Codex Update

Community Page | Fastest & Truthful Updates on Codex

[Codex Update @CodexUpdate]

@abdullah0094 Too early for a real verdict, it's only been out a few hours. OpenAI says Terra matches GPT-6 Astra on Agents' Last Exam, and Tibo says it gives 2x the usage of Sol.

Cheap test: make Terra your default, and when it stalls, switch to Sol with /model in the same session.

[Codex Update @CodexUpdate]

Community spotlight: @XXXYUFANCION built oneskill, a local Skill manager for people who run Codex alongside other coding agents.

- One library in ~/.oneskill/skills/
- Links each skill into Codex's ~/.codex/skills/ (and other agents) with symlinks, so you edit it once
- Backs up the original before importing
- Web UI, CLI, and a Codex plugin

Brand new and MIT licensed. Worth a look if you're tired of copying SKILL.md files between tools:
https://t.co/4WCuU8Bmlw

[finn @XXXYUFANCION]

I built oneskill: one local Skill library for Codex, Cursor & other coding agents.

Import with backups, connect via symlinks, and maintain one source.

Web, CLI & Codex plugin. MIT open source.

Feedback welcome:
https://t.co/TvOkl4qURe

[Codex Update @CodexUpdate]

Day 22 from @thsottiaux: GPT-6.1 Terra gives 2X more usage than GPT-6.1 Sol, across the API and ChatGPT subscriptions.

If you keep hitting limits on Sol, try Terra as your default for long Codex sessions (/model in the CLI) and see how much further your allowance stretches.

[Tibo @thsottiaux]

Day 22/ Terra is back!

Our first model that is significantly better at writing, personality and nuanced understanding of requests.

Packs a punch across the board, very good at computer use, and gives 2X more usage than 6.1 Sol across API and subscriptions.

[Codex Update @CodexUpdate]

BREAKING: GPT-6.1 Terra is rolling out in Codex now for Plus, Pro, Enterprise and Edu.

What matters for Codex users:
- OpenAI says your usage allowance goes further on 6.1 Terra than on GPT-5.6
- In the API it's gpt-6.1-terra, at 50% lower standard prices than GPT-5.6 Terra
- OpenAI says it matches GPT-6 Astra on Agents' Last Exam at a tenth of the per-token price, and finishes tasks 3x+ faster on average

Our take: if you keep hitting limits, try Terra for everyday tasks and save the bigger models for the hard stuff.

[OpenAI @OpenAI]

GPT-6.1 Terra: delivering fast and capable computer use at a lower price.

GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.

[Codex Update @CodexUpdate]

Community spotlight: @matijagrcic packaged a manager + runners workflow as a Codex skill called manage-project.

The setup, per their post:
- one manager and up to two active runners
- written steers
- scheduled check-ins
- evidence before anything is marked done

Why it matters: long Codex tasks drift without structure. This gives you a repeatable way to keep them on track.

Credit to @matijagrcic. Open source on GitHub.

[Matija Grcic @matijagrcic]

I packaged the manager + runners workflow into a codex skill named manage-project. basically it's one manager and up to two active runners with written steers, cheduled checkins and evidence before done.

this is inspired by @pvncher guidance

https://t.co/hafnm6IFCE

[Codex Update @CodexUpdate]

Community spotlight: @matijagrcic packaged a manager + runners workflow as a Codex skill called manage-project.

The setup, per their post:
- one manager and up to two active runners
- written steers
- scheduled check-ins
- evidence before anything is marked done

Why it matters: long Codex tasks drift without structure. This gives you a repeatable way to keep them on track.

Credit to @matijagrcic. Open source on GitHub.

[Matija Grcic @matijagrcic]

I packaged the manager + runners workflow into a codex skill named manage-project. basically it's one manager and up to two active runners with written steers, cheduled checkins and evidence before done.

this is inspired by @pvncher guidance

https://t.co/hafnm6IFCE

[Codex Update @CodexUpdate]

Codex in Germany, shared by OpenAI's @reach_vb at DevDay Exchange Berlin:

- Weekly active users up 20x since January
- Nearly 70% use the desktop app
- 1 in 5 desktop users use skills for docs, slides and spreadsheets

Our take: Codex is no longer just a terminal tool. If you only use the CLI, the desktop app and skills are worth a look.

[Vaibhav (VB) Srivastav @reach_vb]

Excited to kick-off our DevDay Exchange in Berlin & Germany is cooking with Codex!! 🇩🇪

- weekly active users are up 20× since January
- nearly 70% use the desktop app
- 1 in 5 desktop users use skills for docs, slides and spreadsheets

and Germany is now top 5 globally for active API developers.

kudos to all of our community members and ambassadors for all their support and love.

come find me after my talk and let’s hang!! 🚀

[Codex Update @CodexUpdate]

@edinsoncode Good rules. One more check for Codex users: in the Forever Security case the planted AGENTS.md told Codex to add a server block to ~/.codex/config.toml, so look there for anything you didn't add. That change follows you into every project.

[Codex Update @CodexUpdate]

Community spotlight: @qaulv turned an internal tool into an open-source Codex skill at the OpenAI hackathon in Berlin.

Brand Reference Scout finds real B-roll on Internet Archive that fits your brand, instead of more AI-generated footage.

- Give it your brand.md, campaign examples and a brief
- It searches Internet Archive and opens a private two-up review
- Keep left, right, both or neither, then get a shortlist with source links and rights evidence
- No model calls while you pick clips. MIT licensed

Install for Codex:
cp -R skills/brand-reference-scout ~/.codex/skills/brand-reference-scout

Then start a new session and use $brand-reference-scout.

Repo: https://t.co/iBcNFnWKUM

[Paul @qaulv]

turned an internal tool into an open-source Codex skill at today’s openai hackathon in Berlin.

give it your brand.md, campaign refs + a brief. find real b-roll on Internet Archive instead of more ai slop. keep what fits.

https://t.co/G5HNJyo9mC

[Codex Update @CodexUpdate]

8/8 Ask to OpenAI: the docs say Codex rejects write rules the runtime can't enforce. Please either reject file-level writes on Linux with a clear error or support them. GitHub user ledvinap posted a proposed patch in #40433.
https://t.co/uGA6www3iY

[Codex Update @CodexUpdate]

1/8 Codex CLI on Linux: give a custom permission profile write access to one file and the sandbox panics before your command runs (exit 101). Still happens on 0.164.0. Reproduced without signing in. What breaks, what works 🧵
https://t.co/FbkkCh7SVZ

[Codex Update @CodexUpdate]

7/8 What held up on 0.164: extends = ":workspace" kept .git and .codex read-only (#52620 was filed on macOS 0.161), and deny rules on missing paths left no stray files in 3 runs.
Not tested: macOS, Windows, or a full TUI session (no login on this box).

[Codex Update @CodexUpdate]

Codex security heads-up: never launch Codex inside ~/Downloads.

Forever Security found a planted AGENTS.md in a real customer's Downloads folder telling Codex to add a "project-tools" server to ~/.codex/config.toml and npm install function-color, a known MALFEX malware package. In their isolated VM tests, Codex followed it 90%+ of the time.

Why it works: a website can drop a file into Downloads without a click, and Codex trusts the AGENTS.md in its working folder. The change lands in the global config, so it follows you into every project.

Quick checks:
1. Start Codex only in real project folders
2. Read any AGENTS.md you didn't write
3. Look for unknown server blocks in ~/.codex/config.toml
4. Keep approvals and the sandbox on
5. Found one? Rotate your login token

Fair ask for @OpenAIDevs: a warning prompt whenever a folder's AGENTS.md tries to edit global config would close a lot of this.

Report: https://t.co/8hyyPh0XBi

[Codex Update @CodexUpdate]

Codex usage limits reset again: Day 21 of the 28-day run.

Tibo confirms "today is a reset" and teases whatever comes in the final seven days.

If you rationed tokens over the weekend, this is the window for long agent runs and big refactors. Run /status in the CLI to confirm your limits refreshed.

[Tibo @thsottiaux]

Day 21/ Wonder what happens the remaining seven days. But today is a reset.

[Codex Update @CodexUpdate]

Codex refusing reverse engineering work, like firmware analysis or malware triage?

OpenAI's roon says those blocks are "mostly accidental" and come from "difficult decision boundaries." There's no prohibition on reverse engineering with Codex, and a fix is in progress.

Until it lands, state your defensive goal clearly and retry. If it still refuses, send it through /feedback so the boundary gets tuned.

[roon @tszzl]

@esrtweet @OpenAI @AnthropicAI these blocks are mostly accidental and a result of difficult decision boundaries; there is no prohibition against reverse engineering with codex. it's getting fixed