literature-reader

Literature Reader — Literature Reading and Summary Generation

Instructions

Step 1: Obtain PDF Content

Preferred method: Use Cursor's built-in Read tool to directly read the PDF file.

Read tool → path: "/path/to/paper.pdf"

The Read tool automatically converts PDF to text. If the PDF is too large and gets truncated, read it in segments:

Read tool → path: "/path/to/paper.pdf", offset: 1, limit: 200
Read tool → path: "/path/to/paper.pdf", offset: 201, limit: 200

Alternative method: If the Read tool cannot read the PDF (returns empty content, garbled text, severe truncation), or when encountering scanned documents, complex layouts, etc., please refer to PDF Text Extraction Guide to use the alternative extraction script.

Quick usage:

pip3 install pdfplumber  # 首次使用需安装
python3 ~/.cursor/skills/literature-reader/scripts/extract_pdf.py "/path/to/paper.pdf" --output /tmp/paper_text.txt

Expected output: the terminal displays Extracted text saved to: /tmp/paper_text.txt, then use the Read tool to read that file. For complete parameter descriptions, judgment criteria, and troubleshooting, see references/pdf-extract.md。

Step 2: Content Analysis

Analyze the extracted text one by one according to the following dimensions:

1. Metadata extraction — title, authors, institutions, publication year, journal/conference
2. Research question identification — what question the paper attempts to answer, what pain point it addresses
3. Methodology analysis — what research methods, technical approaches, and experimental designs are used
4. Core contribution summary — the paper's main innovations and academic contributions (usually explicitly listed at the end of the Introduction or in the Abstract)
5. Experiments and results — key experimental settings, datasets, metrics, quantitative results
6. Limitations and future directions — shortcomings acknowledged by the authors, follow-up work proposed

Step 3: Generate Literature Summary

Use the template to generate structured output. See the template at summary-template.md。

The output format is a Markdown file, suggested filename:[第一作者姓氏]_[年份]_[关键词缩写].md

Step 4: Quality Check

After generating the summary, perform a self-check:

• Are the metadata (title, authors, year) accurate
• Is the research question clearly stated
• Does the method description capture the core technical approach rather than trivial details
• Are the core contributions consistent with the paper's Abstract/Introduction
• Are key data and conclusions supported by specific numerical values
• Is the personal evaluation objective and well-founded

Special Scenario Handling

Batch Literature Reading

When users provide multiple papers:

1. Process each paper according to the above workflow
2. At the end, additionally output a cross-comparison table, including: title, year, method, dataset, core metrics, main contributions
3. Briefly explain the relationships between the papers (complementary/competitive/progressive)

See the cross-comparison table template at summary-template.md bottom.

In-depth Reading of Specific Sections

When users only care about a certain part of the paper (such as methodology, experimental results):

1. Only expand the detailed analysis of that part
2. Keep the remaining parts as a one-sentence summary
3. Interpret key formulas and algorithm pseudocode

Chinese papers

For Chinese papers, output the summary in Chinese. For English terms cited in Chinese papers, keep the original text and annotate them in parentheses, e.g.: 注意力机制（Attention Mechanism）.

Output specifications

• Literature summary length: 1000–2000 characters (adjusted according to paper complexity)
• Language: consistent with the paper's language, or follow the language specified by the user
• Format: Markdown, using clear hierarchical headings
• Annotate the original English text when key terms first appear
• Data and conclusions must be annotated with their source section (e.g. "see Table 3" "Section 4.2")

Examples

Example 1: Close reading of a single paper

The user says: "Help me read this paper /Users/me/papers/attention.pdf" Operation:

1. Use the Read tool to read the PDF content
2. Analyze one by one according to the dimensions in Step 2
3. Generate a complete literature summary according to the template Result: Output a structured Markdown summary file

Example 2: Batch comparison

The user says: "Compare these three papers /papers/a.pdf /papers/b.pdf /papers/c.pdf" Operation:

1. Read and generate a summary for each paper one by one
2. Output a horizontal comparison table
3. Analyze the complementary/progressive/competitive relationships among the papers Result: Three independent summaries + one horizontal comparative analysis

Example 3: In-depth reading of a section

The user says: "How exactly does the method section of this paper work?" Operation:

1. Read the PDF and locate the methodology section
2. Interpret in detail the technical approach, key formulas, and algorithms
3. Summarize the remaining sections in one sentence each Result: A deep analysis focused on the methodology

Troubleshooting

The Read tool reads the PDF as empty or garbled text

Cause: The PDF is a scanned copy (image format) or uses a special encoding Solution: Use an alternative extraction script instead; see PDF Text Extraction Guide

PDF content is truncated

Cause: The PDF has too many pages, and a single read exceeds the limit Solution:

1. Use the Read tool to read in segments, 200 lines each time, until all content is covered
2. Or use the alternative extraction script's --pages parameter to extract by range; see references/pdf-extract.md

pdfplumber installation fails or the script reports an error

See PDF Text Extraction Guide - Troubleshooting for the complete troubleshooting steps