What problem does it solve? Engineering tasks often receive mixed inputs—PDFs, scans, Word and Excel files, drawings, charts, and photographs—where values extracted without provenance can silently enter simulations or safety decisions. This Skill classifies each source, plans the right extraction operations, and packages every fact with original text, page or locator, confidence, and review status so nothing unverified reaches downstream NeqSim workflows. ## Core Features & Use Cases - Source Classification and Routing: Detects PDF, Office, spreadsheet, text, and image formats and builds an ordered ExtractionPlan covering native text, tables, OCR, rendering, and vision steps. - Evidence Contract: Every fact carries original text, unit, page or locator, extraction method, confidence, and an automatic review gate for safety-critical, ambiguous, or low-confidence values. - Conflict Detection: Compares normalized values across multiple sources and reports disagreements as review items instead of silently picking one. - Use Case: Given a scanned equipment datasheet PDF plus a conflicting Excel specification, plan OCR and table extraction, capture design pressure with its page citation, and flag the mismatch between sources for human review before running a NeqSim simulation. ## Quick Start Ask the agent to classify the attached engineering PDF and produce a source-traceable evidence package of its design values with confidence scores and review flags.
