Every session, you tell Claude who you are, and it runs on that self-report. It has never watched how you actually decide, delegate, and verify — the person in the transcripts, not the one in the intro. Out of Matrix reads your own Claude Code history and shows you that person.
An evidence-cited profile of the observed you: how you decide, what you delegate, how you verify, what drives and derails you. Every claim backed by ≥2 quotes from your real sessions. Anything unproven is labeled a hypothesis — never dressed up as insight.
The Mirror wired into Claude's memory. Every future session starts already knowing how you work — no re-briefing, no re-explaining who you are.
prompt-creator briefs any agent or session to your standards. project-manager runs your work the way you would, at the autonomy you choose. Both stay installed.
An illustrative excerpt — yours is built from your own sessions, every claim carrying its citations.
“You ask for options, then override them and ship your first instinct — the exploration is a warm-up, not the decision.” (2 sessions cited)
“You don't call anything done on someone's word — you make it prove it ran, then ask what's still broken.” (3 sessions cited)
“May work in long focused blocks late at night.” (thin evidence — labeled a hypothesis, not an insight)
That last one is the point: a mirror that flatters is broken. Anything it can't back with ≥2 quotes from different sessions is labeled a hypothesis, never dressed up.
That is your entire involvement — Claude does everything else and debriefs you at the end.
Your transcripts are read from your own disk and shown only to the same Claude that already saw them when you wrote them. No third-party service. No telemetry. Nothing published, nothing committed anywhere. Secrets are redacted before any model reads a byte. Everything lives in one folder you can delete in a minute.
Founders NFT utility: lifetime updates incl. v2 · a credit in CREDITS.md · an async review of your Mirror debrief.
Yes. Transcripts are read locally, shown only to the same Claude that wrote them, and never sent to a third party or published. Secrets are redacted before any model reads them. You can remove everything in about a minute.
No — it can't use it. It needs your local session transcripts, so it runs in Claude Code: terminal, desktop app, or IDE extension.
The Mirror scales to how much you've used Claude. A two-week corpus makes a thinner mirror than a six-month one — and it tells you so. A mirror that flatters is broken.
Yes. Everything lives in one folder. Delete it and it's gone in about a minute.
No. Out of Matrix runs in Claude Code but is not made, endorsed, or sponsored by Anthropic.