Codex just dropped Hooks in the latest release - so I built a tiny macOS toolbar app, Loopndroll, that makes sure your Codex Agent never stops working. It’s open source. Link below.

@thekitze now you don’t need queue 10 messages till it finish 😂

love the name of this 😂 feels like playing Elden rings

I don’t understand the purpose?

Codex like many other agents wants to finish the job before it’s actually finished, even if you tell them “don’t stop until the tests pass” or “make sure you checked the feature in a real browser with playwright”. So I often queue multiple tasks in a row like “Keep working, you have not finished yet” because for anything less trivial agent basically can’t finish things in one go and response often with things it has not finished.

Okay 🤷‍♂️ I have never experienced that

Hooks plus a tiny macOS toolbar is exactly the kind of glue that turns Codex from demo energy into usable leverage.

@Dimillian 😮

Thank you very much, very useful

What are hooks

this is exactly the kind of thin glue layer that makes agent tooling actually usable. hooks are the primitive, but the UX of "never let it idle" is what people actually need. nice execution

Good for autoreseach

@MartinShkreli Is this possibly helpful regarding your recent post?

so it would just keep dumping wasting token

a toolbar app on top of that is a smart move. makes the invisible guardrails visible.

Keep it Ralph loop

Hooks as the connective tissue for persistent agents is underrated. What's your retry logic when the hook fires on a failed run?

Loopndroll with Hooks changes everything for agent automation workflows. Clean implementation

Are you aware that tokens don’t grow on trees?

Ahh cool stuf

Nice quick POC

is it jsut ralph loop?

And, um parallel,@AnthropicAI @claudeai reducing the limit of use…. #movingtocodex

does it auto retry on failures or just keep the agent alive between tasks?

@maciejewskii0 Just keeps that mf working. It just a quick POC, just to check how Hooks work and that it’s possible now force them to work without queuing multiple same messages.

RIP tokens. lol codex is actually pretty good and working until it actually gets the job done. I’ve never had to prompt it to just continue or anything.

Hooks are the right primitive here. Once you can intercept steps around the agent loop, a tiny guard app can keep runs alive without changing the core workflow. The part that matters long term is auditability, being able to see what hook fired, why it fired, and what it changed.