1. What does it do?
PrepOrder Brief is designed to generate an evidenced pre-order brief for restaurant procurement. It will compare current quotes against older archives, highlighting changes in price, units, and validity. The proposed flow never silently converts units or auto-selects sources, requiring operator confirmation to review uncertain terms.
2. Target user and problem?
Restaurant owners and procurement officers in Hong Kong. The problem hypothesis is difficulty reconciling complex supplier quotes with varying currencies, pack sizes, and delivery terms. Actual frequency and costs remain unvalidated.
3. Solution?
A proposed NAS module to extract terms from authorized quote documents and reconcile them against historical records, and flags discrepancies. It would provide a bilingual (English/Cantonese) exception report, requiring explicit operator confirmation before saving a final briefing document.
4. Core flow?
- Operator selects authorized NAS folder and date. 2. System finds quote/SOP files. 3. Agents extract supplier, item, currency, units, and dates with source evidence. 4. Reconciliation identifies changed fields or incompatible units. 5. Operator confirms valid source. 6. Briefing is saved only after exact path confirmation.
5. Difference?
The planned distinction is a shift-ready bilingual exception brief with an operator-selected authoritative version. The flow cites source files and pages, marks missing data as HOLD, and requires operator-selected authoritative versions. It does not claim auto-ordering or production reliability.
6. NAS capabilities and steps?
Uses ug-filesystem for file-agent find/ls/stat. Uses ug-document for document-agent extract/qa/reconcile. Uses ug-knowledge for knowledge-agent save-text. Each operation is delegated to registered experts; no direct shell or Python runtime is used in the entry.
7. Technical approach?
The Skill orchestrates only registered experts, passing confirmed scope and completed upstream evidence to each step. The official creation kit guided authoring and static validation. Local Ollama assisted drafting with Codex review. Host-model configuration, access controls and data flows remain subject to assigned-device validation.
8. Current status?
A newly authored SKILL.md and two linked Markdown references are complete. The official v0.7.3 checker returned static-pass with zero errors and warnings. The earlier PrepOrder Python application informed the idea only; no previous code or submission files are reused. NAS execution remains untested.
9. Validation?
Core1: Two synthetic English quotes with different prices/dates to test change table generation. Core2: Cantonese alias and SOP requirements to test HOLD flags for missing quantities. Edge cases: absent files or extraction failures must fail closed with missing evidence, no fabrication.
10. Limitations?
No stock calculations, payment or order sending. File discovery is bounded and must not silently omit relevant sources. Extraction quality, page evidence, model behaviour, permissions and saved-file readback require NAS validation. Missing evidence remains visible; no measured customer savings are claimed.
Authorship and third-party disclosure
Ho Tung Cheng, individual entrant. Created 26 October 2026. Local Ollama assisted bounded drafting; Codex supervised factual and static checks. UGREEN Uliya Skill Creation Kit v0.7.3 and capability catalog 2.0.0 were used under vendor terms, not redistributed or relicensed. The entry contains original Markdown only; no model weights or third-party libraries. Copyright 2026 Ho Tung Cheng; rights retained subject to contest terms. Host experts and model remain subject to UGREEN runtime terms. No real supplier records, API keys or private customer data are included.
Log in or sign up for Devpost to join the conversation.