research-plan

Research Plan

Core Features

This Skill systematically investigates top-tier journal/conference papers to formulate a personalized and actionable research analysis strategy for users.

Core Value

Evidence-driven: Based on the methodology of top-tier journal papers, rather than designing from scratch
Systematic investigation: Automated literature search, screening, and download workflow
Deep learning: Extract analysis methods and technical details from PDFs
Actionability: Output clear objectives/steps that can be directly executed

Use Cases

Formulate experimental design plans
Plan data analysis workflows
Design technical roadmaps
Determine statistical method selection
Formulate visualization strategies

Input Specifications

Required Input

Requirement description: Clearly describe the problem to be solved and the objectives

Optional Input

Working directory: Defaults to the current directory
Literature year range: Defaults to the most recent 5 years
Target journals/conferences: Specify particular sources
Language preference: Defaults to English
Maximum number of papers: Defaults to 30

Workflow

Input

Provide the necessary input according to the user's request and the configuration file; missing information should be explicitly listed and steps depending on that input should stop.

Execution Steps

For bugs caused by design flaws in this skill, first use bensz-collect-bugs to record them in a standardized way to ~/.bensz-skills/bugs/, do not directly modify the user's locally installed skill source code; if there is a workaround, record the bug first, then continue completing the task.
Only when the user explicitly requests public reporting such as "report bensz skills bugs", use the local gh to upload new bugs to huangwb8/bensz-bugs; do not pull / clone the entire repository.

Research Analysis Strategy Planning Assistant Based on Literature Investigation

Old name make-research-plan is retained only as a prompt compatibility alias; historical .make-research-plan/ is only for explicit compatibility reading, migration, or cleanup, and new runs uniformly use the task-level workspace.

Phase 0: Initialization

Input:

User requirement description
Working directory path (optional, defaults to the current directory)

Operations:

Verify that the working directory exists and is writable
Create a task-level hidden working directory .bensz-api/task-{yyyymmdd-hhmm}-{简短描述}/research-plan/
Clearly inform the user of the location and purpose of the created directory

Output:

.bensz-api/task-{yyyymmdd-hhmm}-{简短描述}/research-plan/ directory structure

Directory structure:

.bensz-api/task-{yyyymmdd-hhmm}-{简短描述}/research-plan/
├── input/papers/        # 下载的 PDF 文献
├── input/metadata/      # 调研元数据
│   ├── theme.json      # 主题和关键词
│   └── search_history.json # 检索历史
├── output/extracted/    # 提取的文献信息
│   └── papers_info.json # 论文结构化信息
├── output/analysis-framework.md # 分析框架总结
├── output/plan.md             # 任务内草稿
└── log/                       # 命令、验证与错误日志

Phase 1: Topic Extraction and Literature Investigation

Step 1.1: Topic Extraction

Objective: Structurally extract the research topic and keywords from the user's requirements

Method:

Analyze the user's requirement text
Identify the core research field
Extract key concept terms
Generate multiple related topics (if applicable)
Generate 3-5 keywords for each topic

Output: metadata/theme.json

{
  "primary_topic": {
    "name": "主题名称",
    "description": "主题描述",
    "keywords": ["关键词1", "关键词2", "关键词3"]
  },
  "secondary_topics": [
    {
      "name": "相关主题1",
      "keywords": ["关键词1", "关键词2"]
    }
  ],
  "extracted_at": "2026-01-19T09:30:00Z"
}

Step 1.2: Literature Search

Search Strategy:

Multi-source Search：
- PubMed (biomedical)
- Google Scholar (general)
- IEEE Xplore (engineering and technology)
- arXiv (preprints)
- Semantic Scholar (AI-driven)
Search Query Construction:
- Main topic keywords AND analysis method-related terms
- Prioritize literature from the last 5 years
- Limit to high-impact journals/top conferences
Quality Filtering:
- Prioritize peer-reviewed articles
- Exclude non-English literature (unless specified by the user)
- Prioritize highly cited papers

Output: metadata/search_history.json

{
  "searches": [
    {
      "topic": "主题名称",
      "query": "检索式",
      "source": "数据源",
      "results_count": 150,
      "timestamp": "2026-01-19T09:35:00Z"
    }
  ]
}

Step 1.3: Literature Screening and Scoring

AI evaluates each paper:

Semantic relevance score(1-10 points)
- Title match
- Abstract relevance
- Keyword overlap
Subtopic grouping：
- Methodology papers
- Application cases
- Review articles
- Tools/software
Prioritize selection of high-scoring papers：
- Keep papers with a score ≥ 7
- Ensure each subtopic has representative papers
- Keep the total number between 15-30 papers

Output: metadata/papers_scored.json

{
  "papers": [
    {
      "title": "论文标题",
      "authors": ["作者1", "作者2"],
      "year": 2024,
      "journal": "期刊名",
      "relevance_score": 9,
      "subtopic": "方法学",
      "doi": "10.xxxx/xxxxx",
      "url": "https://doi.org/..."
    }
  ]
}

Step 1.4: Generate Reference List

Output: references.bib

Format: standard BibTeX, including all necessary fields

Step 1.5: Download PDFs

Method:

Prioritize downloading via open access links
Use the Unpaywall API to find legal free versions
Save to the papers/ directory, naming format:{第一作者姓氏}_{年份}_{期刊简写}.pdf

Output: papers/*.pdf

Error Handling:

Record DOIs that failed to download
Do not block the subsequent workflow
State in the final report which papers could not be obtained

Phase 2: Deep Learning Analysis Strategy

Step 2.1: PDF Information Extraction

Goal: Extract analysis methods and technical details from each PDF

Extracted Content:

Study design: experiment type, control setup, sample size
Data preprocessing: cleaning steps, normalization methods, quality control
Statistical methods: test methods, multiple comparison correction, effect size
Visualization strategy: chart types, color schemes, tools
Software tools: R/Python packages, versions, parameter settings
Validation methods: cross-validation, bootstrap, sensitivity analysis

Method:

Use PDF reading tools to extract text and tables
AI semantic analysis identifies the content of the Methods section
Store the extracted information in a structured format

Output: extracted/papers_info.json

{
  "papers": [
    {
      "id": "paper_001",
      "title": "论文标题",
      "methods": {
        "study_design": "队列研究",
        "preprocessing": ["质量控制", "归一化"],
        "statistical_tests": ["t检验", "ANOVA"],
        "correction": "FDR",
        "visualization": ["箱线图", "热图"],
        "software": {
          "R": ["limma", "ggplot2"],
          "parameters": {"adjust.method": "BH"}
        }
      },
      "strengths": ["大样本量", "预注册"],
      "limitations": ["单中心研究"]
    }
  ]
}

Step 2.2: Comprehensive Analysis Framework

Goal: Synthesize all papers to distill best practices

Analysis dimensions:

Methodological consensus：
- Methods shared across multiple papers
- Standard procedures within the field
- Recommended best practices
Methodological differences：
- Applicable methods for different scenarios
- Emerging methods vs traditional methods
- Points of controversy
Tool ecosystem：
- Most commonly used software packages
- Comparison of tool pros and cons
- Version compatibility
Quality control：
- Data validation standards
- Result robustness checks
- Reproducibility practices

Output: analysis-framework.md

See template:research-plan/references/output-templates.md

Phase 3: Develop a personalized plan

Step 3.1: Requirements alignment

Analyze user requirements:

What is the research question?
What special constraints are there?
What is the expected output?
What is the skill level?

Map to the analysis framework:

Match relevant methodology sections
Identify applicable tools
Determine quality control points

Step 3.2: Plan generation

Output: plan.md

See template:research-plan/references/output-templates.md

Step 3.3: Plan review

Present to the user:

Plan overview
Explanation of key decision points
Comparison of alternative options
Invitation for revision suggestions

Review questions:

Does the plan cover your requirements?
Are the steps clear and feasible?
Is there any analysis missing?
Is there anything that needs adjustment?

For users

Be specific in describing requirements: avoid vague statements
Clarify constraints: data type, sample size, skill level, etc.
Review the plan carefully: this is your research, and you have the final say
Keep the working directory: for subsequent adjustments and reproduction

For AI execution

Transparency: clearly state what is being done at each step
Traceability: all recommendations have literature support
Conservatism: admit uncertainty rather than fabricate
User-centered: the plan must fit the user's actual constraints

For implementation/script status and responsibility boundaries, see:research-plan/references/implementation-notes.md

For change records, see:research-plan/CHANGELOG.md(version number subject to research-plan/config.yaml:skill_info.version )

Output

Core output

plan.md: final analysis strategy plan (user-friendly format)
analysis-framework.md: analysis framework summary (detailed technical document)
references.bib: reference list (citable)

Auxiliary output

metadata/: research metadata (reproducible)
extracted/papers_info.json: structured data of paper information (reusable)

Output management

Intermediate files for new tasks of this Skill are uniformly written to ./.bensz-api/task-{yyyymmdd-hhmm}-{简短描述}/{skill名}/input|output|log/. Reuse one task root directory for the same task; only create a new one when multiple Skills collaborate shared/Formal deliverables are not written to this directory; the hidden historical directory is only allowed for explicit compatibility reading, migration, or cleanup.

Validation

After completion, execute the Skill's existing static checks, script validation, or manual review, and record the pass criteria.

Failures and Recovery

Common Issues

Q: Unable to download certain PDFs A: Record the failed DOI; you can explain it in the final report, or try to obtain it manually

Q: Extracted method information is incomplete A: Use supplementary materials or contact the authors

Q: Too few references A: Loosen the search conditions (year, keyword scope)

Q: Methods from different papers conflict A: Mark the points of controversy in the analysis framework and provide multiple options

Constraints

Functional Boundaries

✅ What can be done：
- Survey the literature and summarize the methodology
- Develop analysis strategies and steps
- Recommend tools and best practices
- Generate an executable plan document
❌ Strictly prohibited：
- Perform actual data analysis
- Modify the user's original data
- Write formal scientific papers
- Make scientific conclusions

Usage Limitations

Only provide strategic advice; specific implementation requires user verification
Methodology recommendations are based on the literature, and cannot be guaranteed to apply to all scenarios
PDF availability depends on accessibility; some references may not be obtainable
AI extraction may contain errors; manual review is recommended for key information

Quality Assurance

All recommendations come from peer-reviewed papers
Clearly indicate the strength of evidence (consensus across multiple papers vs. a single paper)
Provide reference traceability
Mark uncertainties or points of controversy

Common Hard Constraints

When a task requires writing to disk, use the unique ./.bensz-api/task-{yyyymmdd-hhmm}-{简短描述}/ root directory; place shared materials in shared/, and place Skill-specific materials in that Skill's input/、output/、log/。
Formal deliverables, source code, and formal plans are saved according to project conventions and are not written to the task workspace; without authorization, do not overwrite, delete, migrate, or write remotely.
Project maintenance changes check BAC availability and record requirements, AI outputs, tool results, file changes, and validation summaries; BAC only performs process auditing and does not replace attribution, responsibility, or compliance judgments.
Do not record API Keys, access tokens, passwords, Cookies, environment/credential files, private Prompts, identity information, local usernames, hostnames, or unnecessary large volumes of raw data.
File paths must be normalized and restricted to the authorized project scope; external URLs, subprocesses, and network access follow least privilege to prevent path traversal, SSRF, and command injection.
The Skill version is uniquely recorded in its own config.yaml:skill_info.version; changes to public APIs, protocols, directories, or configurations are synchronized with documentation and CHANGELOG.md。
Only design flaws in the Skill or Bensz infrastructure itself are reported to bensz-collect-bugs; first write a sanitized version to ~/.bensz-skills/bugs/, the current task is not interrupted, public reporting only occurs when the user explicitly requests it, and directly modifying the user's installed Skill source code is prohibited.

Skill-Specific Constraints

Must not exceed the scope declared in this Skill description and the process above; do not disguise unverified information as definitive conclusions.

About this skill

Core Features

Core Value

Use Cases

Input Specifications

Required Input

Optional Input

Workflow

Input

Execution Steps

Phase 0: Initialization

Phase 1: Topic Extraction and Literature Investigation

Step 1.1: Topic Extraction

Step 1.2: Literature Search

Step 1.3: Literature Screening and Scoring

Step 1.4: Generate Reference List

Step 1.5: Download PDFs

Phase 2: Deep Learning Analysis Strategy

Step 2.1: PDF Information Extraction

Step 2.2: Comprehensive Analysis Framework

Phase 3: Develop a personalized plan

Step 3.1: Requirements alignment

Step 3.2: Plan generation

Step 3.3: Plan review

For users

For AI execution

Output

Core output

Auxiliary output

Output management

Validation

Failures and Recovery

Common Issues

Constraints

Functional Boundaries

Usage Limitations

Quality Assurance

Common Hard Constraints

Skill-Specific Constraints