yitengchun833-gif/datasheet-cn-reader

Datasheet CN Reader

Create a separate output; never overwrite the source PDF.

Required workflow

1. Run datasheet-cn-reader extract SOURCE.pdf --output work/extraction.json.
2. Run datasheet-cn-reader manifest work/extraction.json --output work/qa_manifest.json.
3. Read translation-rules.md.
4. Create page translations:
   • For Agent translation, run datasheet-cn-reader translate work/extraction.json --output work/translation.json --provider draft, replace every 待人工确认 block, and preserve all protected tokens.
   • For a controlled regression baseline, use --provider baseline-html --baseline accepted.htm.
   • For an external translator, use --provider command --command ...; keep credentials outside prompts and files.
5. Run datasheet-cn-reader build work/extraction.json work/translation.json --output output/reader.html.
6. Read qa-contract.md, then run: datasheet-cn-reader validate work/qa_manifest.json work/translation.json output/reader.html --report output/qa-report.json.
7. Run datasheet-cn-reader review output/reader.html --output-dir output/render-review.
8. Deliver only when validation exits 0 and the QA report has status: passed.

Translation contract

• Treat the uploaded PDF as the only parameter authority.
• Translate English natural language only.
• Do not change product models, pin names, parameter symbols, formulas, values, units, test conditions, table order, or MIN/TYP/MAX column relationships.
• Do not silently update values from a website or another revision.
• If external material differs, report the difference without replacing the uploaded value.
• Mark uncertain extraction or translation as 待人工确认; never guess.
• Preserve every source page as one corresponding HTML page block.
• Use real HTML tables for structured data; do not substitute table screenshots.
• Keep original page images embedded as data URLs for offline traceability.

Failure rules

• Treat a nonzero validator exit as a hard stop.
• Do not rename a failing build as a verified build.
• Fix the exact page/table/token issue recorded in the QA report.
• Do not suppress a missing token or row without recording a narrowly scoped, evidence-based exemption.
• If OCR is required, record OCR use and confidence per page.

Interfaces

• CLI: datasheet-cn-reader --help
• Source-tree wrappers: scripts/*.py
• MCP stdio server: python -m datasheet_cn_reader.mcp_server
• Generic Agent prompt: adapters/universal-prompt.md
• Product adapters: adapters/{codex,claude-code,cursor,gemini-cli,hermes}/

Read technical-terms.md when standard Chinese terminology is uncertain.