Receipt & Warranty Vault

What it does

Receipt & Warranty Vault turns a messy NAS folder of receipt, invoice, and contract scans into a user-confirmed, searchable archive with extracted details and a warranty-expiry watchlist.

The problem

Freelancers, home-office workers, and families keep their purchase paperwork on a NAS as mixed PDFs and phone photos. Facing a refund, a tax filing, or a warranty claim, they open dozens of files to find the document, the amount, or the expiry date.

How it works

The Skill scans an authorized folder, reads each document, and extracts vendor, date, total, category, and warranty or renewal date. It reconciles duplicate scans, and after user confirmation files the originals into a year-and-vendor archive and saves a ledger and expiry watchlist on the NAS.

The main user flow is simple: the user says, "Organize October's receipts and flag warranties expiring soon." The Skill confirms the scope, scans, extracts, and presents a review table flagging duplicates and unreadable files. The user approves, and the Skill files and tags everything, saves the ledger, and ends with a summary.

Why it is different

Cloud AI chat tools make users upload private financial documents off-site and return one-off answers, with no lasting archive. Expense apps add subscriptions and lock-in. This Skill runs where the documents already live, leaves a persistent archive, and treats warranty deadlines as first-class results.

Built on UGREEN

It uses UGREEN's supported UGOS capabilities: Files for intake and filing (find, deduplicate, tag, and organize), Documents for reading, extracting fields, and reconciling duplicates, and Knowledge for saving the ledger and watchlist as text. Reading runs on the local AI model, on-device.

Technical approach

An orchestrator runs four expert agents: Intake (Files), Extraction (Documents), Reconciliation (Documents), and Archivist (Files and Knowledge). Inputs are the target folder, date range, and expiry-warning window. The Skill is read-only until the user confirms every move or overwrite; nothing is deleted, no external services are used, and no credentials are stored.

Current status

Built: the complete SKILL.md package (orchestration workflow, confirmation gates, and completion and failure handling), authored with the official UGOS Skill Creation Tool. Not built yet: extraction tuned on real NAS documents. To validate on a NAS, we will test photo-versus-PDF extraction accuracy, duplicate-matching thresholds, and large-folder performance.

Finalist test plan

Test 1: twenty receipts and invoices, mixed PDFs and photos, extracted and filed with the correct vendor, date, and total for each readable item. Test 2: the same invoice scanned into two folders reconciled into one ledger record, never double-counted. Edge case: a blurred photo reported as skipped, with a reason, while the run completes.

Limits and risks

Extraction depends on the local model and scan quality; handwriting and damaged receipts may fail and are flagged rather than guessed. Totals are recorded exactly as printed, with no currency conversion. The Skill cannot read password-protected or corrupted files, and large folders may need batching.

Disclosures

This Skill uses only UGREEN's supported UGOS capabilities (Files, Documents, Knowledge), authored with the official UGOS Skill Creation Tool; no third-party code, APIs, or external services are used. AI tool used: Claude — principal use: drafting and structuring the SKILL.md workflow and this proposal, reviewed and finalized by the entrant.

Built With

• claude
• creation
• skill
• ugos