Inspiration
I build AI agents for my own work, and their work is merged only after I approve it. Family recordings and photos cannot be replaced, so this Skill follows the same rule: it shows what it found and changes nothing until the user approves.
The UGREEN AI NAS can already transcribe audio, find photos by face, place and date, and create albums, but nothing connects a recording to the photos of the same day. Voice memos from a trip, a birthday dinner, a parent telling old stories or a digitized cassette sit in a folder as REC_0142.m4a, and nobody plays them again.
What it does
Give it one recordings folder and a story title. It transcribes each recording and shows what it heard: the people, places and periods the speakers mention, each with the words it came from. From these it proposes a photo search by date, named faces and places, which the user confirms or edits once. A preview follows, and only after the user's approval does it create one new album and save one Markdown story: for each recording a short summary, three to six verbatim quotes with their time in the recording, the matched photos and the path of the original audio. It never edits, moves or deletes recordings or photos, and it never makes up names, dates or places.
UGOS Voice Memos transcribes and summarizes a recording, and AI Album search and the official ug-photo-album-create Skill start from what the user types. Voice Story Album starts from what the recordings say. A general AI chat cannot see the photo library and may add things nobody said; here every quote is checked against the transcript.
How we built it
Voice Story Album is one orchestration-only SKILL.md made with the official UGOS Skill Creation Tool (Uliya Skill Kit 0.7.3): nine steps and seven delegations to file-agent, knowledge-agent and photo-agent, with no scripts, credentials or outside services. The user is asked twice, to confirm the plan and at the preview. The Skill writes the summaries itself from the transcripts and checks every quote against them, so only creating the album and saving the story change the NAS, each through the expert's own approval. The package adds a user guide, a test plan and disclosures, and passes the kit's static checker with 0 errors and 0 warnings. I used Claude Code (Anthropic) to design and draft the Skill and this text and to run the checker, and I reviewed the result.
Challenges we ran into
Round 1 has no NAS, so every step was checked against the kit's capability catalog instead of a real device. A digitized tape carries the date it was copied, not the date it was recorded, so the Skill proposes the period the speaker talks about, or a named person, instead. After a review the flow was cut to two checks and two approvals, because a long flow is more likely to break on the single test day.
What's next for Voice Story Album
If selected as a Finalist: run the test plan on the assigned iDX6011 Pro with recordings and photos made for testing (trip voice memos, a digitized tape with a wrong file date, and damaged, silent and unmatched files), fix what the NAS shows, run the checker again and record the demo.
Built With
- claude-code
- markdown
- ugos
- ugreen-ai-nas
- uliya-skill-kit
