JJdoesTech

How it started How it's going

Cursor just yoinking Kimi is so funny The wild west in AI is far from over

@ajeetunc I prompt injected my linked-in. But this is smort, gonna add that to any open repo now

an Open Source maintainer prompt injected his own project's Contributing.md to differentiate between human PRs and bot ones. and it worked. most bots started adding '🤖🤖🤖' in the title.

@LLMJunky Yep, amazingly so tbh. now my architect spawns workers to do stuff, those have reviewers, explorers what have you but those are constrained but no nesting loops this way

@JJdoesTech no, does it work?

Started doing something really fun with subagents in #codex: per-worker override of max_depth Simple example: I want my reviewers not to spawn more subagents But my workers should and can spawn sub-reviewers! So until a per-agent whitelist arrives, this is great

So tl;dr: config.toml agents.max_depth = 1 subagent.toml agents.max_depth = 2 This means the subagent (that's depth 1) can spawn one more (depth 2). But the sub-subagent can NOT spawn more.

GPT-5.4-Mini and Nano support are on codexmeter now Not 100% tested yet but looked good on first glance

@rohanvarma Windows CLI - no worktree support so mostly research/architecture agents etc - no good way to copy stuff with line breaks (/copy sucks since full message only) App - performance gets honestly just so much worse with multiple agents and, at least on windows, subagents meh

Some interesting data we pulled today showed that ~40% of Codex users use multiple surfaces, between the App, CLI, and IDE extensions. Everyone seems to have a primary preference, but a bigger-than-expected chunk of users launch codex agents outside of their primary interface. If you use multiple surfaces, I'm curious why? And what could we do to improve the experience?

@thsottiaux CLI - enable worktree feature a la app App - either: open source it or make updating more transparent/easy (Windows lacks behind Mac? idk? I don't see agent feature yet) Generic: - more transparency on spending/usage limit like what agents used what, etc

What are we consistently getting wrong with codex that you wish we would improve / fix?

@OpenAIDevs Is that also live on windows? Would be quite neat to have either OSS repo to build real quick or more clear update comms

Subagents are now available in Codex. You can accelerate your workflow by spinning up specialized agents to: • Keep your main context window clean • Tackle different parts of a task in parallel • Steer individual agents as work unfolds

@Dimillian Yay, the video export should work without chrome as well and only download the chromium blob if needed Everything else is packaged

@JJdoesTech This is working and extremely cooll!!!

Fixed a few more bugs Repos, Models and Daily overviews now respect date filters and massively sped up ingest speed as well Thanks @Dimillian and for the inspo on this one

@Dimillian Haven't done any mac/linux testing yet though, codex with 5.4 promises me "almost for sure" works on other platforms hahaha gonna add smoke CI now for other platforms to make sure it does

@Dimillian Absolutely, open source and npx package npx codexmeter@latest github.com/Pimpmuckl/code…

So what does that mean? Do: - Use Codex-5.3-Spark-medium for small-scale reviews If you have Spark tokens to burn: - Use Codex-5.3-Spark-xhigh for PR reviews but cover with a second reviewer

"PR"-Reviwe: The final gate pre-github PR is a dual pass until both reviewers come back green. Spark xhigh is usable here in theory but shouldn't be compared to 5.3 xhigh. Again, I suspect compaction-hell.

First batch of my #reviewbenchmarks is in with new scoring. Very cool data! Focused this one mostly on 5.3 Codex Spark. tl;dr: - Spark-med works for small reviews - more thinking is not better - Spark-xhigh is okay for PR reviews IF you have spare tokens A few notes: ⬇️

Did a lil thing for the #codex enjoyers out there Usage dashboard with cute animated overview and Model, Repo, Timeline and Session tabs Video Export as well to flex on socials Use: 🖥️ npx codexmeter@latest