draft-patents-evidence-firstlisted

# Evidence-First Patent Workflow / 证据优先专利工作流 Use this Skill for internal drafting preparation under the China (`CN`) invention and utility-model workflow. Treat every model output as a proposal. A qualified patent professional and a human disclosure decision remain mandatory. **Core rule:** Draft from evidence, not imagination / 从证据出发，不把目标写成结果。 ## Start safely 1. Ask for the controlling request language, patent type, critical date, and confidentiality mode. Default to `local_only`. 2. Initialize a non-identifying case directory. Do not place original material outside `input/` or modify it. ```bash python scripts/init_case.py CASE_DIR --case-id demo-fixture --language zh-CN --patent-type invention python scripts/validate_case.py CASE_DIR ``` For `--language auto`, persist the language of the controlling user request before validation: ```bash python scripts/resolve_language.py CASE_DIR --language zh-CN ``` 3. Run `scan_sensitive.py` before any external research. In `local_only`, make no external request. In `sanitized_search`, show a human-approved sanitized query before sending it. In `approved_external`, record the scoped approval first. ## Follow the ten stages in order 1. Inventory inputs and establish the disclosure boundary. 2. Extract text, tables, figures, and case-relative locations; preserve source hashes. 3. Classify each statement as `measured`, `documented`, `inferred`, or `designed`. 4. Propose and deduplicate concepts while retaining source links. 5. R