SKILL.md 全文

Frontmatter

name	pdf
description	Read, create, extract, and visually verify PDF files in the AgentRun container using Poppler and Python document libraries.

PDF Skill

Use this skill for PDF parsing, creation, conversion, and render checks.

Runtime

The AgentRun image provides:

• python3 with reportlab, pdfplumber, pypdf, pymupdf, pillow
• pdftotext, pdftoppm, pdftocairo
• qpdf
• Noto CJK fonts

Rules

• Batched writing (hard limit): infrastructure kills any streaming response after ~4.5 minutes of continuous generation. Cap every single Write/Edit payload at ~150 lines and build long generation scripts or HTML sources in batches (skeleton first, then append one section per response with Edit); never regenerate a whole file to fix one part.
• Keep all generated files under the current workspace, preferably output/.
• For multi-step work, write a small .py script under tmp/ or output/ and run it with python3.
• Do not use python -c; inline Python is blocked by the runtime policy.
• Render created PDFs with pdftoppm -png and inspect extracted text or page images before reporting success.
• For PDF analysis or form work, start with scripts/inspect_pdf_batch.py. It extracts all pages, labels, annotations, widgets, and checkbox candidates and creates page/detail contact sheets in one command.
• Read summary.json first, then only the required slices of structure.json. Do not print the full structure for a large document.
• Inspect page contact sheets in one parallel visual turn. Do not read one page image per model turn.
• When closer crops are needed, write all rough regions to one JSON manifest and run scripts/create_region_contact_sheet.py once. Do not crop one field at a time.
• For PDF forms, read [forms.md](forms.md) and keep field discovery, filling, and verification batch-oriented.

Common Commands

pdftotext input.pdf output.txt
python3 scripts/inspect_pdf_batch.py input.pdf --output-dir tmp/pdf-inspection
python3 scripts/create_region_contact_sheet.py \
  --images-dir tmp/pdf-inspection/pages \
  --regions tmp/regions.json \
  --output-dir tmp/pdf-regions
python3 tmp/create_pdf.py