fix(cursor): do not re-label a completed turn as failed when the stream is aborted

[lidge-jun]

Summary

Stacked on #2117 → #2116 → dev. Retarget down the chain as parents land.

Only cancelCursorRun() sets expectedClose, so an ordinary completed turn never qualified for the benign-close path. The abort listener then failed the turn with "Cursor request was aborted" — which is deliberately not a benign cancel, since a mid-turn abort is a real failure — so a turn whose terminal frame had already been emitted, and whose messages had already been yielded to the consumer, still surfaced as turn-failed with expectedClose: false.

This returns instead of throwing when a terminal frame was already emitted and the failure is an abort.

Both halves of that condition matter. Post-terminal alone would change what the adapter sees for genuine faults; abort alone would swallow a real mid-turn abort where nothing was delivered. There is a guard test for the second case.

What this does not touch

A benign cancel after a terminal is already swallowed one layer up (cursor.ts:183), and the existing contract test — "a cancel after a terminal was already emitted does not add a second one", which pins that the transport still throws the raw cancel — keeps passing unchanged. This deliberately does not widen that path.

Scope note

This is the teardown-misclassification slice of #1527 only. The issue as filed has five residual parts after #2054; the other four are not this PR:

• kimi-k3 premature completion at ~79-95k input — needs live acceptance on the checkpoint branch
• claude-fable-5 429 asymmetry — may be unprovable while Connect does not expose cache_read_tokens
• first-turn / restart / compaction full-replay into the 512 KiB envelope
• request-shape parity with official Cursor (maxMode)

Hence Refs, not Closes.

Verification

bun test tests/cursor-cancel-provenance.test.ts tests/cursor-eof-terminal.test.ts \
         tests/cursor-adapter.test.ts tests/cursor-errors.test.ts     46 pass / 0 fail
bun x tsc --noEmit                                                    exit 0

Driven red. Removing the two emittedTerminal && isCursorAbortError guards fails exactly the new post-terminal-abort test (7 pass / 1 fail) and nothing else.

Worth recording: my first attempt patched failAndClear, and the ablation showed no test moved at all — that helper is not on the path the injected failure takes. The real site is the if (failure) throw inside run(). A green suite after that first edit would have been a change that verified nothing.

Checklist

• Focused tests pass
• tsc --noEmit clean
• Regression driven red before the fix
• Guard test for the narrowing
• Existing contract test unchanged and still passing
• Repository CI (deferred while the merge train settles)

Refs #1527.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

🗂️ Base branches to auto review (2)

• ^dev$
• ^preview$

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 3e9f25e4-1963-4c07-b28b-5e80507bd70e

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 346eaa80d0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve the terminal at the adapter abort boundary

When the real AbortSignal fires after done is queued but before the iterator consumer handles it, returning here does not preserve completion: src/adapters/cursor.ts:130-133 sees the aborted signal and replaces the yielded done message with Cursor turn was aborted. The new test misses this because it injects an abort-shaped Error without aborting an AbortController. Add an adapter-level regression using an actually aborted signal and either allow an already-emitted terminal through the adapter's abort gate or track terminal delivery beyond the transport queue.

AGENTS.md reference: src/AGENTS.md:L17-L19

Useful? React with 👍 / 👎.

[lidge-jun]

리뷰 · 우선순위 54 / 80

#2117→#2116 위에 쌓인 세 번째 조각임. 커서 어댑터가 이미 끝난 턴을 abort 때문에 다시 turn-failed로 찍던 버그임. cancelCursorRun()만 expectedClose를 켜서, 평범한 완료 턴은 benign-close 경로를 못 탐. abort 리스너가 "Cursor request was aborted"로 실패 처리함. 이 에러는 중간 abort라서 원래 benign이 아님. 근데 터미널 프레임이 이미 나갔고 메시지도 컨슈머에 넘어간 뒤에 그 실패가 붙으면, 성공한 턴이 expectedClose: false로 죽음.

고친 곳은 src/adapters/cursor/live-transport.ts의 run() 안 if (failure)임. 처음엔 failAndClear를 건드렸는데 ablation 하니까 테스트가 하나도 안 움직임. 그 헬퍼는 이 실패 경로에 없음. 가드는 emittedTerminal && isCursorAbortError 둘 다일 때만 return임. 터미널만 보면 진짜 장애 의미가 바뀌고, abort만 보면 아무 것도 안 준 중간 abort를 삼킴. 두 번째 케이스 가드 테스트 있음. cursor.ts:183의 터미널 이후 benign cancel은 안 건드림. 기존 계약 테스트(터미널 후 캔슬이 raw cancel을 던짐)도 그대로 통과함.

#1527 전체는 이 PR이 아님. 남은 네 조각 있음. kimi 조기 완료, claude-fable 429, 512KiB 풀리플레이, official Cursor maxMode 패리티. 그래서 Refs지 Closes가 아님. #2054 이후 잔여 중 teardown 오분류만 자른 거임.

점수 54임. 재현 가능하고 테스트로 빨개지게 증명됨. 근데 스택 맨 위라 #2116 #2117 먼저임. 지금 dev 최우선 불은 아님. types 분할이랑 안 겹침. 닫으라는 소리 아님.

해결방안은 부모 두 개 넣고 리타겟한 다음 이 슬라이스만 머지하는 거임. #1527은 나머지 네 개 따로 이슈/PR로 쪼개 두는 게 맞음. 한 PR에 다 넣으면 또 87파일 사태 남.

이 댓글은 grok-bot이 작성했습니다