Patent Claim → Chinese Bilingual Annotated Comparison

Turn a patent's independent claims into a Chinese-convention deliverable: each
feature carries its specification reference numeral (N), and the result is shown
as English original (top) / Chinese translation (bottom) per claim, with a
proper-noun glossary.

When to use

• User gives a patent number (e.g. US12636975B2) or uploads a patent PDF and asks
  to extract independent claims, annotate features with reference numerals, translate
  to Chinese, or produce a top/bottom bilingual layout.
• User says "中英对照", "权要标号", "独立权利要求", "给特征加附图标记", etc.

Workflow

Follow these steps in order. Skip a step only if the user explicitly scopes it out.

Step 1 — Fetch the full patent text

1. Prefer the connected PatSnap connector mcp__patsnap-search__patsnap_fetch:
   - keys=[<patent_number>], key_type="pn", module=["basic"],
   include_structured=true.
   - This returns bibliographic + full claims + description. The structured payload
   is the most reliable source.
2. Do not rely on WebFetch for Google Patents / USPTO PDF — it is frequently
   blocked or returns only the abstract.
3. Do not use curl for the PDF — the Bash sandbox has no network (exit 7).

Step 2 — Recover the fetched payload into the workspace

The connector result may be persisted to a protected path that the Read tool
cannot open directly.

1. Copy it into the workspace with PowerShell (Bash cp / coreutils are often
   missing in this environment):
   powershell Copy-Item -Path "<protected-path>" -Destination "E:\05-WorkBuddy\<session>\patent_struct.json" -Force
2. The claims/description fields are long single lines that get truncated by Read.
   Use the managed Python to slice them out:
   bash PYTHONPATH="C:\Users\Administrator\.workbuddy\binaries\python\envs\default\Lib\site-packages" \ "C:\Users\Administrator\.workbuddy\binaries\python\versions\3.13.12\python.exe" extract.py
   See references/workflow.md for the extraction snippet.

Step 3 — Identify independent claims

• Independent claims open with A/An <subject> comprising: and do not reference
  any other claim.
• Dependent claims begin The <subject> of claim N, wherein ... (or according to claim N). Record which claim each dependent claim cites; the cited roots are the
  independent claims.
• Extract the verbatim text of every independent claim (preserve English exactly).

Step 4 — Map features to reference numerals

• US claims usually carry no numerals. Pull the numerals from the first
  embodiment in the detailed description (most complete set, e.g. FIG. 1–2).
• Search the description for phrases like vehicle body 3, floor 9,
  first seat base 41, second seat bottom 32, armrest 34, contact portion 61.
• For a generalized claim feature (e.g. claim 14 "a seat"), pick the
  representative embodiment (e.g. rear seat 31/32/33) and explicitly note the
  alternative (front seat 27/28/29) in the deliverable.
• Composite surfaces keep the spec's own label (e.g. loading surface (SF1)).

Step 5 — Annotate and translate (top / bottom layout)

For each independent claim, emit two blocks:

英文原文（带附图标记）： the verbatim English, with (N) inserted after each
feature word. Repeat the numeral every time the feature recurs (Chinese-patent
style: vehicle body(3), second seat bottom(32)). Do not alter the English.

中文译文： sentence-by-sentence Chinese translation directly below. Keep
terminology consistent across claims (e.g. seat bottom→座垫, seat back→椅背,
loading surface→装载面, normal/retracted position→正常使用位置/收起位置).

Step 6 — Proper-noun glossary

At the end, add a table preserving verbatim originals + Chinese notes:
US12636975B2, KAWASAKI MOTORS, LTD., SF1, and any non-standard translated
term (e.g. developable ← Japanese 展開可能, meaning "可展开/可转换" — flag it
explicitly so it is not misread).

Step 7 — Deliver

Write the result to a Markdown file in the workspace (e.g.
<pn>_独立权利要求_中英对照.md) and call present_files.

Notes / extensions

• To also annotate dependent claims (2–13, 15–17, …), apply Steps 4–6 per claim;
  the same numeral set from the first embodiment usually carries over.
• If the user wants Word/DOCX output, route the Markdown through the
  tencent-docx skill after this step.
• See references/example-us12636975b2.md for a fully worked mapping, and
  assets/template_bilingual.md for the output skeleton.