Skip to content

Repository files navigation

OpenSocialScience

An Open-Source Agent for Quantitative Social Science Research

Version Docs Examples License

Python Claude Code DeepSeek Harness MCP Word

English · 简体中文

Pages from the three example papers fanned out, labelled Causal, Descriptive and Measurement

OpenSocialScience is a toolkit for quantitative social science research that runs inside agents such as Claude Code and Codex. Given a research direction and a dataset, it carries out literature search, topic selection, econometric analysis, writing and cross-model review in turn, and exports a Word manuscript formatted for the target journal.

🔥 News

  • 🚀 2026-10-21 · v0.1 released. The first public release, covering causal, descriptive, measurement and simulation studies, with Word export in Chinese journal styles. (v0.1.0)
  • 🤖 2026-10-04 · Install through your agent. The agent follows INSTALL.md to set everything up, see Quick start.
  • 📄 2026-09-27 · Three example papers. One descriptive, one measurement and one causal study on CGSS data, each formatted for a Chinese journal, see Examples.
  • 📑 2026-09-26 · Chinese journals and literature. Style packs for Sociological Studies, Journal of Management World and Social Sciences in China set the layout, fonts and word limits, and Chinese papers can be imported from web full text.

🚀 Quick start

Install with your agent

Send this message to your agent.

Install https://github.com/openSocialScience/openSocialScience by following docs/INSTALL.md in the repo.

Install manually

Requires Python 3.11+ and pandoc.

git clone https://github.com/openSocialScience/openSocialScience
cd openSocialScience
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev,figures,stats-formats,excel,causal,reading]"

cp .env.example .env    # add your API keys
oss doctor
oss install-skills

Start a study

Open your agent in the project directory and enter a research direction, the data and the discipline.

/social-research Educational expansion and family background --- data: data.csv --- discipline: sociology

The paper is saved to runs/<run>/paper.docx. See the user guide for more options.

🤖 Supported agents

Agent --host
Claude Code claude-code
Codex / OpenCode codex
DeepSeek Harness dsh
WorkBuddy workbuddy
OpenClaw openclaw

Other clients can connect through oss mcp serve, see MCP.

📄 Examples

All three papers were generated by OpenSocialScience.

Paper Type Data Journal style
The Urban-Rural and Generational Distribution of the Differential Mode of Trust in China Descriptive CGSS 2015 Sociological Studies
Whose Gains Does the Public-Service "Sense of Gain" Measure? Measurement CGSS 2013 Journal of Management World
Religious Participation, Charitable Giving and Trust in Strangers Causal CGSS 2012 Social Sciences in China

🔍 Number provenance

Every statistic in the manuscript is written as a placeholder such as {{result:main.coef}} and filled in from results.json at render time. A hand-typed decimal fails the check. See Architecture for the checks at each stage.

paper.md           The effect is {{result:main.coef}} (SE {{result:main.se}}, p {{result:main.p}}).
paper_rendered.md  The effect is 0.188 (SE 0.026, p < 0.001).

📦 Built-in data

QoG, V-Dem, GSS and CGSS are ready to use, for example --- data: pack:qog. Indicators from the World Bank, OECD and Our World in Data can be searched and downloaded with oss data search and oss data fetch. Simulation data comes from /social-sim, see simulation.

📜 License

Apache 2.0

📝 Citation

@software{opensocialscience,
  author  = {{OpenSocialScience contributors}},
  title   = {OpenSocialScience: An Open-Source Agent for Quantitative Social Science Research},
  year    = {2026},
  version = {0.1.0},
  url     = {https://github.com/openSocialScience/openSocialScience}
}