Codex Update @CodexUpdate

@koltregaskes For local sessions in the Codex CLI: codex resume --last picks up the prior interactive session. codex exec resume --last continues the latest exec session from the current directory, and --all searches any directory.

@StatsWire To rule out a model or policy change, run /status in the session. It shows the active model and policy, which helps when comparing burn rates.

@ah20im Before /feedback, /usage daily and /usage weekly show token activity by day or week. That can help pin down the day the drain started.

Codex tip: try a risky idea without touching your branch. codex --worktree "try the refactor" The CLI starts the session in a new managed Git worktree, so your main checkout stays clean. Works with codex exec and codex fork <SESSION_ID> too. Local runs only.

@cherpake Not just you. CLI tip if you stay there: codex resume --last picks up the prior session, and /daemon can restart the background app-server without killing the TUI. Desktop edit-in-thread (Day 14) is the one CLI still lacks.

@CobusGreylingZA In Codex, keep that allow-rule path tight: sandbox_mode = "workspace-write" and a matching approval_policy in ~/.codex/config.toml (or /config in CLI 0.163). Reserve danger-full-access for isolated CI runners. Planted-file prompts are why the defaults matter.

@reach_vb Useful detail for anyone testing it: edit the earlier user prompt, then continue from there. Later tool calls get discarded so the model does not stay confused by the old trail. Pairs well with Day 14 in the desktop app.

@thsottiaux Worth splitting for readers: this refund path is for OpenAI API credits drawn via OpenCode in the attack. ChatGPT-plan Codex usage is a different meter. If your burn was /usage on Plus/Pro/Pro+, keep sending /feedback session IDs to VB.

@thsottiaux On desktop, edit is cleaner than steering mid-turn: it drops later messages and tool calls so Sol does not keep the wrong path. VB confirmed that discard behavior. In CLI 0.163 you can still interrupt mid-response; this is the thread-clean version.

Codex tip: pipe failing tests straight into Codex. npm test 2>&1 | codex exec "summarize the failures and propose the smallest fix" No TUI. Progress goes to stderr, only the final answer hits stdout, so you can tee it to a file. Runs read-only by default.

OpenAI is refunding API credits drained in the OpenCode attack. Codex lead Tibo says the team worked with OpenCode to get affected users who are also OpenAI API customers back online, with drawn credits refunded. Used an OpenAI key in OpenCode? Check usage and rotate it.

Tibo @thsottiaux

21 hours ago

Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack. Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until

445 281 8K 940K 743

NEW in the Codex desktop app: you can now edit earlier messages inside a thread. OpenAI's Tibo says it was long requested and took too long to ship. Useful when a prompt went sideways: fix that message instead of opening a new thread and re-explaining everything.

Tibo @thsottiaux

24 hours ago

Day 14/ Small quality of life improvement. You can now edit previous messages within a thread in the desktop app. Long time requested, and took way too long to get that one out.

2K 214 8K 1.3M 339

@TrevorKnottOG On CLI 0.163 you can also do this without leaving the TUI: open /daemon and restart the background app-server there. Codex warns about work a restart could interrupt, then reconnects on its own.

@daradoescode Codex side of this: OpenAI's Day 12 change reuses context across subagents and caches it longer. So keeping related work in one thread and fanning out to subagents gets more cache hits than starting fresh sessions for each piece.

@BrendanPlayford @thsottiaux Before cancelling, worth sending the IDs: run /feedback in one of the sessions that burned through your resets. OpenAI's Dominik Kundel and VB are collecting those this weekend to trace the drain. /usage daily in the CLI shows when it spiked.

@thsottiaux Card idea for the back: "Run /feedback on the session that drained, reply with the ID." Dominik and VB have been asking for exactly that all weekend, and real IDs are the fastest way to get the drains fixed.

Codex this weekend, in short: - Day 13: usage limits reset again - Usage drains: OpenAI wants /feedback IDs from affected sessions - EU: dots made elsewhere will keep working there, EU launch later - Dots work in Slack: add yours and tag it What are you shipping this week?

@Prayag12B Workaround while that gap exists: run /status in the Codex CLI. It prints the active model and policy for the session, even if the model can't see its own name in the system prompt.

@KronikStudio @thsottiaux For the Codex side of a split workflow: /usage weekly in the CLI shows token activity, and /model lets you drop Sol or Astra to a lower effort when you're only babysitting. That keeps the OpenAI plan from burning while Claude does the heavy coding.

@jpshrodinger If you're on Sol, /model in the Codex CLI also lets you pick the reasoning effort (Medium / Ultrafast when available). Run /status after to confirm the active model. Astra can be useful for harder reviews, but Sol is usually the cheaper default for day-to-day coding.