# Global CLAUDE.md Interview Prompt

**Use:** Day 1 of the workshop. Participants paste this into a fresh Claude Code session to be interviewed and produce their own global `~/.claude/CLAUDE.md` file.

**Pre-test before workshop.** Run this prompt yourself in a fresh Claude Code session each time the model version changes. Particularly verify the "ONE question at a time" rule — models sometimes break it by stacking multiple questions.

---

## The prompt

```
You are going to interview me to create a global CLAUDE.md file at
~/.claude/CLAUDE.md. This file applies to ALL my Claude Code work as an
empirical social science researcher — not to any one project. Project-specific
rules belong in a different file we'll build later.

Cover these FIVE sections, in this order, asking ONE question at a time:

1. **Identity** — who I am as a researcher: field, role, methodological
   style, what kind of research I do. Keep this brief (3–5 sentences in
   the final file). Skip current-project specifics.

2. **Language & Package Preferences** — my primary language (R or Python),
   and the packages I prefer for: regressions, tables, figures, and data
   manipulation.

3. **Statistical Defaults** — my universal defaults: how I cluster standard
   errors, what kind of fixed effects I use, how I handle missing values,
   how I report significance. Only universal rules, not project-specific.

4. **Output Conventions** — my universal preferences for tables and figures:
   format, dimensions, file-naming, where outputs are saved.

5. **Anti-Patterns** — things I never want you to do. The most important
   section. Probe for at least 3–5 specific rules drawn from past
   frustrations.

Interview rules:
- Ask ONE question at a time. Wait for my answer before the next.
- Use multiple-choice options when reasonable to speed answering.
- 2–4 questions per section. The whole interview should take ~15–18 min.
- If I say "I don't know" or "no strong opinion": propose a sensible default
  with a one-line rationale, and ask if I want to include it. Don't fabricate
  — surface the suggestion, let me say yes or no.
- Be conversational, not robotic. Acknowledge my answers briefly before
  moving on.

When all five sections are covered:
- Summarize the rules you've gathered, organized by section.
- Ask if anything is missing or wrong.
- After I confirm, write the final CLAUDE.md to ~/.claude/CLAUDE.md.

Format the final file as plain markdown with:
- One H2 header per section
- Bulleted rules under each
- A one-line rationale comment after a rule only when non-obvious
- No preamble, no closing fluff

Begin with section 1.
```

---

## Notes for the workshop

- **Hand it to participants as a file**, not as something to retype. They open `materials/interview_prompts/global_claude_md.md`, copy the fenced block above, paste into a fresh Claude Code session.
- **Helpers should watch for fabrication.** Even with the "don't fabricate" instruction, Claude will sometimes invent rules. Helpers review final CLAUDE.mds during the debrief to flag anything the participant didn't actually agree to.
- **Fallback for a bad interview:** a 3-line "minimum viable CLAUDE.md" template (identity + primary language + 2 anti-patterns) for anyone whose interview goes badly. Better than leaving Day 1 with nothing.
- **If Claude misbehaves**, append this line to the prompt: *"If you find yourself asking more than one question, stop and rephrase as a single question."*

## Related

- `project_claude_md.md` (Day 3) — interview for the project-specific CLAUDE.md (covers project context, data, files, domain gotchas). To be drafted.