--- title: Top Document Processing Skills slug: document-processing type: primary-category canonical: https://top-agent-skills.com/top-skills-for-document-processing skills: 16 rubric: 1.0 updated: 2026-08-31 --- # Top Document Processing Skills **docling** is the top-ranked documents Skill at 66/100: One skill covering every document format the category exists for — including scanned PDFs and RAG chunking — from the maintainers of the parser itself. Create, parse, and edit DOCX, PDF, XLSX and PPTX files. OCR, format conversion, deterministic scripts. 16 Skills indexed, ranked by a published composite score (https://top-agent-skills.com/about/methodology). Content last changed 2026-08-31. ## The ranking ### 1. docling — score 66/100 (verified-org) One skill covering every document format the category exists for — including scanned PDFs and RAG chunking — from the maintainers of the parser itself. - Best for: Reading or converting documents the agent cannot open directly, at 65K stars and MIT: scanned PDFs, Office formats, tables, and chunking a corpus for retrieval. - Not ideal for: Producing polished Office files. Docling is a reader and converter — for generating a styled DOCX, XLSX or PPTX, the Anthropic document skills are the better fit. - Publisher: Docling project (LF AI & Data) · MIT - Verified on: claude-code, cursor, codex ```bash pip install docling && uvx library-skills --claude ``` Full entry: https://top-agent-skills.com/skill/docling (https://top-agent-skills.com/skill/docling.md) ### 2. azure-ai-translation-document-py — score 59/100 (verified-org) Batch document translation that preserves layout — the part that breaks when you translate a PDF by extracting text and re-flowing it. - Best for: Translating a corpus of Word, PDF, Excel or PowerPoint files where the formatting has to survive the round trip. - Not ideal for: A single short document, or anything that cannot leave your infrastructure — it is a cloud batch service with Azure credentials attached. - Publisher: Microsoft · MIT - Verified on: claude-code ```bash npx skills add microsoft/skills --skill azure-ai-translation-document-py ``` Full entry: https://top-agent-skills.com/skill/azure-document-translation (https://top-agent-skills.com/skill/azure-document-translation.md) ### 3. azure-ai-contentunderstanding-py — score 56/100 (verified-org) Multimodal extraction across documents, images, audio and video through one Azure client — Microsoft-authored, with an explicit trigger list. - Best for: Teams already on Azure that need structured extraction from mixed media, not just PDFs — the case where a single-format parser leaves half the corpus unread. - Not ideal for: Local or offline processing. It teaches an agent to write correct Azure SDK code, so it needs Azure credentials and sends content to the service. - Publisher: Microsoft · MIT - Verified on: claude-code ```bash npx skills add microsoft/skills --skill azure-ai-contentunderstanding-py ``` Full entry: https://top-agent-skills.com/skill/azure-content-understanding (https://top-agent-skills.com/skill/azure-content-understanding.md) ### 4. liteparse — score 53/100 (verified-org) Local document extraction with no API key — read the file, not the whole file. - Best for: Looking up a figure or table inside a large document without burning context on the parts you do not need. - Not ideal for: Complex layout reconstruction or scanned-image OCR — the hosted llamaparse skill in the same repo is the heavier option for that. - Publisher: LlamaIndex · MIT - Verified on: claude-code ```bash npx skills add run-llama/llamaparse-agent-skills --skill liteparse ``` Full entry: https://top-agent-skills.com/skill/liteparse (https://top-agent-skills.com/skill/liteparse.md) ### 5. box — score 53/100 (verified-org) Box as an agent surface, plus the contract-review workflows nothing else in the catalog covers. - Best for: Teams whose documents already live in Box, especially contract review and diligence over files stored there. - Not ideal for: Anyone not on Box — every skill routes through Box auth and its MCP or CLI. - Publisher: Box · MIT - Verified on: claude-code ```bash npx skills add box/box-for-ai --skill box ``` Full entry: https://top-agent-skills.com/skill/box-for-ai (https://top-agent-skills.com/skill/box-for-ai.md) ### 6. pdf — score 52/100 (official-anthropic) Extract, fill, merge. PDF manipulation with bundled scripts so Claude does not waste tokens generating them. Full entry: https://top-agent-skills.com/skill/pdf (https://top-agent-skills.com/skill/pdf.md) ### 7. docx — score 48/100 (official-anthropic) Word documents end-to-end. Create, edit, comment, track-change. Full entry: https://top-agent-skills.com/skill/docx (https://top-agent-skills.com/skill/docx.md) ### 8. xlsx — score 48/100 (official-anthropic) Excel spreadsheets with real formulas, charts, and formatting — not flat CSV exports. Full entry: https://top-agent-skills.com/skill/xlsx (https://top-agent-skills.com/skill/xlsx.md) ### 9. pptx — score 48/100 (official-anthropic) PowerPoint decks with layouts, charts, speaker notes — programmatic from prose. Full entry: https://top-agent-skills.com/skill/pptx (https://top-agent-skills.com/skill/pptx.md) ### 10. doc-coauthoring — score 48/100 (official-anthropic) Multi-author edits with suggestion mode and conflict resolution. Full entry: https://top-agent-skills.com/skill/doc-coauthoring (https://top-agent-skills.com/skill/doc-coauthoring.md) ### 11. markdown-exporter — score 48/100 (community) The widest format matrix of any conversion skill we catalogue, and it executes the conversion rather than describing it. Full entry: https://top-agent-skills.com/skill/markdown-exporter (https://top-agent-skills.com/skill/markdown-exporter.md) ### 12. zotero-skills — score 48/100 (community) Programmatic access to the reference manager academia actually uses. Full entry: https://top-agent-skills.com/skill/zotero-skills (https://top-agent-skills.com/skill/zotero-skills.md) ### 13. claude-healthcare — score 47/100 (official-anthropic) FHIR to prior-auth to ICD-10 coding: the official payer/provider/pharma toolkit. Full entry: https://top-agent-skills.com/skill/claude-healthcare (https://top-agent-skills.com/skill/claude-healthcare.md) ### 14. research-co-pilot — score 47/100 (community) The academic writing half nothing else covers: manuscript, grant, peer review, reviewer response. Full entry: https://top-agent-skills.com/skill/research-co-pilot (https://top-agent-skills.com/skill/research-co-pilot.md) ### 15. carbone — score 46/100 (verified-org) Data-driven document generation: one template, a thousand filled documents. Full entry: https://top-agent-skills.com/skill/carbone-templates (https://top-agent-skills.com/skill/carbone-templates.md) ### 16. research-workspace — score 44/100 (community) Zotero plus Obsidian plus NotebookLM as one agent-operable research workspace. Full entry: https://top-agent-skills.com/skill/research-hub (https://top-agent-skills.com/skill/research-hub.md) ## About this category The best agent skills for document processing in 2026 are Anthropic's official docx, pdf, pptx, and xlsx Skills — they handle the formats every business runs on with tested scripts for edge cases (track-changes, password protection, merged cells) that improvised Python often misses. Document-processing Skills give AI agents deterministic control over Word, PDF, Excel, PowerPoint and CSV files. Instead of having Claude or Cursor improvise file manipulation in plain Python every time, a Skill bundles tested scripts, format-specific edge cases (track-changes, embedded images, locked sheets, password protection), and the precise install steps for each agent. The result is reproducible: the same prompt produces the same file, every time. Most teams reach for these Skills the moment an agent has to touch a real customer artifact — a contract redline, a financial model, a board deck, a parsed claim form. Common workflows include extracting tables from scanned PDFs, find-and-replace across hundreds of DOCX files, generating PPTX decks from a Markdown outline, building XLSX models with formulas and conditional formatting, and OCR pipelines that turn scanned paper into searchable, indexable text. The audience is broad: legal ops, finance, sales engineering, customer support, and any engineer wiring document automation into a SaaS product. We rank skills here by composite score — install count, provenance (Anthropic-authored skills like docx, pdf, pptx, xlsx score highest), cross-agent compatibility, recency, documentation depth, and how friction-free the install is on Claude Code, Cursor, Codex, Gemini CLI and Antigravity. ## Common questions **What is a document-processing Skill?** A document-processing Skill is a SKILL.md-format bundle that teaches an AI agent how to read, edit, or generate a specific file type — DOCX, PDF, XLSX, PPTX, CSV — using deterministic scripts rather than ad-hoc code. It ships as a folder the agent loads on demand. **Which agents support document Skills today?** Anthropic's docx/pdf/pptx/xlsx Skills run natively in Claude Code, Claude.ai (Pro/Max/Team/Enterprise), and the Claude API. Codex CLI, Cursor, Gemini CLI and Antigravity all support the SKILL.md spec, with Antigravity bundling 1,000+ pre-installed. **Do document Skills work offline?** Yes for the file manipulation itself — the bundled scripts run locally inside the agent's sandbox. OCR Skills that call cloud vision APIs (Anthropic's file-search, Google Vision) require network access; pure pdfplumber/python-docx flows do not. **How do I pick between docx, pdf, pptx and xlsx Skills?** Pick by output format, not input. If you need to produce a Word file, use docx. If you need a styled deck, use pptx. xlsx Skills handle formulas and pivot tables that generic CSV writers cannot. PDF skills handle both reading (extraction) and writing (form fill, watermark, encrypt). **Are these Skills better than just asking the model to write a script?** For one-off tasks, prompting works. For anything reproducible — same input, same output, every time — a Skill is meaningfully more reliable. It bundles tested scripts, handles format edge cases the model often gets wrong (encrypted PDFs, .docx tracked changes, .xlsx merged cells), and avoids prompt-to-prompt drift. --- _Top Agent Skills. Machine-readable twin of https://top-agent-skills.com/top-skills-for-document-processing._ _JSON: https://top-agent-skills.com/top-skills-for-document-processing.json · whole catalog: https://top-agent-skills.com/skills.json (CC BY 4.0)_