What it does
insurance-case-brief turns an insurance agent's scattered client files on the NAS, including scanned PDFs and phone photos, into a source-linked case brief, a missing-document checklist, and an unsent follow-up draft.
Target users and problem
Independent and small-team agents who keep client files on a NAS reread policies, receipts, insurer letters and photos of forms before every meeting or claim follow-up, to find what changed and what the insurer still needs. These include health and financial records; with a suitable local model configured, the workflow can reduce reliance on external AI services.
Solution
One reusable workflow with four modes: meeting brief, factual comparison, claim or underwriting checklist, and continuation from a saved brief. An administrator-managed shared library supplies generic product documents and templates; client evidence and outputs stay in the agent's personal area.
Core user flow
The agent names a case and task and selects its folder. The Skill lists the files it will read, reads the documents and the photos the agent confirms, then shows findings, gaps and conflicts with sources. The agent approves saving the brief as Markdown. Next time, the saved brief plus new files produce an updated brief.
Difference from existing tools
A general chatbot needs files uploaded each time and blurs evidence. Here evidence status drives every checklist: if an insurer requests three documents and two are found, the brief names the missing one and marks a photographed form's signing status for manual confirmation. Billed and receipt-evidenced amounts, illustrations and issued policies, and conflicting versions stay distinct across saved briefs.
UGREEN AI NAS capabilities
ug-filesystem finds the case and shared files (discovery). ug-document reads, compares and cites policy documents and extracts receipts, including scanned PDFs (reading). ug-photo vision-ocr reads phone photos of forms and receipts (reading). ug-knowledge saves the approved brief with read-back verification (saving), and the saved brief anchors later continuation.
Technical approach
An orchestration Skill delegates only to file-agent, document-agent, photo-agent and knowledge-agent, passing verified results forward. Inputs: case identifier, task, authorized paths, optional personal destination. NAS permissions govern reads; platform approval governs the only write. A suitable local model is preferred; actual routing is disclosed before confidential use. No external API.
Development status
Done: Skill instructions, evidence rules, shared-library boundaries, photo and receipt routing, synthetic text fixtures, and this proposal. The 0.7.7 checker reports zero errors and warnings. Not done: PDF and photo fixtures, direct OCR of discovered photo paths, installation and execution, OCR quality, citation accuracy, and account isolation, all pending assigned-NAS validation.
Validation plan
Core 1: meeting brief comparing two currencies, using shared references, a receipt photo, and verified saving. Core 2: insurer checklist with a photographed hospital receipt, then continuation after a signed form arrives. Edge: conflicting policy versions, a blurred photo, and a shared output destination; expect visible uncertainty and no shared write.
Limits and dependencies
Depends on administrator-configured folders, platform experts and models, and readable inputs. Photos are read one confirmed image per request; OCR may miss values and cannot confirm signatures, so these need manual checks. No email import, reminders, insurer integration, calculations, or automatic sending. Outputs require agent review and make no coverage or payout decision.
Built With
- markdown
- ugos
