Document Skill
Routes
| Route |
Trigger phrases |
Tool |
| HTML to PDF |
"create PDF", "generate report", "render HTML as PDF", "save this as PDF" |
document.pdf-from-html |
| Markdown to PDF |
"turn this markdown into PDF", "convert markdown to PDF", "PDF from notes" |
document.markdown-to-pdf |
| Extract text |
"get text from this PDF", "read the PDF", "extract content from PDF", "parse PDF" |
document.pdf-extract-text |
| Extract tables |
"pull tables from PDF", "get the tables", "extract table data from PDF" |
document.pdf-extract-tables |
Decision Rules
- User has HTML → use
document.pdf-from-html directly. Supports A4/Letter, landscape, custom margins.
- User has Markdown → use
document.markdown-to-pdf. Internally converts Markdown→HTML→PDF.
- User wants to read a PDF → use
document.pdf-extract-text. Returns full text with layout preservation.
- User wants table data from a PDF → use
document.pdf-extract-tables. Returns structured rows/cells.
- Page range needed → both extraction tools accept
pages: "1-5" or pages: "3".
- Format preference →
document.pdf-extract-text accepts format: "json" to return per-page arrays.
Output Path Rules
PDF creation tools (pdf-from-html, markdown-to-pdf) save to:
/tmp/<filename>.pdf — for temporary outputs<project-root>/data/documents/<filename>.pdf — for persistent outputs
Always use an absolute path in the outputPath parameter.
Examples
Generate a PDF report from HTML
{
"tool": "document.pdf-from-html",
"params": {
"html": "<html><body><h1>Monthly Report</h1><p>Results for Q1...</p></body></html>",
"outputPath": "/tmp/monthly-report.pdf",
"format": "A4",
"landscape": false,
"margins": { "top": 15, "right": 15, "bottom": 15, "left": 15 }
}
}
Convert Markdown notes to PDF
{
"tool": "document.markdown-to-pdf",
"params": {
"markdown": "# Meeting Notes\n\n## Action items\n- Follow up with team\n- Send report",
"outputPath": "/tmp/notes.pdf",
"title": "Meeting Notes 2026-04-13"
}
}
Extract all text from a PDF
{
"tool": "document.pdf-extract-text",
"params": {
"pdfPath": "<project-root>/data/documents/report.pdf",
"format": "text"
}
}
Extract pages 2-4 as JSON (one entry per page)
{
"tool": "document.pdf-extract-text",
"params": {
"pdfPath": "<project-root>/data/documents/report.pdf",
"pages": "2-4",
"format": "json"
}
}
Extract tables from first page of a PDF
{
"tool": "document.pdf-extract-tables",
"params": {
"pdfPath": "/tmp/financial-data.pdf",
"pages": "1"
}
}
Error Cases
| Scenario |
Behavior |
outputPath outside /tmp/ or data/documents/ |
Returns error with instructions |
| PDF file not found |
Returns error with path |
| pdftotext/pdftohtml not installed |
Returns error with apt install poppler-utils hint |
| No tables found in PDF |
Returns tables: [] (not an error) |
| Invalid page range |
Returns validation error |
| Playwright/Chromium crash |
Cleans up browser, returns error message |
| HTML is empty |
Returns validation error before launching browser |
Dependencies
| Component |
Source |
Notes |
| Playwright Chromium |
playwright-core npm package |
Already installed; browser binary at ~/.cache/ms-playwright/ |
| pdftotext |
system poppler-utils |
which pdftotext to verify |
| pdftohtml |
system poppler-utils |
Same package as pdftotext |
| marked (optional) |
marked npm package |
Used if installed; falls back to inline converter |
Performance
- PDF generation: ~2-5s for simple pages, up to 30s for complex HTML
- Text extraction: <5s for most PDFs (15s timeout)
- Table extraction: <10s for most PDFs (15s timeout)
- All tools clean up resources (browser, temp files) on both success and error
Matrixx0070