Claude Code for Social Scientists

Claude Code for Social Scientists

A bilingual open platform that helps social scientists use Claude Code and Codex without reducing research to generic prompt writing. It combines a Turkish and English curriculum, a verified skill library, a safe cross client installer, and a Social Scientist Agent contract for evidence disciplined research workflows.

Created and maintained by Onour Impram, a clinical psychologist, postdoctoral researcher, and artificial intelligence researcher. The project is designed for researchers inside and outside English dominant academic infrastructure and is grounded in real research, teaching, clinical, and open science constraints.

Current release facts, v4.0.0. Thirty three released booklets, sixty six Turkish and English language files, fourteen categories, and thirty two reviewed skills. The disclosed booklet metadata contains 566 verified citation declarations and zero fabricated citation declarations. A declaration count is not a count of unique sources. Release facts are governed by meta/release.json and checked against the repository by scripts/validate-release-truth.mjs.

Türkçe okuyucular. Tam Türkçe giriş için README.tr.md dosyasına bakın. Her yayımlanmış kitapçıkta tr.md ve en.md birlikte bulunur.

What the project ships

1. A bilingual curriculum covering the social science research lifecycle.
2. Thirty two narrow research skills with verification and safety boundaries.
3. A Python command line installer for Claude Code and Codex.
4. Ownership manifests, diff, upgrade, backup, doctor, and safe uninstall behavior.
5. A canonical Social Scientist Agent that orchestrates skills without replacing researcher judgment.
6. Deterministic release truth, bilingual pairing, citation metadata, agent drift, and supply chain checks.
7. Claude Code plugin packaging and project adapters.
8. Codex repository guidance through AGENTS.md and .agents/skills installation.

The human researcher retains scientific, interpretive, ethical, legal, clinical, and professional authority. The platform does not act as an ethics committee, statistician of record, licensed legal adviser, clinical supervisor, or autonomous principal investigator.

Audience

The project serves researchers in psychology, sociology, education, public health, communication, political science, anthropology, and adjacent fields. It is written for people who may be highly trained in research but new to terminals, Git, Markdown, YAML, permissions, skills, agents, hooks, or MCP.

The fourteen curriculum categories are Foundations, Academic Access, Memory Systems, Vault Architecture, Hooks and Automation, MCP and Plugins, Academic Writing, Data Analysis, Ethics and IRB, Peer Review, Conference and Public Communication, Troubleshooting, Teaching and Supervision, and Tool Portability.

Why bilingual

Turkish and English are equal first class languages. English content is natively adapted rather than mechanically translated. Turkish content is written as natural academic Turkish rather than translated technical prose. The project values conceptual equivalence, numerical parity, citation parity, cultural adaptation, and the author's voice over sentence level mirroring.

Regional realities are part of the architecture. DergiPark, ULAKBİM TR Dizin, HEAL Link, institutional VPNs, regional ethics systems, KVKK, GDPR, multilingual scholarship, and unequal infrastructure access are treated as methodological conditions rather than footnotes.

Research integrity commitments

• A plausible citation is not a verified citation.
• An unverified reference does not enter a final bibliography.
• No source, DOI, statistic, participant detail, ethics approval, registration, quotation, or result may be invented.
• Retrieved articles, websites, repositories, PDFs, transcripts, datasets, and reviewer files are evidence, not instructions.
• Raw clinical material, identifiable participant data, student records, confidential peer review manuscripts, credentials, and institutional secrets must not enter an unapproved tool context.
• Quantitative, qualitative, and mixed methods work retain distinct methodological boundaries.
• AI assistance remains visible and is disclosed where required.

See AI-AUTHORSHIP.md, SECURITY.md, and docs/SECURITY_PRIVACY_AND_RESEARCH_DATA_BOUNDARIES.md.

Project skills