NAS Knowledge Inbox — UGREEN NAS Skill Proposal
Entrant: CHEN,MIN-YU · Track: Automate Your Workflow · Skill: nas-knowledge-inbox
1. What the Skill does
It turns a NAS inbox into sentence-level Markdown knowledge files, pre-categorized against the user's existing category tree, and answers questions from that library with sources.
2. Target user and problem
People who collect PDFs, voice notes, and screenshots into a personal knowledge base spend hours transcribing, splitting, and filing them. Cloud note AIs require uploading private research.
3. Solution
One request ("process my knowledge inbox") converts documents, audio, and images already on the NAS to text, splits sentences, drops library duplicates, pre-categorizes, and writes category_source.md files.
4. Core user flow
User names the inbox and knowledge folders → Skill lists files and reads categories from existing Markdown filenames (the part before the underscore) → converts files to text → splits, dedupes, and proposes categories → user reviews the draft → after confirmation, files are written. Questions are answered with sources.
5. Differentiation
Cloud tools need uploads and a live network. This Skill starts after NAS-native downloaders have filled the inbox (the kit has no web-download action), keeps files on the device, writes nothing without approval, and uses a filename convention a later local front end can read.
6. NAS capabilities by step
Inventory and exact-file scan: ug-filesystem, ug-filemgr (file-agent). Text conversion: ug-document extract, ug-knowledge audio-transcription, ug-photo vision-ocr. Library dedupe and sourced answers: ug-knowledge collect-evidence. Drafts: ug-document. Confirmed writes: cli.invoke --cli ug-filesystem --action mkdir and --action echo (gateway approval), or cli.invoke --cli ug-knowledge --action save-text (workflow approval).
7. Technical approach
A Markdown orchestration Skill built with the official uliya-skill-kit. The parent uses only delegate to registered experts in a fixed order; each step receives only the real result of the previous one. Inputs: inbox folder, knowledge folder, optional language and question. Review steps are read-only; writes need confirmation and never overwrite or delete. Models are those built into the NAS experts; whether they run fully on-device must be verified on test day. All five delegation states are handled.
8. Development status
Built: SKILL.md and two reference files, passing the official static checker with zero errors. Not built: web download (absent from the catalog; NAS-native tools do this outside the Skill); a card-swiping front end; scheduling. Pending NAS validation: extract, transcription, and OCR quality; sentence boundaries; collect-evidence as a near-duplicate check; pre-categorization; save-text filenames.
9. Validation plan
Test 1: an inbox with a PDF, an audio note, a text screenshot, and known duplicate sentences in the library; check sentences, skipped duplicates, and written category_source.md files against a hand-labeled key. Test 2: a question answered in one existing file; the reply must cite that source and invent none. Edge case: an unreadable file and a refused approval; mark the file uncertain, write nothing, and never delete.
10. Limitations, risks, dependencies
Ingest starts at the inbox; the kit cannot download from the web. Sentence splitting and pre-categorization are not published actions and must be validated. collect-evidence is query-based retrieval, not a dedicated near-duplicate API. Writes use mkdir/echo or save-text. All listed actions have unknown stability and permissions. Depends on UGOS and catalog 2.0.0; no external services. Fully offline AI is not assumed until test-day verification.
Disclosure (excluded from word count)
Third-party components: none bundled; the Skill relies only on UGREEN's official uliya-skill-kit (capability catalog 2.0.0; format and static checker) and NAS built-in experts. AI assistance: an AI assistant helped draft the Skill text and this proposal; the entrant defined the concept, reviewed and edited all content.
Built With
- markdown
- ugreen-nas
- uliya-skill-kit
