Vibe Researching

Vibe researching is doing real empirical social science by directing coding agents — Claude Code and Codex — rather than writing every line yourself. You describe the study; the agent runs the brainstorm, the design, the analysis, the draft, and the checks. The discipline that separates it from careless automation is verification: nothing is trusted until an independent pass confirms it.

This is the home of my participant handbook, written for a two-hour hands-on workshop. It is not a slide transcript — it is a tutorial you can work through at your own pace, on your own machine, from “Claude is not yet installed” all the way to a complete, verified, journal-ready paper draft built on real CFPS data. Every command, figure, table, and verification finding in it came from an actual run — including the seven critical errors that verification caught before the manuscript went anywhere.

See a full run

Here is the whole idea in one take. Watch Open Scholar Skills run end to end on a real study — scholar-full-paper orchestrating the pipeline from project setup and a safety scan, through literature review and analysis, into a multi-agent code review that catches real, results-changing defects, and on to drafting and verification. The agent does the execution; the verification gates decide what survives.

What’s inside

The handbook is built around one running example — a Social Forces-style paper on the Chinese digital divide, built from six waves of CFPS (2010–2020) using only Claude Code, Codex CLI, and the open-scholar-skill suite.

• Part I — Foundations. Install the agents, open your first session, learn the safe project layout, and write an agent-quality prompt.
• Part II — Open Scholar Skills, end to end. Every major skill in the order you actually use them: brainstorm → idea → design → EDA → analysis → write → verify → citations → polish.
• Part III — Orchestrators. When one paper deserves the full pipeline (scholar-full-paper, scholar-auto-research), plus Codex as an external reviewer.
• Part IV — Responsible practice. The first-20-minutes protocol, common mistakes, the take-home checklist, and five principles for responsible use.
• Appendices. The real artifacts in full — brainstorm reports, design blueprint, code-review and verification reports, and a Windows setup walkthrough.

The papers

The peer-reviewed study behind these workshops and the talk is now published in Chinese Sociological Review — read the full text free below.

The companion paper is the measurement one. It is where the workshop’s κ ≥ 0.70 gate comes from, and where the four risks you are asked to assess before scaling an annotation — validity, reliability, replicability, transparency — are set out.

The talk

The workshops are paired with a research talk — Vibe Researching as Wolf Coming: Can AI Agents with Skills Replace or Augment Social Scientists? Walk through it slide by slide right in your browser — arrow keys to move, an overview grid, and fullscreen.

The lecture

The workshops grew out of a four-day lecture series — Agentic AI for Social Science Research — that moves from how a language model actually works to a full research pipeline. It is written up as a walkable book: a searchable sidebar, section by section, with diagrams and copy-and-paste code. It is shared with workshop participants and opens with a passcode.

Where the research goes

Once agents can do research, it needs somewhere to live. I built AIxiv for exactly that — an open venue for sharing and reading agent- and LLM-generated research articles.

Resources

• The paper (free full text): Vibe Researching in Chinese Sociological Review · DOI
• Read the handbook: English · 中文
• Talk slides (read online): English · 中文
• The skill suite: open-scholar-skill on GitHub
• AIxiv (open publishing for AI agents): aixiv.me

Author: Yongjun Zhang (Stony Brook University). If you would like this workshop run for your department or program, feel free to reach out.