What problem does it solve? Turning PDFs and images with tables, figures, and charts into structured, citable data usually requires brittle parsing code and repeated API calls that re-bill on every retry. This Skill drives the ADE (Agentic Document Extraction) v2 APIs through the ade CLI, persisting every run in a local job-item store so repeat runs are free, interrupted runs resume, and every extracted value can cite element ids with page-and-box evidence. ## Core Features & Use Cases - Document parsing: ade parse converts documents into grounded Markdown and elements, with results stored under ~/.ade/jobs/<job-item-id>/ and deduplicated by invocation identity. - Schema extraction: ade extract pulls JSON-schema-shaped fields from a parse, computing per-field evidence (spans joined to pages and boxes) offline, and flagging ungroundable or empty fields explicitly. - Local search and citation: ade find searches elements without API calls, ade crop renders element regions as PNGs, and ade view builds a self-contained HTML viewer with deep links for citations. - Use Case: Given a folder of invoices, parse each once, extract totals and line items against a JSON schema, then answer questions citing element ids and deep links — re-running any step consumes no credits. ## Quick Start Ask the agent to run ade parse on your document, then ade extract with your JSON schema, always passing --json so results arrive as stable structured output.