SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsfund-literature-review-writer

fund-literature-review-writer

Writing & research

View repository

Updated 10-15 22:15

About this skill

NSFC Literature Review and Literature Commentary Writing Skill

Core Positioning

This skill is not "writing literature," buttransforming literature into "a part of the NSFC research argumentation structure"。

On the basis of the user's existing research concept, learn the "literature review—literature commentary" organization approach in successful NSFC projects, and carry out academic reconstruction, logical organization optimization, critical synthesis, and standardized output oriented to the NSFC expression system for the literature section.

Trigger Conditions

Triggered when the user needs to write the following content:

  • "1.2 Domestic and International Research Status and Development Trend Analysis" of an NSFC proposal
  • The "literature review" section of an NSFC proposal
  • The "literature commentary" section of an NSFC proposal
  • The user explicitly requests "write a literature review," "write literature commentary," or "write research status"

Core Principles

  1. Do not presuppose a fixed structure: The structure of the literature review is "learned from" 16 successful NSFC proposals, not "hard-coded"
  2. Do not stack literature abstracts: Literature abstract-style output is prohibited; literature must be integrated into the argumentation
  3. Do not depart from the NSFC context: Literature serves the NSFC research argumentation structure
  4. Emphasize structural synthesis ability: Synthesize organizational logic from proposals, do not solidify templates
  5. Emphasize critical commentary ability: Literature commentary is structural judgment, not summary
  6. Emphasize the ability of literature to serve the research question: A complete argumentation chain from literature to research question/method/innovation

Skill Scope

This skill is only responsible for the 1.2 Domestic and International Research Status and Development Trend Analysis + Literature Commentary section of the project basis in an NSFC proposal. It is necessary to read the user's research concept to ensure that the literature serves the research argumentation.

Knowledge Base References

Before executing this skill, the following knowledge base files must be read:

knowledge_base/

  1. literature_review_style_guide.md: Literature Review Writing Style Guide

    • The overall narrative logic of the literature review
    • Theme division strategies (4 methods)
    • Paragraph organization norms (argumentation function, not fixed sentence patterns)
    • Literature citation format
    • Patterns of connective usage between literature items
    • Table usage norms
    • Transition methods from review to commentary
    • Patterns of critical expression
    • Word count norms

  2. literature_review_patterns.md: Literature Review Organizational Logic Paradigms

    • Typical literature review organization for five project types
    • Differences in literature organization for method-type/mechanism-type/system-type
    • The construction logic of literature clusters
    • Mapping patterns from literature to research question/method/innovation
    • The 4 types of critical judgment in literature commentary
    • Decision tree for literature organization choices

  3. literature_mapping_rules.md: Rules for Mapping Literature to Research Question/Method/Innovation

    • Rules for mapping literature to research question
    • Rules for mapping literature to method rationality
    • Rules for mapping literature to innovation rationality
    • Alignment mechanism with the user's research concept

references/

  1. info_form.md: User input information table
  2. output_skeletons.md: Output skeleton (argumentative function, not a fixed template)
  3. validation_menu.md: Validation menu
  4. anti_patterns.md: Anti-pattern checklist
  5. literature_search_strategy.md: Literature search strategy
  6. fund-search-workflow.md: Complete search workflow (script call chain, fallback strategy)

search/ (internalized search package)

  • search/scripts/: Search scripts (search_all.py, citation_finder.py, rank_and_filter.py, format_bibtex.py, dedup.py, plus internal dependencies such as tier_utils/exa_search/search_google_scholar, which do not need to be called directly)
  • search/references/: Search reference documents (search-strategy.md, api-reference.md, data-schema.md, journal-tiers.md)
  • search/data/: Journal data (priority_journals.csv, blacklist_journals.csv)
  • search/.env.example: API key template (must be copied to .env and filled in)
  • search/requirements.txt: Python dependencies (requests, exa-py, scholarly)

Workflow

Step 1: Understand the research objectives

  • Read the user's research concept (research topic, research objectives)
  • Determine the overall direction of the literature review

Step 2: Parse the research content

  • Read the user's research content (usually 3-5 items)
  • Determine the thematic division of the literature review

Step 3: Identify the key scientific questions

  • Read the user's key scientific questions (usually 1-3)
  • Determine the research gap direction for the literature commentary

Step 4: Search the literature

  • Read references/fund-search-workflow.md for the complete search process

4a. Construct topic search queries: Generate English query groups from the research topic/content/methods/key scientific questions

4b. English literature search(four sources in parallel):

# 设置环境变量（从 search/.env.example 复制为 search/.env 并填写）
# 对每个英文查询组执行：
python search/scripts/search_all.py --query "{英文查询}" --email "$OPENALEX_EMAIL" --output runs/cf/query_N.json
# 合并所有查询组：
python search/scripts/citation_finder.py merge runs/cf/query_*.json --output runs/cf/merged.json
# 补算期刊层级：
python search/scripts/citation_finder.py enrich-tiers --input runs/cf/merged.json --priority-csv search/data/priority_journals.csv --blacklist-csv search/data/blacklist_journals.csv --output runs/cf/enriched.json

4c. Zotero MCP / local file search(optional): follow fund-search-workflow.md Step 4/5

4d. Merge, deduplicate, and rank across all sources：

python search/scripts/dedup.py --input runs/all_merged.json --output runs/deduped.json
python search/scripts/rank_and_filter.py --input runs/deduped.json --weights relevance=0.4,journal_tier=0.3,citation_count=0.2,recency=0.1 --output runs/final.json
python search/scripts/format_bibtex.py --input runs/final.json --output runs/nsfc_references.bib

Step 5: Confirm the literature set

  • Directly use the runs/final.jsonoutput by Step 4d (already deduplicated, ranked, and filtered)
  • To adjust the filtering weights or priorities, see references/literature_search_strategy.md

Step 6: Automatically summarize the literature organization approach

  • Based on literature_review_patterns.md , select the literature organization approach
  • Select a typical organization pattern based on the project type
  • Divide literature themes based on the research content
  • Determine the focus of the methods introduction based on the research methods

Step 7: Generate the literature review

  • According to literature_review_style_guide.md and output_skeletons.md , generate the literature review
  • Thematic division strategies (4 approaches, selected based on research content)
  • Paragraph organization (argumentative function: introduction -> examples -> comparison -> transition -> summary)
  • Literature citation format (author-year system preferred)
  • Table usage (used by default for methods-research type)
  • Embedded commentary (at the end of each thematic paragraph)

Step 8: Generate the literature commentary (independent module)

  • According to output_skeletons.md , generate the literature commentary
  • Commentary approach (embedded/independent, selected based on project type)
  • Critical judgment (methodological/data applicability/model explanatory power/research boundaries)
  • Lead into this project's research direction

Step 9: Validate consistency with the user's research concept

  • According to validation_menu.md , verify item by item
  • Consistency check between the literature review and the research content
  • Correspondence check between the literature commentary and the research questions
  • Closed-loop check of literature → method → innovation
  • Citation format consistency check
  • Word count check
  • Literature quality check
  • Anti-pattern check (according to anti_patterns.md）

Adaptation mechanism

User already has literature

  • Prioritize the user's existing literature
  • Supplementarily search for missing literature
  • Verify the quality and relevance of the user's literature

User has no literature

  • Fully rely on automatic retrieval
  • According to literature_search_strategy.md execute retrieval
  • Generate a complete BibTeX citation library

User has partial literature

  • Build on the user's existing literature
  • Supplementarily search for missing literature
  • Ensure the literature covers all research content

Output format

Output 1: Literature review main body + literature commentary (MD file)

# 1.2 国内外研究现状及发展动态分析

## 1.2.1 主题1（对应研究内容1）
[引入段]
[子主题1.1 文献群]
[子主题1.2 文献群]
[主题评述段（嵌入式评述）]

## 1.2.2 主题2（对应研究内容2）
[引入段]
[子主题2.1 文献群]
[子主题2.2 文献群]
[主题评述段（嵌入式评述）]

## 1.2.3 主题3（对应研究内容3）
[引入段]
[子主题3.1 文献群]
[子主题3.2 文献群]
[主题评述段（嵌入式评述）]

## 1.2.X 文献评述（独立评述小节，可选）
[开头段]
[主题评述段1]
[主题评述段2]
[主题评述段3]
[总结段]

## 参考文献
[参考文献列表]

Output 2: BibTeX format citation library

% NSFC申请书文献综述引用库
% 生成时间：YYYY-MM-DD
% 项目主题：XXX

@article{key1,
  title = {文献标题},
  author = {作者},
  journal = {期刊},
  year = {年份},
  doi = {DOI}
}

Citation format requirements

In-text citation

  • Author-year system(preferred):作者等(年份) or Author et al.(年份)
  • Numbered system(adaptive to user needs):[1]、[2]

Reference output (sole format)

  • BibTeX format(mandatory)
  • Must have complete fields (title/author/journal/year)
  • Preserve DOI whenever possible
  • Missing key information is not allowed (if unknown, it must be marked)
  • For direct import into Zotero / BibTeX

Language constraints

  • Except for necessary terminology, file names, titles, etc., use Chinese
  • Mixed Chinese-English citations (for Chinese literature use "Name et al. (year)", for English literature use "Author et al. (year)")

Prohibited actions

  1. ❌ Abstract-style output of literature is prohibited
  2. ❌ Simply listing papers is prohibited
  3. ❌ Timeline-style review is prohibited
  4. ❌ Fixed templated writing is prohibited (paragraph/sentence templates must not be preset)
  5. ❌ Organizing literature disconnected from the research topic is prohibited
  6. ❌ Literature forming an independent system is prohibited
  7. ❌ Literature commentary that is only a summary without critical judgment is prohibited
  8. ❌ Calling sub-agents is prohibited
  9. ❌ Using PubMed and arXiv for retrieval is prohibited
  10. ❌ Marketing articles, unsourced content, secondary reposted data, and non-academic summary materials are prohibited
  11. ❌ Using claim_extractor.py and support_llm.py is prohibited (item-by-item claim evaluation, not the purpose of this skill)
  12. ❌ Downloading full texts is prohibited (CNKI/ScienceDirect only collect metadata, do not click Download/View PDF)

Final principle

The structure of the literature review should be "learned" from successful NSFC proposals, not "hard-coded."

Literature does not form an independent system; it serves the user's research conception. Every piece of literature should find its place in the argumentation chain of "literature → research question/method/innovation."