Literature Close Reading

Purpose

Use this skill to turn one academic paper, review article, thesis chapter, report, or article link into a rigorous close-reading Word report. Never return only a generic summary. Always produce a .docx file with exactly the seven required sections.

The default reader profile is: a doctoral student at Tsinghua Shenzhen International Graduate School, Institute for Ocean Engineering, researching digital ocean. Use this profile when judging transferability, research opportunities, and methodological relevance.

Inputs

Accept any of these:

• Local PDF file.
• Local Word file (.docx).
• Web URL or DOI.
• Plain text, abstract, or copied article sections.

If a URL, DOI, or current online article is provided, browse the web and use reliable source pages. If the paper itself is paywalled or inaccessible, clearly state the limitation and base the report only on accessible abstract/metadata unless the user provides the full text.

Workflow

1. Extract bibliographic information: title, authors, year, venue, DOI/URL if available.
2. Read the full text when available. For PDF/Word, inspect the document directly. For web articles, fetch the page and any accessible PDF.
3. Load references/seven-part-rubric.md and apply every section.
4. Avoid paper-by-paper generic paraphrase. Identify the paper's argument structure: why this problem matters, what is unresolved, what the authors actually did, what evidence supports it, and where the claim stops.
5. Write a report in Chinese by default unless the user asks for English.
6. Save the output as a Word .docx file. Use scripts/create_close_reading_docx.py when possible:
   powershell python scripts/create_close_reading_docx.py input.json output.docx
7. Return the absolute path to the .docx file and a 3-5 bullet preview of the most important insights.

Required Seven Sections

Always output these exact section headings, in this order:

1. 研究背景
2. 研究问题
3. 方法路线
4. 关键证据
5. 创新边界
6. 局限与争议
7. 对自己课题的可迁移启发

Do not add extra top-level analysis sections. Put metadata, citation, and one-line verdict before the seven sections.

Quality Bar

For each section, be specific enough that the reader can discuss the paper in a doctoral meeting:

• Name the engineering/scientific context, not just the field.
• State the exact unresolved problem or hypothesis.
• Describe the method as a route: data/model -> processing/simulation/algorithm -> validation -> conclusion.
• Separate evidence from author claims.
• Define the innovation boundary: what is new, what is inherited, what only works under certain conditions.
• Include limitations, uncertainty, assumptions, dataset/model bias, scalability, reproducibility, and possible contrary interpretations.
• Map transferability to digital ocean: ocean data assimilation, marine environmental simulation, ocean engineering digital twins, sensor networks, multi-source heterogeneous data fusion, marine structure monitoring, ocean-meteorology coupling, visualization platforms, and decision support.

Output File Naming

Use a concise safe filename:

YYYYMMDD_文献精读_<short-title>.docx

If the title is unknown, use:

YYYYMMDD_文献精读_report.docx

Save in the current workspace unless the user specifies another output directory.

Script Contract

The helper script expects JSON with this shape:

{
  "title": "Paper title",
  "authors": "Author A; Author B",
  "year": "2026",
  "venue": "Journal or conference",
  "doi_or_url": "https://...",
  "reader_profile": "清华大学深圳国际研究生院海洋工程研究院博士研究生，研究方向：数字海洋",
  "one_line_verdict": "一句话判断",
  "sections": {
    "研究背景": ["..."],
    "研究问题": ["..."],
    "方法路线": ["..."],
    "关键证据": ["..."],
    "创新边界": ["..."],
    "局限与争议": ["..."],
    "对自己课题的可迁移启发": ["..."]
  }
}

Use arrays of paragraphs or bullets. The script validates that all seven sections exist.