PDF

Use this skill for project PDFs, uploaded datasheets, scanned references, manuals, drawings, and image-heavy documents.

Workflow

1. Start from the PDF path and excerpt in project state.
2. Use pdf.search when the relevant pages are unknown. It returns ranked page hits, nearby headings, and exact continuation calls.
3. Use pdf.read for bounded exact text from known pages or ranges. Follow continuation_calls when the result is partial.
4. Use pdf.render_pages when visual content matters: embedded images, diagrams, charts, scans, screenshots, page layout, package drawings, pinout figures, callouts, dimensions, and tables whose layout changes meaning. Rendered page images are attached to the result automatically.
5. Render only relevant pages. If a request exceeds the per-call limit, follow the returned suggested_batches.

Rules

• Do not read PDF binary files with fs.read.
• Do not use fs.read on .tool-output as the normal way to continue a PDF result. Prefer its exact continuation_calls; archives are for audit recovery.
• Do not infer embedded-image, scanned-page, or diagram content from pdf.read; text extraction may omit or reorder visual facts.
• Treat no_match and broad_match as completed search outcomes, not generic success. Broaden or refine the query once before concluding the PDF lacks the information.
• Distinguish extracted text from visual observations in reports.
• Cite page numbers when using PDF evidence.
• If rendered text or diagrams are unreadable, render fewer pages at a higher dpi or ask the user for a clearer source.

Common Patterns

Datasheet dimensions:

1. pdf.search with queries such as package, mechanical drawing, dimensions, recommended footprint, pinout, or mounting.
2. pdf.read for the matching page numbers.
3. pdf.render_pages for the matching package drawing, footprint, or pinout pages and inspect the attached images.

Scanned or image-only PDF:

1. pdf.render_pages for the needed pages.
2. Inspect the attached rendered page images.
3. Use pdf.search or pdf.read only to confirm whether any text layer exists.