Назад към всички

research-report

// Research technical projects/papers and generate comprehensive reports with PDF export. Modes: lite (analysis + writing) or full (+ environment setup + experiments). Params: iterations, mode, project-path, output. Use for paper analysis, code review, technical reports, research documentation.

$ git log --oneline --stat
stars:1,933
forks:367
updated:March 4, 2026
SKILL.mdreadonly
SKILL.md Frontmatter
nameresearch-report
descriptionResearch technical projects/papers and generate comprehensive reports with PDF export. Modes: lite (analysis + writing) or full (+ environment setup + experiments). Params: iterations, mode, project-path, output. Use for paper analysis, code review, technical reports, research documentation.
metadata[object Object]

Research Report Generator

Analyze technical projects/papers and produce comprehensive reports with PDF export.

Modes

Lite Mode (default)

  • Literature search + paper analysis
  • Code reading (local or remote)
  • Multi-iteration report writing
  • PDF generation via md2pdf skill
  • No environment setup or experiment runs

Full Mode

  • Everything in lite mode +
  • Conda/virtualenv setup
  • Dependency installation
  • Experiment execution
  • Result analysis

Usage

bash {baseDir}/scripts/research-report.sh \
  --topic "Spatial Forcing" \
  --mode lite \
  --iterations 3 \
  --output both

Parameters

ParameterDefaultDescription
--topic(required)Paper/project name or arXiv ID
--modelitelite or full
--iterations3Report revision iterations
--outputbothmd, pdf, or both
--project-path(auto)Local code directory (optional)
--workspace(current)Workspace directory

Workflow

Phase 1: Discovery

  1. Search arXiv + project pages
  2. Fetch related papers (citations + references)
  3. Identify key technologies and dependencies

Phase 2: Analysis

  1. Read source code (if --project-path provided)
  2. Analyze architecture from docs/code
  3. Map technology stack

Phase 3: Report Writing (× iterations)

  1. Draft outline
  2. Write sections iteratively
  3. Add diagrams (Mermaid/ASCII)
  4. Refine explanations

Phase 4: Export (Full mode only)

  1. Setup conda env
  2. Install dependencies
  3. Run experiments
  4. Append results to report

Phase 5: PDF Generation

  1. Call md2pdf skill
  2. Send to user via Telegram

Output Structure

<workspace>/
├── reports/
│   ├── <topic>_report_v1.md
│   ├── <topic>_report_v2.md
│   ├── <topic>_report_final.md
│   └── <topic>_report_final.pdf
├── memory/YYYY-MM-DD.md (appended)
└── logs/<topic>_research.log

Report Template

The generated report follows this structure:

  1. Executive Summary - 100-word overview
  2. Motivation - Problem statement + why it matters
  3. Background - Prerequisites explained intuitively
  4. Core Method - Technical details with analogies
  5. Code Analysis - Key files walkthrough
  6. Experiments - Setup + results (full mode)
  7. Troubleshooting - Common issues + fixes
  8. References - Papers + repos + docs

Dependencies

Required:

  • pandoc (for PDF export)
  • texlive-xetex (CJK + math support)

Full mode only:

  • conda/miniconda
  • CUDA toolkit (if GPU experiments)

Integration

This skill automatically:

  • Uses md2pdf skill for PDF conversion
  • Appends to memory/YYYY-MM-DD.md
  • Creates structured report directory

Examples

Lite mode, 5 iterations:

research-report --topic "VGGT" --iterations 5 --mode lite

Full mode with local code:

research-report --topic "Spatial Forcing" \
  --project-path ~/Spatial-Forcing/openvla-SF \
  --mode full \
  --iterations 3

PDF only output:

research-report --topic "OpenVLA" --output pdf

Troubleshooting

IssueFix
PDF generation failsCheck pandoc --version, install texlive-xetex
CJK characters missingInstall fonts-noto-cjk, verify with fc-list :lang=zh
Math formulas not renderedEnsure markdown uses $...$ / $$...$$ syntax
Full mode conda failsRun conda update -n base conda first