anydocSkill
Claude Code skill that converts documents to Markdown and HTML while preserving correct text order, built specifically to solve the Hebrew/right-to-left text extraction problem that scrambles output in most document extractors.
Overview
anydocSkill is a Claude Code skill that converts PDFs, Word, PowerPoint, Excel, OpenDocument, and EPUB files into Markdown and HTML. It wraps Firecrawl's anydoc library for office formats, which it credits with the best structural fidelity, but deliberately avoids anydoc's PDF extractor in favor of pdf.js — because the author found Firecrawl's PDF path scrambles Hebrew and other right-to-left text beyond repair. The skill's core design principle is to refuse to output scrambled text rather than silently produce a wrong conversion, verifying its own output against the source document when the --verify flag is used. A companion anydoc-ocr skill handles scanned documents via ocrmypdf.
The Verdict
Who Should Use anydocSkill?
Best For
- Claude Code users converting Hebrew or other RTL-language documents, where mainstream extractors scramble reading order
- Legal or formatted documents needing preserved clause numbering and removed repeated headers/footers
Not Ideal For
- Production pipelines needing a proven track record — this is a one-month-old personal project with no visible adoption beyond its author
- Non-Claude-Code workflows — it's built specifically as a Claude Code skill, not a standalone CLI or API
- Teams needing active community support or a maintained roadmap
What's Great
- Directly addresses a real, specific gap: scrambled Hebrew/RTL text extraction that mainstream tools (including Firecrawl's own PDF extractor) get wrong
- Verification mode (--verify) compares output against the source document instead of assuming correctness
- Preserves legal clause numbering and strips repeated headers/footers
- MIT licensed, free to use
Watch Out For
- Very early-stage: created September 2026, 0 forks, built and maintained by a single developer
- No public adoption signal, third-party reviews, or community discussion found
- One open issue flags missing test fixtures for malformed packages and mixed-script RTL content
- Tied to the Claude Code skill ecosystem rather than usable as a general-purpose tool
Pricing
View all features & details
Supported Formats
- PDF (via pdf.js, not anydoc's PDF path)
- DOCX, PPTX, XLSX, ODT, EPUB (via Firecrawl anydoc)
- Plain text
Key Behaviors
- Refuses scrambled Hebrew/RTL output instead of producing silently-wrong text
- --verify flag checks output against the source document
- Recovers section headings from hand-formatted documents
- Preserves table structure and legal clause numbering
Related Skill
- anydoc-ocr — companion skill for scanned documents using ocrmypdf
How It Compares
| Feature | anydocSkill | Anydoc (Firecrawl) | MarkItDown.tech |
|---|---|---|---|
| RTL/Hebrew Handling | Purpose-built refusal logic | Known to scramble (per author) | Not verified |
| Interface | Claude Code skill | Library/CLI/API | Browser app |
| Maturity | 1 month old | Established, active | Commercial, established |
| Output Verification | Built-in –verify flag | No | No |
| License | MIT | MIT | Commercial (freemium) |