Auto Quotation System

Problem

Quotations often stall because requirements are fragmented, historical quotes cannot be reused directly, and the final number feels like a guess. This skill turns requirement documents and historical quotation files into a structured quotation draft, emphasizing module-level estimates, explicit assumptions, and stable JSON output rather than a single total price.

How It Works And Boundaries

Core workflow:
- Normalize requirements: use Markdown or plain text directly, or extract DOCX text, then organize project name, delivery channels, feature list, and non-functional requirements.
- Calibrate samples: use extract_docx_corpus.py to mine historical DOCX files for sections, tables, and keywords, then build sample libraries, calibration profiles, and domain price bands.
- Generate the quote: use generate_quote_draft.py to produce module detail, role effort, payment schedule, delivery boundaries, similar cases, and open questions; optionally render a Word file with render_quote_docx.py.

Keep historical quotes as reference data, not copy targets. If requirements are incomplete, keep pending questions visible instead of silently guessing. Review third-party costs, AI, OCR, ERP, hardware, and private deployment as high-risk items. Also check whether administration, testing, deployment, and maintenance phases are missing, and whether change-control rules are stated.