Office PDF Toolkit

office/pdf-toolkit — skill payload

Machine-readable contract: ../pdf-toolkit.manifest.json. This file is the narrative a host LLM reads before deciding to invoke the skill.

What this skill does

Wraps the standard Python PDF stack (pdfplumber for text/tables, pypdf for structure and split/merge, reportlab for new-document generation) behind a single operation parameter, so an agent can manipulate PDFs without learning three libraries.

Supported operations:

• extract_text — full text, optionally per-page.
• extract_tables — structured rows, one list per detected table.
• fill_form — fill an AcroForm with a {field: value} map and flatten.
• split — produce N output PDFs by page range.
• merge — concatenate multiple PDFs.

When to use it

• You need data out of a PDF in machine-readable form.
• You need to fill a standard form (tax form, expense claim) at scale.
• You're stitching PDFs together for an audit pack.

When not to use it

• The document is an invoice and you want structured invoice fields — call finance/invoice-extract instead; it composes this skill internally and also does field-mapping.
• You need OCR on a scanned image-only PDF — this skill can extract text layers but does not run OCR.
• The output target is editable Word — author with office/docx-toolkit and export to PDF separately.

Determinism and risk

Determinism: high — same input, same output. Risk: low — pure in-memory file manipulation, no external side effects.

How it composes

Acts as a building block: pdf.extract ← finance/invoice-extract, pdf.write ← office/docx-toolkit (export path), and any audit-pack workflow that needs to concatenate evidence PDFs.