I have read the CLA Document and I hereby sign the CLA

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

@jif-oai would you mind taking a look when you have a chance?

This PR fixes the "interrupt mid-turn → later turns repeat aborted work" issue by recording a model-visible <turn_aborted> marker in history (and includes a regression test).

Appreciate any review — thank you!

I have read the CLA Document and I hereby sign the CLA

I have read the CLA Document and I hereby sign the CLA

@codex review

Codex Review: Didn't find any major issues. Keep them coming!

I want to explicitly call out that this fix was implemented by @swordfish444 (my brother), and I’ve reviewed the full diff + tests closely because this bug had a real, material impact on my day-to-day use of Codex.

In v0.80.0, interrupted turns not being persisted as model-visible state caused follow-up turns to resume or repeat aborted work, often re-triggering side effects. It was disruptive enough that I had to downgrade to v0.78.0 to stay productive.

What I really appreciate about this PR is that it addresses the problem at the correct abstraction layer:
• A durable, model-visible <turn_aborted> marker is recorded into history at interrupt time
• The marker is treated consistently as a session prefix (alongside environment_context)
• The marker is flushed before emitting TurnAborted, ensuring deterministic visibility
• Event ordering edge cases are handled explicitly rather than assumed
• Comprehensive regression tests assert that follow-up requests carry the marker and do not repeat aborted side effects

This isn’t a workaround or a heuristic — it’s a clean, well-scoped fix that aligns the event stream, session history, and model context in a way that’s robust under interruption, retries, and concurrency.

From a user perspective, this dramatically improves reliability for longer or interruption-prone Codex workflows. From a code perspective, the solution feels intentional, well-tested, and future-proof.

Strongly hope this lands soon — it’s a big quality-of-life improvement for anyone using Codex seriously.

Thank you OpenAi/Codex team for your wonderful product!

This contains a behavioural change of the model so we are discussing internally on the best approach to take here and will come back to you shortly

@codex review

Codex Review: Didn't find any major issues. Delightful!

@swordfish444, looks like this PR is close. There are a few CI failures that still need to be addressed. You can ignore the Bazel ones, but the other linter issues need to be fixed. Once that's done, I think this will be good to merge!

Thanks! I fixed the non-Bazel CI failures (missing text_elements in a couple UserInput::Text initializers in codex-rs/core/tests/suite/abort_tasks.rs).

Pushed a small follow-up commit; locally just fmt, just fix -p codex-core, and cargo test -p codex-core --all-features -- --test-threads=1 are all green.

Happy to rerun CI (non-Bazel) once approved.

@codex review

Quick update: merged latest main to resolve conflicts and re-ran local checks.

• cd codex-rs && just fmt
• cd codex-rs && just fix -p codex-core
• cd codex-rs && cargo test --all-features -- --test-threads=1

All green locally.

Also addressed the compaction concern from the Codex bot: local compaction now preserves <turn_aborted> markers (with a regression test in codex-rs/core/src/compact.rs).

@etraut-openai This is ready to go!

@etraut-openai Do you need me to keep updating the branch or should I let you all take it from here?

@swordfish444, thanks again for the contribution. Apologies for taking a while to approve it. This change touches some areas that have high regression risk, so we wanted to be extra cautious before merging this one.