1. What does your Skill do?

It turns a small business owner's overnight voicemails and call notes, stored on the NAS, into one morning callback list with each caller's name, number and need.

2. Who is the target user, and what is their problem?

Owners of small trades shops, like roofers, plumbers and HVAC techs. After-hours and on-the-job calls land in voicemail or scattered notes. Each morning they replay every message and write down who called. Callers who wait too long hire someone else.

3. What is your solution?

The owner keeps voicemail recordings and call notes in one NAS folder. They say "Make my callback list." The Skill finds the new files, transcribes the recordings, pulls out each caller's details and saves one dated callback list back to the NAS. Nothing leaves the NAS unless the owner asks.

4. What is the main user flow?

Owner says "Make my callback list for last night." The Skill finds new files, transcribes audio, reads notes, extracts name, number, need and time, flags urgent words like "leak" or "no heat", and saves the list. It shows the caller count, urgent count and saved path, and asks before archiving handled files.

5. How is your idea different?

Phone apps transcribe one voicemail at a time and keep audio in the provider's cloud. This Skill batches a night of mixed voicemails and notes into one ranked list, urgent jobs first, writes "not given" instead of guessing a missing number, and keeps customer recordings on the owner's NAS.

6. Which UGREEN AI NAS features does your Skill use, and at which step?

Files: find new files in the calls folder (step 1) and archive handled files after confirmation (final step). Knowledge: transcribe recordings and save the callback list as text (steps 2 and 5). Documents: read text notes and extract caller details (steps 3 and 4).

7. What is your technical approach?

SKILL.md defines one orchestration that delegates to file-agent, knowledge-agent and document-agent: Files finds inputs by folder and date; Knowledge transcribes each audio file; Documents extracts fields from each transcript and note; Knowledge saves the list. Required input: the calls folder path and a date range. Permissions: read the calls folder, write a callback-lists folder, move only after approval. AI models: only the UGREEN-provided knowledge-agent and document-agent; no custom or third-party model. Recordings and lists stay on the NAS. If a field is missing, the list says "not given."

8. What have you built so far? What is not built yet?

Built: the full SKILL.md (trigger, inputs, three-expert orchestration, parent urgency ranking, confirmation, failure and empty-result handling) and labeled synthetic sample test files. It passes the Skill Developer Package checker with zero errors or warnings. Round 1 was built from that package without a personal NAS. Pending on the assigned Finalist NAS: transcription accuracy on real voicemail audio and file-move permissions.

9. How will you test it as a Finalist?

Core 1: five voicemails and two text notes from one night produce one list with all seven callers. Core 2: an urgent message ("water coming through the ceiling") lands at the top. Edge: a voicemail with no number spoken shows "number not given" and is not dropped.

10. Known limits, risks or outside dependencies?

Transcription can mishear names and numbers in noisy audio, so the list marks low-confidence fields. The Skill does not pull calls from a phone service; the owner or answering service saves files to the folder. It never calls or texts anyone.

Third-party components

None beyond UGREEN-provided capabilities.

AI tools used

An AI assistant (Grok Bot) helped draft and edit the proposal, SKILL.md, and the sample test files.

Built With

Share this project:

Updates

Submission history