survey-generation
lingzhi227/agent-research-skills
Generate comprehensive academic survey papers with structured outlines, RAG-based writing, and validated citations.
What is survey-generation?
Automates creation of complete academic survey papers by collecting relevant literature, generating multi-perspective outlines, writing subsections with citations, and validating coherence. Use this when you need to produce a structured literature survey on a research topic with proper citations and logical flow.
- Parallel multi-LLM outline generation to capture diverse perspectives
- RAG-based subsection writing with inline citations from collected papers
- Citation validation using NLI to ensure claims are supported by sources
- Local coherence enhancement across subsections with smooth transitions
- Automatic BibTeX generation and citation key management
- Structured output with main.tex, references.bib, outline.json, and section files
How to install survey-generation
npx skills add https://github.com/lingzhi227/agent-research-skills --skill survey-generation- Python environment with semantic-scholar and arXiv access
- Access to deep-research or literature-search skill for paper collection
- LLM API credentials for parallel outline generation
- BibTeX-compatible LaTeX environment for output
How to use survey-generation
- 1.Provide a survey topic or research area as input
- 2.Collect 50-200 relevant papers using the literature search script (search_semantic_scholar.py)
- 3.Run the multi-LLM parallel outline generation to create N independent outlines
- 4.Merge outlines into a single comprehensive structure with subsections
- 5.For each subsection, retrieve relevant papers and generate content with citations
- 6.Validate all citations using NLI to ensure claims are supported
- 7.Enhance local coherence by refining transitions between subsections
- 8.Convert citations to BibTeX format and generate the final references.bib file
Use cases
- Writing a comprehensive literature survey for a research proposal or thesis
- Generating background sections for academic papers on emerging topics
- Creating structured overviews of a research area for team onboarding
- Producing survey papers for publication with validated citations
- Synthesizing 50-200 papers into a coherent narrative with proper attribution
- Academic researchers and PhD students
- Literature review authors
- Research team leads synthesizing domain knowledge
- Authors preparing survey papers for publication
survey-generation FAQ
Only papers from the collected paper list provided in Step 1. The skill enforces this rule and will not hallucinate citations to papers not in your collection.
The workflow recommends 50-200 papers filtered by relevance and citation count. Start with a semantic scholar search and filter by quality.
Yes. The output includes individual section files and an outline.json, allowing you to review and refine content before final compilation.
A complete LaTeX survey with main.tex, references.bib (BibTeX), outline.json (structure), and individual section files for modularity.
The skill uses Natural Language Inference (NLI) to verify that cited papers actually support the claims made in each subsection, removing or correcting unsupported citations.
Full instructions (SKILL.md)
Source of truth, from lingzhi227/agent-research-skills.
name: survey-generation description: Generate complete academic survey papers using multi-LLM parallel outline generation, RAG-based subsection writing, citation validation, and local coherence enhancement. Based on AutoSurvey pipeline. Use for writing comprehensive literature surveys. argument-hint: [topic]
Survey Generation
Generate complete academic survey papers with structured outline, RAG-based writing, and citation validation.
Input
$0— Survey topic or research area
Scripts
Literature search
python ~/.claude/skills/deep-research/scripts/search_semantic_scholar.py \
--query "relevant search query" --max-results 50
References
- Survey prompts (outline, writing, citation, coherence):
~/.claude/skills/survey-generation/references/survey-prompts.md
Workflow (from AutoSurvey)
Step 1: Collect Papers
- Search Semantic Scholar / arXiv for papers on the topic
- Collect 50-200 relevant papers with titles and abstracts
- Filter by relevance and citation count
Step 2: Generate Outline (Multi-LLM Parallel)
- Generate N rough outlines independently (parallel)
- Merge outlines into a single comprehensive outline
- Expand each section into subsections
- Edit final outline to remove redundancies
Step 3: Write Subsections (RAG-Based)
For each subsection:
- Retrieve relevant papers for the subsection topic
- Generate content with inline citations
[paper_title] - Enforce minimum word count per subsection
- Only cite papers from the provided list
Step 4: Validate Citations
For each subsection:
- Check that cited paper titles are correct
- Verify citations support the claims made
- Remove or correct unsupported citations
- Use NLI (Natural Language Inference) for claim-source faithfulness
Step 5: Enhance Local Coherence
For each subsection:
- Read previous and following subsections
- Refine transitions and flow
- Preserve core content and citations
- Ensure smooth reading experience
Step 6: Convert Citations to BibTeX
- Replace
[paper_title]with\cite{key} - Generate BibTeX entries for all cited papers
- Validate all citation keys exist in .bib file
Output Structure
survey/
├── main.tex # Complete survey paper
├── references.bib # All citations
├── outline.json # Survey outline
└── sections/ # Individual section files
Rules
- Only cite papers from the collected paper list — never hallucinate citations
- Each subsection must meet minimum word count
- No duplicate subsections across sections
- Citation validation is mandatory before final output
- Local coherence enhancement must preserve all citations
- The survey should be comprehensive and logically organized
Related Skills
- Upstream: deep-research, literature-search, literature-review
- See also: related-work-writing
Related skills
More from lingzhi227/agent-research-skills and the wider catalog.

symbolic-equation
Discover interpretable scientific equations from data using LLM-guided evolutionary search.

table-generation
Generate publication-quality LaTeX tables from experimental results with booktabs styling and bold best results.

algorithm-design
Design algorithms with LaTeX pseudocode and UML diagrams for formal method documentation.

atomic-decomposition
Decompose research papers into atomic concepts with bidirectional math-to-code mapping.

linkfox-multimodal-recognize-image
基于多模态AI的图片识别与分析。当用户想分析、描述、从图片URL中提取信息、image recognition, image analysis, image description, image content understanding, OCR text recognition, visual Q&A时触发此技能。当用户提到图片识别、图片分析、图片描述、识别图片内容、分析产品图、从图片中读取文字、描述图片、提取视觉内容或理解照片内容时触发。当用户提供图片URL并就其视觉内容提问时,即使未明确说"图片识别",也应触发此技能。

anti-render
Generate ideal-vs-reality visual comparisons by analyzing images and creating side-by-side contrasts.