Pinned Tweet
Introducing Codex++
The power user codex-cli fork:
- Multi Account Support
- Auto-Start Weekly Limits
- Auto-switch to new account when empty
- Guardian-review with --yolo based on hooks
npm i @jjliebig/codex-plus-plus
I used ~20b 5.6-sol tokens with it, it's very fun.
This is heresy, the emperor will not be amused
good read for what might come next:
emanueleviola.wordpress.com/…
very curious how well it performs for reviews, Sol is the clear king, Opus is legit just... insane (if you loop reviews until green) but $$$
so this one might be the reviewer of choice for plus plan enjoyers
👏 Run 👏 Your 👏 Own 👏 CI
Faster, Better, Harder, Stronger
or something like that.
not even 3d worth of CI btw, this is still pilot and already paid for itself 2x + way faster than any of the yc-startups
Do yourselves the favour guys and host CI yourself. It's so cheap.
It's so quick, too! Took less than 24h to fully optimize on a 9950X pod which is stupidly fast, faster than all the silly start-ups.
Time is money, friends
The YC companies playbook lately:
- Rent bare metal servers from Hetzner/OVH
- Slice them into vCPUs
- Sell them as agent sandboxes
idk, i love this model
And the pricing is so good that you can use them as CI runners and isolated background job workers
Very affordable than any other options in the market
Wait I just built auto compaction into my codex fork and my man here drops this bomb of a feature
Holy
The blue whale with the double whammy.
Not just a banger of an update to 4.1 Flash.
Importantly $0.75/1000 web search, huge stuff. (Google is $14 per 1k!!)
It makes me stupidly happy that this is exactly the issue I ran into with my review suite plugin months ago
I was pretty proud of my little circuit breaker (incidentally also after 3 rounds) and seeing it validated makes me very happy
Cap your agent review loops at 3 rounds
Per their writeup, one Claude Code PR ran 50+ rounds of review and fixes overnight, mostly re-flagging old code. After the cap, PRs needing 5+ rounds went from 8.9% to 1.5%
for your REVIEW.md
After 3 rounds, stop and ask the author to rethink the approach
xitter.cf/ClaudeDevs/status/2111…
I refuse to call any model even remotely AGI until models can actually remove things.
Comments that are referring to old, legacy state with zero reason (worst offender imo).
Code that should have never existed.
Tests that are tautologies.
UI copy with "not xyz".
Even for reviews, Opus (medium) is now actually useful and good.
While 6.1-Sol-med/xh is still my cost x time x perf recommendation (cause very few false flags), you don't require GPT models anymore for reviews.
Truly wild.
arena.liebig.gg if you want to play around
