科研 Skill

实证经济学文献综述

empirical-economics-literature-review

获取全文并按分支撰写实证经济学文献综述

官方收录发布者 local-empirical-economics-literature-review安全审核通过v0.1.1更新于 2026-08-28

Skill 说明

根据研究方向和文献列表准确获取并阅读全文,将研究方向拆分为 3-5 个细分方向,再依据文献内容把全部文献梳理为通常 2-3 个分支;逐支总结核心内容,评述通常 2-3 项现有研究不足,并对应说明本研究如何填补研究空白 | 输入:研究方向;文献列表,每篇至少包含标题、DOI 或引用格式之一 | 输出:结构化文献综述 Word 文档和可选 LaTeX 文档

SKILL.md

Empirical Economics Literature Review

Keep every reviewed claim traceable to one canonical paper, matching full text, and a page or section locator.

References

  • Before search or PDF acquisition, read references/literature-retrieval.md.
  • Before reading papers, read references/evidence-extraction.md.
  • Before classification, read references/branch-classification.md.
  • Before drafting, read references/review-synthesis.md.
  • Before Word or LaTeX output, read references/document-export.md.

Workspace

Use the session workspace for all new files unless the user explicitly names another output location. Treat input paths as read-only. Persist request_snapshot.json, papers.json, retrieval_status.json, paper_evidence.json, branch_map.json, critique_map.json, review.md, references.bib, literature_review.docx, pdf/, and .cache/literature/. Create literature_review.tex only after the user requests LaTeX.

1. Normalize input

Require a research direction plus a literature list or discovery request. Accept pasted text and runtime-readable text, tables, documents, and citation files; never require JSON input.

Define “this study” only from the user's confirmed study_brief in request_snapshot.json. Require a concrete research object, focal relationship, and outcome dimension. Ask the user to expand umbrella directions such as “enterprise characteristics and enterprise innovation” before search; never infer design or contributions.

Detect format from content. A TXT may contain BibTeX and percent-comment groups. Preserve citation keys and groups as hints, not final branches. Each entry needs a DOI, identifiable title, author-year citation, or keywords. Ask for missing information, then use scripts/literature_library.py to persist normalized records.

2. Search and confirm candidates

The first literature tool call must discover, activate, or use a real Browser. Search by DOI, then title, then author-year plus research direction, then keywords.

Use CNKI for Chinese literature and Google Scholar for English literature. Verify candidates on real CNKI, journal, publisher, working-paper, author, or repository pages; snippets are discovery only.

Candidate-list confirmation

Show title, authors, year, venue, DOI, match basis, confidence, and version relation. Wait for confirmation before PDF acquisition. Never silently choose an ambiguous result.

3. Acquire and confirm full text

Acquire open publisher PDFs, author manuscripts, repository copies, or CNKI full text through the user's authenticated Browser.

If CNKI is unavailable, mark affected records login_required, finish independent English/open-access work, atomically save state, show completed and pending records, and ask the user to log in. Resume only pending and login_required records. Never inspect browser secrets or bypass access controls.

Full-text-list confirmation

Show every paper's acquired, abstract_only, upload_required, or inaccessible status and provenance. Wait for confirmation before reading. Only matching acquired full text may proceed.

4. Read and audit

Complete one paper's evidence record before opening the next. Record data, sample, variables, measurement, model and identification, fixed effects, standard errors, findings, innovations, research relevance, and review-writing structure.

Attach a page or section locator to every fact. Visually verify important OCR output. Leave absent fields empty. Run scripts/evidence_audit.py and resolve every hard error.

5. Classify and synthesize

Compare language, content, theory/empirics, macro/micro, method, and corpus-supported schemes. Prefer 2–3 branches; assign each paper to at most two. Use full-text evidence, not filenames or user hints, and record a reason for another branch count.

Draft by branch rather than as an annotated bibliography. Weight detail by relevance, order branches logically, use natural transitions, and learn general review structure without copying prose. Use canonical citation keys only.

Generate references.bib with scripts/literature_library.py export-bib, then run scripts/review_audit.py. Resolve missing confirmed papers, unknown citations, evidence gaps, and invalid memberships.

After branch synthesis, follow the synthesis reference for literature critique. Map evidence-bound shortcomings to responses supported by non-empty study_brief fields in critique_map.json. Run scripts/critique_audit.py and resolve every hard error.

6. Export Word

Read the export reference and run scripts/export_review.py. User templates override defaults. Treat export or structural-validation errors as blockers.

7. Offer LaTeX

After Word is ready, ask whether the user needs LaTeX and whether they have a template. Use it when supplied; otherwise generate the minimal default. Deliver the TeX and references.bib together.

Final checks

Every confirmed paper is cited or excluded with a reason; detailed claims resolve to the correct evidence; abstract-only records are not fully read; shortcomings and study responses correspond; memberships and branch count are valid; Word precedes optional LaTeX; all audits pass.

常见问题

实证经济学文献综述能做什么?

根据研究方向和文献列表准确获取并阅读全文,将研究方向拆分为 3-5 个细分方向,再依据文献内容把全部文献梳理为通常 2-3 个分支;逐支总结核心内容,评述通常 2-3 项现有研究不足,并对应说明本研究如何填补研究空白 | 输入:研究方向;文献列表,每篇至少包含标题、DOI 或引用格式之一 | 输出:结构化文献综述 Word 文档和可选 LaTeX 文档。

如何安装实证经济学文献综述?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「实证经济学文献综述」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

实证经济学文献综述需要哪些权限?

该科研 Skill声明需要:本地文件。安装前会再次列出,由你确认后才会生效。

实证经济学文献综述经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。