Am I using agentic coding wrong with Rozenite or is it...not quite there yet?

[spasmodiumspam-web]

The idea of having Rozenite Agent handle operations, debugging, and performance profiling - with full access to all the data - sounded amazing. I could already picture myself fixing bugs with just a few prompts, improving component performance, and so on.

So I went ahead and set everything up with all the plugins I needed. The documentation is well written, but I think it could really benefit from a step-by-step section that explains how to actually use Rozenite with Codex/Claude (app or CLI). For someone who’s newer to AI tools, that kind of guidance would be really helpful.

Everything seemed to be working fine. For example, I checked my MMKV storage and saw that app-theme = dark.

I thought, “Cool - let’s see if Rozenite can read this information,” just as a starting point, from the Codex app.

So I set up a prompt using ChatGPT 5.4 mini (it’s a simple task, after all), and after it spent some good 5 minutes on the task...

Whoa. But hey, might be a lucky guess. After all I only have light and dark theme, it was a 50/50 chance. Let's change theme to light and try again.

Welp...

Thought I gotta admit something, after starting a new thread and metro server fresh again, it started to get the correct values quite consistently.

My questions actually are:

1. Do we have to specify (at least in Codex) it should use Rozenite agent every single prompt? - meaning using the @, it's quite annoying

2. Is it supposed to take a lot of time? The reasoning chain always make me believe it tries a lot of things instead of simply calling the functions that i supposed are "hooking" it to DevTools - a lot of back and forth calling commands it shouldn't

I told it to get my current theme value and my current language and after 3 minutes, it did, but it did all this back and forth in its chain of thought - chain-of-thought-codex-rozenite.txt

3. Any tips or tricks (and could we get a section for that, please?)

I feel like the average dev will find all of this quite confusing.

[V3RON]

First of all, thank you for such a detailed report and for sharing the transcription of your conversation! It was extremely helpful in figuring out what’s going on and what I can do to improve the speed. Based on what I found, I added the necessary changes to the refactoring PR: #216. I'm already seeing a noticeable improvement locally in terms of performance.

[spasmodiumspam-web]

Fantastic news! Can't wait to test it.