How You Prompt — Codex

Shows you how you actually talk to Codex.

What you get

One page, built from your own history. Every kind of prompt you actually send gets a row — how often it comes up, a real fragment of what it looks like, and whether that kind tends to land first time or spiral into back-and-forth. Above the table sits the read that matters: a few honest sentences about what you reach for, what you never say, and where you're precise versus vague — the thing that's actually worth knowing about how you work.

Nothing is sent, changed, or acted on. It only reads.

How to look at it

Open the Your Style app in this cabinet. It comes pre-filled with a made-up example so you can see the shape of it straight away — the page says clearly that the numbers are fake. Your first real read takes its place.

Before it can read your real history

Import your Codex history from Cabinet's integrations screen. It's a one-off import, not a live connection, so run it again when you want a fresher read.

Set aside a minute for it once. Until then the page keeps showing the example.

What's inside

• Style Reader — the one agent. Reads the imported transcripts, writes the read.
• Monthly Style Read — the one routine. Runs at 10:00 on the 1st of every month.
• Your Style — the one page. Opens on the latest read, with a picker for earlier ones.