Inspiration
A family photo archive holds questions that only its elders can answer. Face recognition can group a stranger across hundreds of photos, but it cannot say who she was. Most tools ask the owner to label things. This project flips that: the archive asks the questions.
What it does
Ask Grandma ranks what a family's NAS archive cannot explain (recurring unnamed faces, undated scans, unknown places, unexplained names) and builds a short photo interview for an elder. A recording of the session is then filed as quotes with timestamps, checked against the archive, and used to write better next questions. If Grandma says 1966 and a letter says 1964, both stay and the gap becomes a follow-up. Narrators can withhold anything.
How it was built
It is a Markdown-only Uliya Skill: SKILL.md plus seven references, delegating to registered experts (photo, document, knowledge, and file agents). It uses no scripts and calls no external service, and it never deletes, moves, or changes the timestamps of originals.
Claude Code (Anthropic) generated the three candidate ideas, selected the core concept, and named it Ask Grandma. It also wrote the proposal answers, SKILL.md, every reference file, and this text, and it ran the official checker and the word count.
Challenges
No NAS is available in Round 1, so the design has to state what it does not know. The capability map lists eight open questions for the finalist day, including unnamed-group fields, transcript timestamps, and which model serves each vision action.
What we learned
The headings in the Skill Kit template did not pass the checker in the same kit, so the package uses the headings that the checker accepts.
Status
The package passes the official static checker (uliya-skill-kit 0.7.3). It has not run on a NAS. The walkthrough in the references is fictional and is not a test result.
Built With
- claude-code
- markdown
- uliya-skill-kit
