AILatest-Studio/ailatest-skill

Write Natural Science Fund Proposals

Write as a proposal co-author and final editor. Preserve the user's latest document, established terminology, and formal mechanism-oriented voice.

Core Workflow

1. Locate the newest user-edited DOCX. Treat it as the editing base and preserve it.
2. Identify style references. Convert referenced PDFs to Markdown before substantive reading, then inspect original PDF pages for layout-sensitive details.
3. Extract the proposal title, two scientific questions, three research contents, corresponding methods, limitations, figures, outcomes, annual plan, and achievements.
4. Build a consistency matrix before rewriting. Use references/consistency-and-facts.md.
5. Revise surgically in the existing structure. Follow the user's voice in references/voice-and-structure.md.
6. Verify publications, journals, years, volumes, pages/article numbers, SCI claims, rankings, and cited facts with primary sources.
7. Apply proposal-specific Word formatting and final checks from references/docx-and-final-qa.md.
8. Render the whole DOCX, inspect every page, fix layout defects, and rerender.
9. Run scripts/audit_proposal_docx.py and resolve errors before delivery.

Logic Contract

• Use exactly two scientific questions for this small project.
• Use exactly three major research-content blocks.
• Map each research-content block to one expanded research-scheme block and one corresponding limitation/gap.
• Keep content focused on what is studied; put detailed tools, models, sensor specifications, and procedures in the scheme/method section.
• Keep figures, captions,正文 references, outcomes, and annual tasks consistent with the same three-block logic.
• Keep the annual plan separate from the technical route.
• Keep the final application/service target to one sentence.
• Reconcile every later edit across the full proposal instead of changing one isolated paragraph.

Writing Contract

• Use formal Chinese academic prose with explicit objects, mechanisms, variables, causal links, evidence, and validation paths.
• Move from limitation or contradiction to the proposed mechanism and then to expected scientific contribution.
• Prefer precise verbs such as 构建, 表征, 解析, 识别, 量化, 揭示, 验证, and 界定.
• Use inline numbered logic such as ①②③ when it improves dense proposal prose.
• Avoid marketing language, inflated claims, generic AI phrasing, and decorative headings.
• Never write process language such as 国家基金本子, 参考已有文件, 沿用某本子, or 根据修改指令 in the submission text.
• Avoid a paragraph ending with a last visual line containing only one Chinese character. Fix wording or spacing after rendering.

Personal Formatting Rules

• Bold 翁建涛 and Jiantao Weng in the achievements section.
• Put 6.1.7 知识产权与设备方法基础。 on its own bold line. Start the explanatory paragraph on the next line.
• Format numeric in-text citations such as [10,15-21] as superscript. Do not superscript bibliography entries or [J], [M], and [P] document-type markers.
• Keep 申请软件著作权2项 consistent across annual plan and expected outcomes.
• Include noise level whenever listing temperature/humidity, CO2, PM2.5, and illuminance environmental sensors.
• Remove unresolved placeholders such as 需填, 待补, or bracketed sensor specifications.
• Use the figure-caption and diagram rules from $design-natural-science-fund-figures when figures are part of the task.

Evidence And Integrity

• Never invent citations, DOI values, indexing status, quartiles, TOP status, corresponding-author roles, equipment ranges, or accuracy values.
• Browse current primary sources for time-sensitive bibliographic and ranking claims.
• Prefer publisher pages, Crossref, official journal pages, Web of Science/Clarivate records when accessible, and manufacturer datasheets.
• Mark uncertain facts for verification instead of presenting them as settled.
• Keep quoted source text minimal and paraphrase in the user's voice.

Deliverables

• Save a new versioned DOCX; do not overwrite the user's source.
• Preserve unrelated formatting and user edits.
• Deliver only the requested final artifact unless the user asks for audit reports or intermediates.

Bundled example

See assets/proposal-core-logic.png for a reconstructed diagram of the proposal's argument skeleton: basis/gap → two scientific questions → three research contents and evidence → contribution and application boundary. The example is synthetic and does not use a user-supplied proposal or project data.

Resources

• Read references/voice-and-structure.md before drafting or rewriting正文.
• Read references/consistency-and-facts.md before changing research logic, sensors, achievements, or bibliography claims.
• Read references/docx-and-final-qa.md before finalizing a DOCX.
• Run python scripts/audit_proposal_docx.py proposal.docx for deterministic checks.