Inspiration
I take screenshots instead of notes: error dialogs, settings, serial numbers, order confirmations. Weeks later, when the same fault comes back, I know I screenshotted it, but I can't find it. No search reads inside images. I had already solved a similar problem for my clipboard with a small local desktop tool (typed filters, search, recoverable trash), and wanted the same for the screenshot pile on a NAS.
What it does
snapit is a Uliya Skill with three modes:
- Process: every new screenshot gets a short text note (title, summary, key facts, recognized text) and one type tag: error, code-config, receipt-order, chat, web-article or other. Runs in batches of 20 with an interim report.
- Ask: "where was the printer error I screenshotted?" is answered from the notes, always with the path of the original screenshot. No guessing, no general knowledge.
- Tidy: duplicates and screenshots with an expired date (tickets, deliveries, vouchers) are proposed and moved into a recoverable trash folder after confirmation. Nothing is deleted.
Screenshots that show passwords, one-time codes or banking data get only a sensitive tag and no note. Everything runs on the local model of the NAS.
How I built it
With the official uliya-skill-kit 0.7.3. I checked every step against the capability catalog, mapped each CLI to its single registered expert (photo-agent, file-agent, document-agent, knowledge-agent) and wrote a pure orchestration SKILL.md: self-contained delegations, only real upstream results passed on, approval for every write, and explicit handling of completed, failed, cancelled, waiting_user and waiting_approval. The package passes the kit's static check with zero errors and zero warnings.
Challenges
Making runs repeatable without a database: a screenshot only counts as processed when its note is verified as saved and its tags are set, so an interrupted run picks it up again. And deciding what must never end up in a text file.
What's next
Testing on a real AI NAS: path hand-over between photo search and tagging, runtime per batch, and coverage of the semantic lookup across all note folders.
Built With
- claude
- markdown
- ocr
- ugos
- uliya
- uliya-skill-kit
