Paste one prompt and answer a handful of real questions — about ten minutes. It writes the files that brief every future AI session on who you are, what you're working on, and how you like to work.
How: copy the prompt below, paste it into your agent, and answer honestly. It asks three questions per round, for about three rounds, then writes the files.
I want you to interview me so you can build the context files for our
future work together.
## Interview rules
- Ask 3 questions at a time (don't overwhelm me)
- Wait for my answers before continuing
- Ask deep questions, not surface-level ones:
- Bad: "What's your job?" (too vague)
- Good: "What's the single most important thing you do each week
that takes too long?"
- Probe follow-ups when I give short answers
- We have about 10 minutes. Aim for three rounds, then write the files.
## What to discover
About me: role, company, what I do day-to-day; current projects and
priorities; how I like to communicate; work rhythm; tools I live in;
what I hate doing; what I wish I had more time for.
About my workspace: tasks I repeat (daily/weekly); what I've tried with
AI and what worked or didn't; guardrails — what you should never do
without asking; off-limits systems; how I want errors handled.
## Output
When you're done, write:
1. USER.md — who I am. Portable; I can take this anywhere.
2. AGENTS.md — how you should work with me: tools, conventions,
guardrails.
Save these **globally**, not in this folder, so they apply everywhere:
~/.codex/AGENTS.md (and USER.md alongside it)
If you can't write files, print them and I'll save them myself.
Keep each under 100 lines. Be opinionated — it's easier for me to edit
than to write from scratch.
Start the interview now.
On Claude Code or Claude Desktop? Claude reads CLAUDE.md, not AGENTS.md — save only the file above and Claude never sees it, silently. Use this version instead: the same interview, one extra line of output.
I want you to interview me so you can build the context files for our
future work together.
## Interview rules
- Ask 3 questions at a time (don't overwhelm me)
- Wait for my answers before continuing
- Ask deep questions, not surface-level ones:
- Bad: "What's your job?" (too vague)
- Good: "What's the single most important thing you do each week
that takes too long?"
- Probe follow-ups when I give short answers
- We have about 10 minutes. Aim for three rounds, then write the files.
## What to discover
About me: role, company, what I do day-to-day; current projects and
priorities; how I like to communicate; work rhythm; tools I live in;
what I hate doing; what I wish I had more time for.
About my workspace: tasks I repeat (daily/weekly); what I've tried with
AI and what worked or didn't; guardrails — what you should never do
without asking; off-limits systems; how I want errors handled.
## Output
When you're done, write:
1. USER.md — who I am. Portable; I can take this anywhere.
2. AGENTS.md — how you should work with me: tools, conventions,
guardrails.
3. CLAUDE.md — containing exactly one line: @AGENTS.md
(Claude Code reads CLAUDE.md and doesn't read AGENTS.md. The
one-line import means both tools read the same instructions, with
only one file to keep up to date.)
Save these **globally**, not in this folder, so they apply everywhere:
~/.claude/AGENTS.md
~/.claude/CLAUDE.md (and USER.md alongside them)
If you can't write files, print them and I'll save them myself.
Keep each under 100 lines. Be opinionated — it's easier for me to edit
than to write from scratch.
Start the interview now.
@AGENTS.md; only needed on Claude, which reads CLAUDE.md and not AGENTS.md.