Stock Intake Trace

Inspiration

Warehouse stock controllers often have to compare a purchase order, a receiving or delivery note, and a supplier invoice by hand. Even a small difference in SKU, quantity, or unit can take time to trace back to its documentary source. Stock Intake Trace addresses this focused, everyday workflow.

What it does

Stock Intake Trace is a proposed UGREEN AI NAS Skill for an authorized folder containing a specific inbound delivery's documents. It identifies eligible files, extracts supported fields, and prepares a source-linked exception report. Each item is classified as MATCH, MISMATCH, MISSING, or UNVERIFIABLE. A mismatch is a difference between documents, not proof that physical inventory is wrong. The Skill never changes stock, purchase orders, invoices, WMS, or ERP records.

How we built it

The original Skill definition and evidence policy were newly drafted for this hackathon using UGREEN's official Uliya Skill Developer Package 0.7.3. The planned orchestration delegates file discovery to file-agent via ug-filesystem, document extraction and supported reconciliation to document-agent via ug-document, and optional user-approved text report storage to knowledge-agent via ug-knowledge. Each delegated task stays within the user's explicitly approved NAS scope.

OpenAI ChatGPT/Codex assisted with requirements analysis, drafting, and quality review. There are no bundled scripts, binaries, credentials, third-party runtime libraries, or external network integrations in the Skill. The official Uliya static checker reports a clean pass against capability catalog 2.0.0 (zero errors and warnings). This is an initial Round 1 package, not a claim of operational NAS deployment.

Challenges and lessons

The hard problem is reliable reconciliation under uncertainty: inconsistent units, ambiguous item identities, scanned documents, incomplete invoices, and missing source references. We designed a conservative evidence policy that reports uncertainty instead of guessing or silently converting packs to units. Native permissions and user approval remain central to the workflow.

Validation plan and next steps

If selected as a finalist, we will run the same Skill on the assigned UGREEN AI NAS using synthetic documents: (1) fully matching order/receipt/invoice lines; (2) a documented 100 ordered / 92 received / 100 invoiced discrepancy; and (3) an edge case with unreadable SKU or unknown unit conversion. We will check provenance, safe failure behavior, and the optional report-saving approval flow. Live execution, OCR reliability, and performance have not yet been validated.

No confidential employer documents or personal supplier records were used in this entry.

Built With

• ai
• chatgpt
• markdown
• ugos
• ugreen
• uliya