Ppx Parse

Parse PDFs and images into Markdown/JSON using the `ppx` CLI. Use when the user asks to OCR scanned PDFs or screenshots, extract tables from PDFs, convert PDF/image to Markdown, preserve document layout, inspect parsing output. Also triggers on: 解析PDF、图片转文字、扫描件识别、扫描件转文字、提取表格、 PDF转Markdown、文档解析、OCR识别、识别图片文字、解析图片、提取文档内容。

Use the local ppx CLI to parse PDFs and images into structured Markdown and JSON.

Runtime Requirements

• Use Python >= 3.12.
• Prefer installing PPX into a virtual environment instead of the system Python.
• If ppx is missing, read references/troubleshooting.md and create a virtual environment before installing dependencies.
• Keep this skill's frontmatter version synchronized from the repository pyproject.toml with scripts/sync_version.py.

Workflow

1. Confirm the runtime uses Python >= 3.12.
2. Check the runtime with scripts/check_ppx_env.sh.
3. If ppx is missing, create or use a virtual environment and install PPX there.
4. Choose parsing options:
   • Use --ocr auto by default.
   • Use --ocr yes for scanned PDFs or screenshots.
   • Use --ocr no for native PDFs when OCR causes noise.
   • Use --table auto by default.
   • Use --table llm only when the user needs highest table accuracy and an LLM backend is configured.
5. Run ppx parse <input> -o <output>.
6. Inspect the output folder and report the main artifacts:
   • doc.md
   • doc.json
   • pages/
   • images/ when figures are extracted
7. If parsing fails, summarize the failing step and load the relevant note from references/.

Common Commands

ppx parse report.pdf -o output/
ppx parse scan.pdf --ocr yes -o output/
ppx parse figure.png -o output/
ppx parse report.pdf --pages "1-5,10" -o output/
ppx parse report.pdf --table llm --backend deepseek -o output/

Output Contract

• Prefer returning the absolute output directory.
• Mention whether the result came from doc.md, doc.json, or page-level files.
• Call out OCR mode, table mode, and backend when they materially affect accuracy.

References

• Read references/cli-options.md when choosing parse flags.
• Read references/backend-config.md when using DeepSeek, Paddle, or GLM backends.
• Read references/troubleshooting.md when PPX is missing, Python is too old, or runtime dependencies fail.