Dr Cook:grant Writer

grant-writer

1. Overview

grant-writer drafts competitive NSFC grant proposals and general research proposals. Three modes: section mode drafts one section (rationale, innovation, methods, or budget); full mode drafts all sections sequentially; review mode evaluates a draft against NSFC criteria. When upstream modules (literature-reviewer, research-gap-finder) have run, grant-writer inherits their key_findings and summary automatically — no re-collection needed.

2. Parameters

Required

Optional

Parameter collection rule

If required parameters are missing and cannot be inherited from upstream context, ask one at a time: grant_type first, then section. Do not ask for optional parameters unless the user has mentioned them or they are needed for the current section.

3. Workflow

Step 1 — Check upstream context. Inspect context_output before asking anything. Extract as already-provided: domain, key_findings, summary, parameters.target_journal, language, all parameters.*. If upstream status is failed, halt and inform the user.

Step 2 — Collect missing required parameters. Identify which required parameters (grant_type, section) are absent. Ask for each one at a time: grant_type first, then section. Wait for the user's answer before asking for the next.

Step 3 — Collect optional context. Ask for research_area and prior_work only if not inferable from upstream context. If upstream summary from literature-reviewer or research-gap-finder is available, present it to the user as the proposed research_area basis and ask for confirmation before proceeding. Do not ask for domain if it is already present in context_output.parameters.domain.

Step 4 — Load reference files. Load the reference files appropriate for the requested section:

• All sections → load references/nsfc-format.md for format constraints and character limits.
• section: rationale or section: innovation → also load references/significance-innovation.md.
• section: budget → also load references/budget-guidelines.md.
• section: full → load all three reference files.
• section: review → load references/nsfc-format.md and references/significance-innovation.md for review criteria.

Step 5 — Draft the requested section. Compose the grant text using structural frameworks from loaded references, content from upstream key_findings and summary, and the user-provided research_area and prior_work.

• Rationale (立项依据): Background → current research status → gaps → how this project addresses gaps → expected significance.
• Innovation (创新性): 3–5 numbered points. Each: [研究角度/技术方法/理论框架] + [what is new] + [comparison to prior approaches]. Apply taxonomy from references/significance-innovation.md.
• Methods (研究方案): Hypothesis → specific aims → design per aim → expected results → contingency plans.
• Budget: 劳务费 → 设备费 → 耗材费 → 国际合作与交流费 → 出版/文献/信息费 → 间接费用. Apply constraints from references/budget-guidelines.md. Present as a table.
• Full: Execute rationale → innovation → methods → budget sequentially, each within character limits. Output all four sections in one continuous response, separated by --- horizontal rules and bold section headers (立项依据, 创新性, 研究方案, 经费预算). Issue the iteration prompt (Step 7) once at the end.
• Review: Evaluate draft against four NSFC criteria (科学意义 20%, 创新性 30%, 研究方案可行性 30%, 申请人研究基础 20%). Rate each ✅ / ⚠️ / ❌ with specific revision suggestions. Check for: (a) missing/thin Paragraph 3 (gap statement), (b) reference count <5, (c) missing Paragraph 5 impact statement, (d) innovation points with no comparison to existing approaches, (e) missing contingency plans in 研究方案.

Step 6 — Apply character limits and language. For Chinese grants (language: zh), output in Chinese using correct NSFC terminology. Verify the drafted text against character limits from references/nsfc-format.md:

• nsfc_general: 立项依据 ≤3000 characters, 研究方案 ≤8000 characters.
• nsfc_young: 立项依据 ≤2000 characters, 研究方案 ≤6000 characters.
• nsfc_key: 立项依据 ≤5000 characters, 研究方案 ≤12000 characters. Show character count at the end of each section draft.

Step 7 — Present draft and offer iteration. After presenting the draft, ask: "Would you like me to expand, compress to the word limit, translate to English for collaborators, or move to the next section?"

4. Output Format

Header line on every output:

[Grant: <grant_type_label> | Section: <section_label> | Language: <ZH|EN>]

For Chinese output (language: zh):

• Section heading in Chinese (e.g., 二、立项依据, 三、创新性)
• Paragraph text in Chinese following the structural framework

For innovation section — numbered list with bold aspect labels:

**创新点1：[研究视角的创新]**
本研究首次…现有研究多采用…本研究通过…

**创新点2：[方法学创新]**
…

For budget section — table format:

| 经费类别 | 金额（元） | 数量依据 |
|---------|----------|--------|
| 劳务费   | …        | …      |

Character count shown at section end:

[字符数: 1847/3000]

For section: review — four-criterion assessment block, each with ✅ / ⚠️ / ❌ rating, followed by a ranked improvement list.

5. context_output

Reads from upstream

Writes to output

{
  "module": "grant-writer",
  "summary": "NSFC general grant innovation section drafted, 3 innovation points",
  "raw_text": "<drafted grant section(s)>",
  "parameters": {
    "grant_type": "<collected>",
    "section": "<collected>",
    "language": "<collected>"
  },
  "status": "success | partial | failed",
  "error_message": null
}

Downstream readers of raw_text: chinese-polisher, english-polisher (if translation requested).

6. References

See references/ for:

• nsfc-format.md — Grant type comparison, character limits, review criteria weights, common rejection reasons, and format requirements
• significance-innovation.md — 立项依据 structural framework, innovation point structure and taxonomy, strength/weakness indicator phrases
• budget-guidelines.md — NSFC budget category breakdown, sample budget table, justification rules, and common budget mistakes