Official SoMark skills collection for document parsing, image OCR, and intelligent extraction — built for AI agent workflows.
npx skills add https://github.com/SoMarkAI/skillsWorks with Claude Code, Cursor, Cline, OpenCode, and 40+ other agents.
| Skill | Description |
|---|---|
| somark-document-parser | Parse PDFs, Word, PowerPoint, and images into structured Markdown, JSON |
| pdf-online | Parse a document once with SoMark and publish it as an editable Feishu, DingTalk, or Notion document, spreadsheet, table, or database |
| image-parser | Core image OCR capability that returns text with precise coordinates (OCR + location awareness) |
| illustrations-local-storage | Persist remote SoMark Markdown images as sequentially named local JPEG files with rewritten Markdown references |
| get-catalog-tree | Extract a document's heading hierarchy as a reusable nested catalog JSON, preserving page and source-block references |
| document-diff | Compare two documents and generate a structured diff report showing changes, additions, and deletions |
| contract-reviewer | Review contracts for risks, unfair clauses, missing provisions, and key obligations with severity ratings |
| resume-parser | Parse resumes and CVs into structured JSON profiles with opinionated candidate assessment |
| tender-analyzer | Extract qualification requirements, scoring criteria, deadlines, and submission checklists from procurement documents |
| paper-digest | Deeply analyze academic papers into structured research cards covering methods, results, and contributions |
| financial-report-analyzer | Extract financial metrics, risk signals, and management commentary from annual reports and earnings releases |
| pitch-screener | Screen startup pitch decks from a VC/angel investor perspective — parses the deck, runs background research via web search, and produces a pre-meeting investment memo |
When you share a document with your AI agent, SoMark parses it into structured Markdown, JSON that the agent can actually reason over — not just OCR'd text, but proper headings, tables, formulas, and layout.
The image-parser skill goes further: it returns every text block with its exact pixel coordinates on the original image, enabling field extraction, region location, and document automation.
Supported formats:
| Type | Formats |
|---|---|
| Documents | PDF, DOC, DOCX, PPT, PPTX |
| Images | PNG, JPG, JPEG, BMP, TIFF, WEBP, HEIC, HEIF, GIF |
Example triggers:
- "Parse this PDF for me"
- "Publish this PDF to Feishu, DingTalk, or Notion as an editable document"
- "Extract the key clauses from this contract"
- "Review this contract for risks"
- "Parse this resume and give me a candidate assessment"
- "Analyze this annual report"
- "Summarize the paper I just uploaded"
- "What changed between these two documents?"
- "Analyze this tender document — what are the qualification requirements?"
- "Convert this document to Markdown"
- "Extract the catalog tree from this document"
- "Turn this SoMark JSON into a nested document outline"
- "What does this image say?"
- "Extract all text with bounding boxes from this image"
- "Find the invoice amount and its position on the page"
Get an API key:
Then set it as an environment variable for your terminal:
# macOS / Linux (Bash, Zsh)
export SOMARK_API_KEY="sk-your-api-key"# Linux (Fish)
set -x SOMARK_API_KEY "sk-your-api-key"# Windows PowerShell
$env:SOMARK_API_KEY = "sk-your-api-key"# Windows Command Prompt (CMD)
set "SOMARK_API_KEY=sk-your-api-key"Or add it to your agent's settings. The skill will guide you through setup on first use.
Free quota: SoMark offers a free tier:
- Mainland China (中国大陆): Visit the purchase page
- Outside mainland China: Visit the purchase page
Follow the instructions there to claim it.
Most agents struggle with documents because raw PDF/image data loses structure. SoMark preserves:
- Heading hierarchy — agents can understand document sections correctly
- Tables — fully reconstructed instead of flattened into plain text
- Formulas and diagrams — converted to LaTeX or described accurately
- Multi-column layouts — reading order is preserved
The result: your agent gives accurate, context-aware answers instead of hallucinating from garbled text.
| Constraint | Limit |
|---|---|
| Max file size | 200 MB |
| Max pages | 300 pages |
| QPS per account | 4 |
MIT