Inspiration
Families can keep years of photos while the stories behind them remain in separate conversations and recordings. Story Keeper connects those two collections and helps a family decide what to ask next.
What it does
Listen mode starts with one user-selected recording. The family reviews the transcript, confirms one event, reviews its matching photos, and approves a new album and story note. Ask mode checks a selected small album against approved story records, then drafts questions for photos with no recorded story link in that scope. The next answer can become the next recording. Suggested questions never count as completed stories.
How it is built
The initial Skill is Markdown orchestration for UGREEN AI NAS, using the supported file-agent, knowledge-agent and photo-agent capabilities. It works in bounded sets of at most twenty fully listed photos and preserves original media. A failed note save is reported as partial work without automatically repeating writes.
Current status and challenges
The initial workflow and supporting references are written and pass the official local static checker. They have not yet been installed or executed on a NAS. Device validation must establish real transcription, photo identity matching, processing permissions, album creation and verified note saves. The proposal and package distinguish planned test outcomes from observed results.
What I learned and what comes next
Designing against an actual capability catalog requires careful handling of incomplete results, approvals and partial writes. If selected, I will test Listen and Ask on the assigned NAS, record actual results, and address the documented interruption cases.
AI and components
Claude assisted with the entrant-provided proposal drafts/review. OpenAI Codex assisted with capability review, workflow and document revisions, packaging and static checks. No third-party executable, runtime library, model weight or media asset is bundled in the Skill ZIP.
Built With
- ai
- markdown
- ugreen
