I had to narrow the initial scope to supported operations and clearly distinguish indexed search results from verified visual content.

Another challenge was validating the Skill without having physical NAS hardware. I used UGREEN's supplied static checker to validate the package structure, while keeping actual NAS execution and search-quality testing for the assigned finalist testing environment.

Accomplishments that we're proud of

I'm proud to have independently developed the Phase 1 Skill and its supporting documentation around four documented UGREEN NAS capabilities.

The package passed the supplied UGREEN static checker with zero errors, warnings, and informational diagnostics. I also verified the extracted ZIP and checked that the documentation, capability mappings, and proposal consistently describe the same implementation.

Most importantly, I established a practical foundation that prioritizes traceable results and technical accuracy instead of claiming visual AI features that I have not yet implemented or validated.

What we learned

Through this project, I learned that developing a useful NAS Skill involves more than connecting an AI assistant to file searches. It requires understanding the platform's actual capabilities, organizing a reliable workflow, and ensuring that every result is grounded in real system output.

Working independently also taught me the importance of managing scope, validating each component, and distinguishing between implemented features, static validation, and functionality that still requires testing on actual hardware.

What's next for AI B-Roll Scout

My next step is to validate the Skill on the assigned UGREEN AI NAS. I plan to test natural-language searches, subtitle searches, media inspection, permissions, and the handling of empty or inaccessible results.

For future development, I envision expanding AI B-Roll Scout with visual scene recognition, keyframe extraction, visual similarity ranking, timestamp-based discovery, and optional clip export.

These enhancements represent my long-term direction for the project, not features currently claimed by Phase 1.

My goal is to make footage discovery faster, more transparent, and more reliable, eventually helping creative professionals find not only the right video files but also the right moments within their footage.

Built With

• ai
• markdown
• skillsugreen
• ugos
• ugreen
• uliya