Inspiration

Not every user wants to tidy up the same way, and why tidy up later at all? A home NAS fills up with phone backups, downloads and scans: duplicates, blurry burst series, names that say nothing. Cleaning up afterwards feels risky, so nobody does it, and finding things again becomes the daily pain.

What it does

OctoFile is an orchestration Skill for UGREEN AI NAS that works like an octopus: every arm acts on its own, one head keeps them together.

  • Intake (core): the inbox works like a quarantine. New files pass your own gates and rules: exact duplicates go to a duplicates archive, picture sequences get uniform numbered names, photo series get a suggested best shot ("keep 1, 3, 4"), and every passing file gets its target folder, all with one approval. Unclear files stay in quarantine.
  • Find: photos by named people and place with a measured fallback chain, and answers from your own files with the passage that proves them ("Who is my current car insurance contact?").
  • Doc: summaries, comparisons, medical letters (diagnoses, medication, lab values over time), tax receipts by category, contracts with notice periods, and source-checked dossiers.
  • Collect: everything on a topic into one folder, as copies, as their new home or as a tag, and share-safe copies with personal data replaced.
  • Album: bind pictures into albums by topic or time and archive the originals instead of deleting them.

Every change waits for one approved package and ends with a receipt.

How I built it

Pure Markdown orchestration following the Uliya Skill Development Standard: one SKILL.md plus references for rules and starter profiles, receipts, setup and tests. Every step delegates to a registered NAS expert (file, photo, document and knowledge agents). Downstream steps use only real upstream results, models run on the device, and no outside service is used. The package passes the kit's static checker without diagnostics.

Challenges

Mapping the ideas honestly onto the capability catalog. What the catalog cannot do, such as drawn frames on pictures, PDF photo books, PDF redaction or a trigger on file arrival, is documented as a drop-out instead of being promised.

What's next

On the NAS test day: measure duplicate-scan behaviour, album behaviour after archiving, the medical-report fields and the runtime per delegation, then tune limits and rules.

Disclosures

  • AI tools: Claude Code (Anthropic) helped turn the author's ideas and wording into English text, structure the Skill files and run the kit's static checker. The author directed the work and reviewed the proposal PDF.
  • Third-party components and external services: none. Built with the UGREEN Skill Developer Package 0.7.3.
  • Prior work: The idea builds on the author's earlier concept for purpose-based file organization. The Skill and its core workflow were newly developed during the Hackathon, starting 2 November 2026.

Built With

Share this project:

Updates

Submission history