I spent nine years as CTO of a direct-response company, and this is how I'd set Overvibe up for copy. Claude drafts the advertorial. Codex goes after the hook, the lead and every claim that could get the ad rejected. They trade drafts until you have versions worth testing, and you pick which one runs.

Paste the draft back and ask what's weak, and you get polite edits to the same idea. The hook it picked is still the hook. The claim it wrote is still there, because the writer thought it was fine. You end up as the second reader on every draft, pasting text between two chat windows.
Behind it, one command. You don't type this. Claude runs it when you ask for a Codex critic, because it learns the overvibe CLI from the skill that ships with Overvibe. Read the CLI reference
$ overvibe new --attach --tool codex --cwd ~/ads/larkmoor --name "ads: codex critic" \ --prompt "Read brief.md and advertorial.md. Attack the hook, the lead and every claim an ad reviewer would reject. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<notes>\"" started "ads: codex critic" in ~/ads/larkmoor, attached to this session # Claude ends its turn; Codex's notes arrive as a new message $ overvibe send "ads: codex critic" "v2 is in advertorial.md. Kept the 3 pm story lead, cut the pills line. Round two?"
Anything with a hook, a promise and a CTA gets better when someone hostile reads it first.
Claude writes the native story. Codex asks whether the headline reads like news or like an ad, and whether the bridge to the offer is earned.
Hook, problem, mechanism, proof, close. Codex marks the stretch where the story stalls and the line where the pitch shows too early.
Headline, subhead, bullets and button copy, argued against the ad that sends the traffic, so the click and the page make the same promise.
Ask Claude for twenty hooks. Ask Codex to kill fifteen of them and say why each one dies.
Two models pull a draft in two directions, so the variants differ in the idea. Changing three words is a weak test.
Health, money and before-and-after language, read for risk before the platform's reviewer or a client's lawyer reads it.
In supplements, health and finance, one wrong verb gets an ad rejected or an ad account restricted. Put your rules in the brief and Codex reads every draft against them. See the whole setup
Two models catch more than one, and they still miss things. Neither is a lawyer or an ad platform, so have someone who knows the rules sign off on regulated copy.
The files sit in a folder on your computer. There's no Overvibe server and no account, and each CLI talks only to its own provider.
You need the Claude Code and Codex CLIs signed in. Each one runs on your own plan and its limits. Overvibe itself is free.
Yes. agy, Google's command-line agent for Gemini models, takes the same overvibe new and overvibe send commands, so it can be a second critic or a tiebreaker.
It writes drafts and critiques. You bring the offer, the brief and the judgment about what runs.
In the folder you picked, as plain Markdown or text files. Open them in any editor or paste them into your page builder.
Claude and Codex both write in many languages. Ask for the draft in yours and tell the critic which market it is for.
Install it, open your brief in Claude Code, and let Codex take the first swing at your hook.
