Inspiration
Most of a family's memories live in video, and video is the part of an archive a blind person is most shut out of. A photo can be described once and tagged. A video is minutes of moving scenes, and a file called IMG_4092 tells you nothing. In practice, a blind person who wants to revisit a memory has to ask a relative to find the clip and narrate it.
We have built voice-first tools meant to work for blind users with the screen off, and that work showed us how often accessibility stops at menus and labels while the actual content stays out of reach. A NAS is where a family's archive already sits, so it is the natural place to fix this. ToldAgain started from one request a blind user should be able to make on their own: "Find the video where my sister was laughing in the kitchen."
What it does
ToldAgain lets a blind user ask for a family memory in plain language. The Skill finds candidate videos on the NAS, collects what the NAS can actually tell it about each one (who appears, when and where it was recorded, what is said), and returns a short text description that a screen reader can read aloud. If the user approves, it tags the video so it is easy to find next time.
How we built it
ToldAgain is a Skill package for UGREEN AI NAS, written to the Uliya Skill Development Standard. SKILL.md is the single entry point, and it describes the whole flow from trigger to result.
Expert agents for files, media and knowledge run in sequence, and each step uses the real output of the one before it. The Skill handles every outcome explicitly: completed, empty, partial, failed, cancelled, and waiting for approval. Nothing is tagged or saved without the user's approval.
This is the initial Skill package for Round 1. It has not yet been run on a UGREEN AI NAS. Validation on the NAS is planned if we are selected as a finalist.
Challenges we ran into
Not making things up. A confident but wrong description is worse than none. The Skill reports only what the NAS returned, and says so when a transcript or file is missing instead of guessing.
Staying inside supported capabilities. Custom scripts are not allowed, so every step has to map to a capability the NAS actually supports. That meant cutting ideas that sounded good but could not be built this way.
Accomplishments that we're proud of
We designed a complete flow that handles failure as carefully as success, so a user always hears a clear outcome and never a silent dead end. We also kept the approval step in the flow, so a blind user stays in control of what gets saved to their own archive.
What we learned
Writing for a screen reader changes what a good answer looks like: short, ordered, and honest about what is unknown. We also learned that designing the empty, partial and failed paths first made the happy path simpler.
What's next for ToldAgain
If we are selected as a finalist, we will install ToldAgain on the assigned NAS, test it against real family video archives, and refine the descriptions based on how well they serve a blind user. After that, we want to extend it from videos to photos, voice notes and scanned documents, so a whole family archive can be explored without sight.
Built With
- accessibility
- markdown
- screen-reader
- skill-md
- ugos
- ugreen-ai-nas
- uliya-skills
