Yaroslav

Building Foreman — run and manage Claude Code & Codex sessions side by side See what every agent's doing, before it goes off track -

[Yaroslav @yrozumnyi_dev]

This is what four Claude sessions look like when they're not four terminal tabs

[Yaroslav @yrozumnyi_dev]

@codewith55 Haha, I think it just tries to save some tokens for you 😅

[Yaroslav @yrozumnyi_dev]

@itsvishaltwt I don't know, maybe cuz I wrote about profile

[Yaroslav @yrozumnyi_dev]

Are you still sending screenshots to the session to explain what's wrong with UI? It was a pain for me for quite a long time, now things changed.

Agent in session could interact with browser page and do checks automatically. Here is a short demo of it, works for claude and codex

[Yaroslav @yrozumnyi_dev]

@Its_lakshya_ai "New generation is broken, they don't even read books anymore"

[Yaroslav @yrozumnyi_dev]

@Samaytwt Yeah it's too many sessions, impossible to handle. But I don't get your point. If developer runs 5-8 sessions in parallel it means he saves your business time by developing feature faster. Probably it worth paying him even more?

[Yaroslav @yrozumnyi_dev]

@sickdotdev This is why memory should be controlled by developer. Something like that, so I decide when and what to add into session context

[Yaroslav @yrozumnyi_dev]

@Yungkeem4987 @sl1ma4 Yeah, it's something really wrong with 5h limit on 100$, it's to small when weekly limit more theb enough

[Yaroslav @yrozumnyi_dev]

@feifei_qiu @sl1ma4 I have no idea how you manage this. When I was on 100$ plan Fable pushed 5h limit to 34% from one request. On 200$ it's much better

[Yaroslav @yrozumnyi_dev]

@talkaboutdesign @sl1ma4 It could help

[Yaroslav @yrozumnyi_dev]

@sl1ma4 The only way to really be able to use Fable is havin 200$ plan. Even 100$ is not enough, weekly limit is fine but not 5h. I was on 5x Max abd asked Fable to run a security review of the project, 5h limit went from 0 to 34% from one task

[Yaroslav @yrozumnyi_dev]

Built this over the last few months — Claude Code and Codex sessions on one canvas. Mac, Windows, Linux. https://t.co/dQez59rWuu

[Yaroslav @yrozumnyi_dev]

This is what four Claude sessions look like when they're not four terminal tabs

[Yaroslav @yrozumnyi_dev]

The thing I didn't expect from AI review: it's most useful in the language I'm worst at.

In TypeScript it tells me things I knew. In Python it tells me things I'd never have caught — because I don't have ten years of "that looks wrong" in Python.

[Yaroslav @yrozumnyi_dev]

Ran a code-health review on a repo. 30 test files.

Ran it again. Same commit, nothing touched. 50 test files.

Neither run was lying. One had counted tests/. The other had counted every tracked spec in the repo. The metric never said which population it meant, so each run quietly picked its own — and then confidently reported a number.

The bug wasn't in the counting. It was that "how many test files" isn't a question until you say which files count.

Fix: every metric now pins its own population as tightly as it pins its command. Whole tracked repo, filtered by name, same way every time.

A number moves when the code moves. That's the whole contract.

Check it out - https://t.co/S9Xe2eyLK2

[Yaroslav @yrozumnyi_dev]

@pcshipp I am using opus 5 for most tasks. Fable 5 I keep for global, big and difficult tasks. Something like full project review or refactoring most of it

[Yaroslav @yrozumnyi_dev]

@jiya_3063 @Taniyatweets_ Why?

[Yaroslav @yrozumnyi_dev]

@Taniyatweets_ Yeah, why not? Claude, if you running opus or fable it writes good code. Of course it can make mistakes same as developer can

[Yaroslav @yrozumnyi_dev]

@thepoonam0914 Building projects, you will learn a lot int the process

[Yaroslav @yrozumnyi_dev]

@xPraveen07 Mac