Subject: Let Claude interview you (10 min)
BRAINFILE #004 — Thu, Jul 30, 2026
Brainfile — Autopilot for Claude Code. Make Claude Code actually work for you. Free, Tue + Thu.
Preheader: Ten minutes of answering questions beats two hours of writing docs Claude half-reads.
DON'T WRITE CONTEXT FILES. MAKE CLAUDE INTERVIEW YOU.
The reason most people's AI setup has no context files: writing them feels like homework. So flip it — paste one prompt that makes Claude interview YOU, then write its own files from your answers. Ten minutes of talking beats two hours of doc-writing.
THE PATTERN
1. One question at a time
The prompt below forces sequential questions. Batch-questionnaires get skimmed answers; one-at-a-time gets real ones.
2. Answers become files immediately
After the interview, Claude writes MEMORY.md + reference files from your answers and shows you the result to correct. You edit, not author.
3. Re-run quarterly
Your stack, priorities and pet peeves drift. A 5-minute re-interview ("what changed since last time?") keeps the files honest.
COPY-PASTE THIS
----------------------------------------
Interview me to build my context files. Rules:
Ask ONE question at a time, 12 questions max.
Cover: what I do, my stack/tools, how I like replies (length/tone),
what you should never do, my current top-3 priorities,
and the mistakes AIs keep making with me.
Then write MEMORY.md (a one-line-per-fact index) plus one small
file per topic, and show me everything for corrections.
----------------------------------------
WHY THIS WORKS
Interviewing extracts the context you didn't know was worth writing down — the "never do X" answers are always the highest-value files and never appear in self-written docs. This is the exact onboarding flow we built into Brainfile itself.
(This is Brainfile's own onboarding pattern (brainfile.io/onboarding), shared in full.)
FROM THE WORKSHOP: brainfile-lite: the free, open core — now on GitHub
Claude Code is brilliant for a session and amnesiac between them. brainfile-lite is our free, open fix: plain-Markdown conventions (CLAUDE.md rules, a tiny MEMORY.md index, one small file per lesson, brain/ working state) that Claude reads at the start of every session — so decisions and lessons compound instead of resetting. No accounts, no lock-in, no telemetry. Clone it, run ./install.sh on your project, then tell Claude: “Read START-HERE.md and interview me to set up my brainfile” — ten minutes and it writes your files for you. The load-bearing idea: agents don't learn between sessions — files do. Star it, fork it, open issues; community improvements get folded back in fast. And when you want the files to maintain THEMSELVES — Autopilot runs, never-empty backlog, end-of-run briefs — that's Brainfile.
The lessons here are free, forever. Brainfile is the maintained layer —
everything it includes:
- The full Autopilot run protocol — tell it what you want and how long it has; it does the work and proves it
- All 10 pattern files (the structured plain-file playbooks the run pulls from)
- All 15 role packs (role-tuned setups; the free trial does not include these)
- The freshness feed — versioned updates every week as we learn
- 30-day money-back guarantee — cancel anytime
$99/mo · $999/yr · 30-day free trial, no card.
Start the free 30-day trial (no card): https://brainfile.io/trial
Vote on what we build next: https://brainfile.io/#vote-ladder
Brainfile · brainfile.io · Tue + Thu, always free · reply or [email protected]
Give your AI a working memory. See how Brainfile packages your context at brainfile.io.