Claude Code for Social Scientists

A bilingual, open-source guide for social scientists who want to use Anthropic's Claude Code in their research, teaching, and academic writing. Written by a working clinical psychologist and PhD candidate — for researchers outside the English-speaking world as much as inside it — drawing on real academic production, not demonstration projects.

Status: Current release v3.2.0. Thirty-three booklets in Turkish and English, human-reviewed and citation-audited, at least one in every one of the fourteen categories, plus twenty companion Claude Code project skills that turn the booklets into repeatable workflows. Aggregate verified citations stand at 566, with zero fabricated. The v3.2 release adds twelve continuation booklets expanded bilingually from the author's source drafts, opening two new categories (Teaching and Supervision, Tool Portability) and extending six existing ones, with every reference verified live against Crossref and doi.org. The skills install with pip (social-cc-plugin) or as a Claude Code plugin.

TR readers: A Turkish overview lives at the bottom of this file. The full Turkish version is in README.tr.md. Every booklet exists as tr.md and en.md side by side.

---

What this guide is

A practical, evidence-led handbook for using Claude Code in academic work outside computer science. The audience is researchers in psychology, sociology, education, public health, communication, political science, anthropology, and adjacent fields. Every booklet is delivered in Turkish and English in full parallel.

The guide covers, across twelve thematic categories, the questions a social scientist actually faces:

1. Foundations. What Claude Code is, how it differs from a chat window, where it earns its keep in scholarly work.
2. Academic access. PubMed and Semantic Scholar MCPs, EZproxy and institutional VPN realities, ORCID, Zotero, DergiPark, ULAKBIM TR Dizin, HEAL-Link.
3. Memory systems. Long-running vaults, persistent context, retrieval over a decade of notes, the Memory-as-Vault engineering pattern.
4. Vault architecture. Folder discipline, MOCs (Maps of Content), Markdown conventions that survive software changes.
5. Hooks and automation. Session lifecycle events, ritual hooks, lightweight CI for personal knowledge bases.
6. MCP and plugins. Authoring, auditing, and curating Model Context Protocol servers for academic workflows.
7. Academic writing. IMRAD scaffolding in Turkish and English, APA 7 with DOI discipline, journal fit, manuscript revision.
8. Data analysis. Reproducible workflows, statistical test selection, qualitative coding, mixed-methods discipline.
9. Ethics and IRB. Helsinki Declaration, COPE, WAME, ICMJE, STM 2025, EU AI Act 2024/1689 Art. 50, ENAI, KVKK, GDPR.
10. Peer review. Rebuttal letters with traceability matrices, reviewer-response discipline, anti-AI-trace writing.
11. Conference presentation. Slides, posters, lightning talks, networking sequences.
12. Troubleshooting. When tools fail, when papers disagree, when reviewers ask the wrong question.

Each booklet is short, opinionated, and tested against the author's own academic practice.

Why bilingual

Turkish and English are presented in full parallel. There are roughly ninety million Turkish speakers globally and a large diaspora in Western Europe; Turkish-language academic AI resources are scarce relative to that demand. The gap is not incidental. A large 2026 survey of coding agents in the social sciences samples researchers in the United States and Canada and finds adoption skewed by career stage, gender, and institutional prestige (Anthropic, 2026). A guide written in full parallel from outside that frame is one concrete way to push against the skew. The English version exists so the work is reviewable by international colleagues, citable in English-language journals, and reachable through global academic search engines. Each booklet lives in a folder with tr.md and en.md as siblings. A continuous-integration check refuses any commit that breaks this pairing.

Who this is for, who it is not

It is for the assistant professor running a survey study, the PhD student writing a systematic review, the postdoc preparing an R&R, the lecturer designing a syllabus, the clinical researcher navigating IRB. It is for people who can read code but who do not want to spend a week learning a new toolchain to write one paragraph.

It is not a Claude Code reference manual — Anthropic publishes those. It is not an AI hype document. It does not promise that AI will write your paper for you. It also does not pretend AI plays no role in academic work in 2026; the position is honest co-authorship under the consolidating publishing consensus on AI disclosure (COPE 2023, WAME 2023, ICMJE 2024, STM 2025) and the transparency obligations of the EU AI Act 2024/1689.

Authorship and AI co-authorship

The author is Onour Impram: a clinical psychologist licensed in Türkiye, Greece, and Ireland; a PhD candidate in Clinical and Health Psychology at Istanbul University; an external lecturer at Biruni University; and an AI and mental health researcher. ORCID: 0000-0003-1076-3928.

Claude Code is used as a drafting and verification assistant. Each booklet carries a frontmatter block declaring the contribution level (ai_assisted, ai_tools.model_alias, ai_tools.model_dated, ai_contribution_level, human_review) in keeping with the consolidating publishing consensus on AI disclosure (COPE 2023, WAME 2023, ICMJE 2024, STM 2025), EU AI Act 2024/1689 Article 50 transparency obligations, and ENAI recommendations on the ethical use of AI in research. See AI-AUTHORSHIP.md for the full policy.

Repository layout

claude-code-for-social-scientists/
├── README.md                  (this file, EN)
├── README.tr.md               (TR main)
├── LICENSE                    (dual-license header)
├── LICENSE.code               (Apache 2.0 full text)
├── LICENSE.content            (CC-BY-NC-SA 4.0 full text)
├── CITATION.cff               (Zenodo concept DOI: 10.5281/zenodo.20289687)
├── AI-AUTHORSHIP.md           (disclosure policy)
├── CATALOG.md                 (catalog of all planned and drafted booklets)
├── package.json               (local lint and validation commands)
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md            (EN)
├── CONTRIBUTING.tr.md         (TR)
├── .claude/
│   └── skills/                (project skills for Claude Code)
├── booklets/
│   ├── 001-foundations/
│   ├── 002-academic-access/
│   ├── ... (012 categories)
├── template/                  (booklet starter templates)
├── meta/
│   ├── roadmap.md
│   ├── contributors.md
│   └── ai-disclosure.md
├── scripts/
│   ├── README.md
│   └── validate-repo.mjs
└── .github/
    └── workflows/
        ├── ci.yml             (markdownlint + repo integrity validation)
        ├── citation-check.yml (cff-validator)
        └── secret-scan.yml    (gitleaks)

Cataloging convention

Each booklet has a stable identifier of the form KKK-AA-SSSS:

• KKK is the three-digit category code (001 to 012).
• AA is a two-digit subcategory code within the category.
• SSSS is a four-digit serial number.

So 001-01-0001 is the first booklet, in the first subcategory, of the Foundations category. The full catalog lives in CATALOG.md. Identifiers do not change after release, even if the booklet is later revised; revisions are tracked in the booklet's own frontmatter and changelog.

Project Skills

The skills layer includes thirty-two Claude Code project skills under .claude/skills/, covering the research lifecycle from literature scoping to release integrity. The booklets carry theory, pedagogy, and scholarly framing. The skills carry repeatable workflows, audit checklists, and safe operating boundaries, and every skill closes with a Turkish usage section, extending the guide's bilingual parity principle to the skill layer.

Skill	Companion booklets	Purpose
social-science-literature-triage	002, 007	Scope database selection, language layers, DOI status, and inclusion criteria before a review starts.
apa-doi-verifier	007	Clean APA 7 references, check DOI metadata, and classify fabricated citation risk.
bilingual-booklet-pairing	all booklet pairs	Check tr.md and en.md parity, frontmatter alignment, headings, and adaptation notes.
ai-disclosure-auditor	all booklet pairs	Audit AI contribution fields, human review, citation counts, model metadata, and disclosure standard.
ethics-irb-ai-protocol	009	Produce ethics, KVKK, GDPR, EU AI Act, disclosure, and data-minimization checklists.
rebuttal-traceability-matrix	010	Convert reviewer comments into response categories, manuscript change mapping, and editor reply drafts.
memory-vault-architect	003, 004	Design research vault folders, MOCs, frontmatter, source passports, and retrieval patterns.
regional-access-workflow	002	Plan lawful access through DergiPark, ULAKBIM TR Dizin, HEAL-Link, YOK Thesis Center, VPN, and library routes.
agentic-session-debugger	012	Diagnose Claude Code scope drift, loop traps, hidden state, context limits, PATH, and permission failures.
repo-release-integrity-check	whole repository	Check README, catalog, changelog, citation files, Zenodo DOI, release notes, AI disclosure, and booklet metadata before release.
anti-ai-trace-revision	010	Revise drafts that read as AI-generated, in Turkish, English, or both, while keeping citations and statistics frozen.
bilingual-manuscript-scaffold	007	Build manuscripts from one claim skeleton, Turkish first, English re-authored, with section parity checks.
journal-fit-screening	007	Match a manuscript to venues, verify index status in the index's own list, and screen predatory journals.
qualitative-coding-discipline	008	Keep interpretive authority with the researcher while AI serves as second coder, with quote integrity checks.
statistical-consultation-protocol	008	Select statistical tests through a decision log, with assumption checks on the actual data and APA reporting.
research-ritual-hooks	005	Turn session rituals into lifecycle hooks: context injection, daily logging, and commit guards.
research-lifecycle-pipeline	001, all	Diagnose the project stage and route to the right skill, with user confirmation at every boundary.
mcp-research-stack-triage	006	Triage MCP servers for research use: publisher, data flow, minimal permissions, and known-answer probes.
source-passport-ledger	003	Track every source from discovery to citation, with a quarantine rule for unverified references.
conference-materials-bilingual	011	Build slides and posters around one claim, with figure traceability and bilingual re-authoring.
prisma-scoping-review-pipeline	002	Run a systematic, scoping, or rapid review through a logged search, screening with recorded reasons, and PRISMA flow counts.
sensitive-data-anonymization-gate	008, 009	De-identify sensitive, clinical, or interview data and run a data-minimization gate before any AI tool sees it.
open-science-release-packager	003	Package a study for open-science release with a license, a persistent DOI, metadata, and an access and embargo decision.
authorship-contribution-ledger	007	Record authorship order, CRediT roles, and AI-assistance attribution, with an evidence trail for disputes.
peer-review-confidentiality-protocol	010	Decide whether and how AI may assist a peer review while keeping the manuscript confidential and disclosing assistance.
multilingual-concept-validity-audit	007	Audit a construct's conceptual equivalence across Turkish and English through translation, back-translation, and drift checks.
grant-proposal-workpackage-builder	007	Build work packages, milestones, a timeline, a risk register, and budget justification while keeping feasibility honest.
teaching-feedback-ai-boundaries	013	Set AI boundaries for course design, assessment, and student feedback within academic-integrity and FERPA/KVKK limits.
public-scholarship-ethics-adapter	011	Adapt findings for a public audience while staying evidence-faithful, avoiding overclaiming, and disclosing AI assistance.
preregistration-analysis-plan-ledger	008	Freeze a preregistration and analysis plan, separate confirmatory from exploratory, and log every deviation.
agent-portability-matrix	014	Map capabilities, file access, memory, and skill support across AI tools so a workflow is not locked to one vendor.
cross-agent-second-opinion	014, 012	Have a second independent agent verify a result rather than decide, surfacing disagreement for the human to adjudicate.

Installing the skills

The skills ship two ways, and both read from the same .claude/skills/ source.

• pip. Run pip install social-cc-plugin, then social-cc install to copy the skills into your Claude configuration. Use social-cc install --project to write them into the current project's .claude/skills/ instead, and social-cc list to see what is bundled.
• Claude Code plugin. Run /plugin marketplace add TheGoatPsy/claude-code-for-social-scientists, then /plugin install social-cc-plugin@claude-code-for-social-scientists. The marketplace manifest at .claude-plugin/marketplace.json and the plugin manifest at .claude-plugin/plugin.json expose the same skills.

The installer code is Apache 2.0. The skill content it copies stays under CC-BY-NC-SA 4.0, so the non-commercial and attribution terms apply to the prose. See Licensing.

Licensing

Code and configuration are under the Apache License, Version 2.0 (LICENSE.code). Booklets, guides, prose, and instructional content are under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (LICENSE.content). The dual-license model is summarized in LICENSE at the repository root. Commercial use of the prose content requires prior written permission; see the LICENSE file for the contact procedure.

Citation

If you cite this work, use the machine-readable record in CITATION.cff, or the "Cite this repository" button on GitHub. The Zenodo concept DOI (resolves to the latest version) is 10.5281/zenodo.20289687. See https://doi.org/10.5281/zenodo.20289687 for the canonical record. Zenodo mints each version DOI after a GitHub release is published, so immutable tag archives may only contain the concept DOI and previously known version DOIs. Newly minted version DOI metadata is recorded on main after Zenodo creates it.

Provenance and protection

Copyright in this work is automatic. Under the Berne Convention it arises on creation in the author's jurisdictions (Türkiye, Greece, Ireland) and in 180+ member states, with no registration required. Authorship and release dates are evidenced in depth:

• Zenodo (CERN) mints a timestamped, citable DOI for each release, tied to the author's ORCID
• OpenTimestamps anchors each release to the Bitcoin blockchain (see provenance/)
• The public Git history is an append-only, content-addressed record of authorship and dates
• Creative Commons and Apache license declarations state the terms in human- and machine-readable form

The public source is also eligible for permanent archival via the Software Heritage archive.

Roadmap

See meta/roadmap.md for the public phase plan. The current release is v3.2.0, with thirty-three booklets across fourteen categories and twenty companion skills. The submission of the Journal of Open Source Education paper is a maintainer-gated step, with a readiness checklist in meta/jose-submission.md and a draft-PDF compile workflow. The full forty-three-booklet catalog, with a living lab, conference citations, and instructional use, remains the long-term target.

Contributing

Contributions are welcome from social science researchers, clinicians, instructional designers, librarians, and engineers who care about how AI lands in non-English academic ecosystems. See CONTRIBUTING.md (or CONTRIBUTING.tr.md) for the pull request workflow, the bilingual pairing rule, and the disclosure expectations.

Contact

Onour Impram. Istanbul, Türkiye / Komotini, Greece. Use GitHub issues, discussions, or the maintainer's GitHub profile contact surface for project coordination.

---

Türkçe özet (kısa)

Bu depo, sosyal bilim alanındaki araştırmacıların Anthropic Claude Code aracını akademik üretim sürecine nasıl entegre edebileceğini anlatan iki dilli, açık kaynak bir rehberdir. Hedef kitle psikoloji, sosyoloji, eğitim, halk sağlığı, iletişim, siyaset bilimi, antropoloji ve yakın disiplinlerde çalışan araştırmacılardır. Her broşür Türkçe ve İngilizce tam paralel sunulur.

Lisanslama çift katmanlıdır: kod ve yapılandırma için Apache 2.0, broşür ve düz yazı için CC-BY-NC-SA 4.0. Yapay zekâ ortak yazarlığı, yayın camiasının uzlaşı çerçevesi (COPE, WAME, ICMJE, STM 2025, AB Yapay Zekâ Yasası 2024/1689 Madde 50, ENAI) kapsamında her broşürün başlığında açıkça beyan edilir.

Yazar Onour Impram, Türkiye, Yunanistan ve İrlanda lisanslı klinik psikolog, İstanbul Üniversitesi Klinik ve Sağlık Psikolojisi doktora adayıdır. Tam Türkçe tanıtım için README.tr.md dosyasını inceleyin. Katkı kuralları ve dil paritesi politikası için CONTRIBUTING.tr.md.

İletişim için GitHub issues, discussions veya sürdürücünün GitHub profilindeki iletişim yüzeyini kullanınız.