DocMind NAS

Inspiration

Research notes, whitepapers and specs pile up in different folders, and nobody reads them side by side. When two documents quietly disagree, such as two different benchmark numbers for the same thing, it usually goes unnoticed until it causes a problem. I wanted a NAS Skill that does the cross-checking and shows its evidence.

What it does

DocMind NAS audits the documents in one NAS folder on a topic you choose. You give it a source folder, a topic and an output folder. It finds the documents, compares them, and saves one Markdown brief to your NAS containing:

• an executive summary
• a comparison table with citations (file and page, or heading for notes)
• contradictions, with both conflicting statements quoted side by side
• gaps, meaning points that no source covers
• a Mermaid mind map of the topic
• a list of any files it could not process

Source files are read-only. Nothing is written outside the output folder, and no external services are called.

How I built it

DocMind NAS is a single SKILL.md that orchestrates three registered UGOS experts:

• ug-filesystem (file-agent) lists the documents in the folder.
• ug-document (document-agent) reads and cross-checks them.
• ug-knowledge (knowledge-agent) collects supporting evidence and saves the brief.

Each step passes only verified results to the next. A linked reference file defines the brief layout. The package passes the Uliya Skill Kit static checker with 0 errors and 0 warnings.

Challenges

My first design assumed capabilities that don't exist, including document OCR and a background indexer. The Starter Kit's catalog showed what is really supported, so I rebuilt the workflow around the real experts. The result is smaller, but every step maps to a real capability. Handling unreadable files honestly was the other challenge: the Skill reports a partial result and never invents content.

What I learned

Skills are orchestration, not code. Good instructions give each expert one clear job and define what happens when a result is empty, partial or fails.

Status and what's next

Done: the full SKILL.md, the brief template and the static check. Not yet validated on a real NAS: Markdown reading, scanned-PDF handling, the approval flow and the saved output. If selected as a finalist, I will test these on the NAS with synthetic documents, including a planted contradiction, mixed PDFs and notes, and a scanned/corrupted file.

Built With

• agents
• ai
• markdown
• mermaid
• ug-document
• ug-filesystem
• ug-knowledge
• ugos
• uliya-skill-kit