fix: remove abort listener leak in abortableSleep

[hobostay]

Summary

• The onAbort listener in abortableSleep was never removed when the sleep timer resolved normally
• Over repeated retry attempts, leaked listeners accumulated on the AbortSignal, causing a slow memory leak
• Fixed by removing the listener in the timer callback, matching the cleanup pattern already used by fetchWithTimeout in the same file

Test plan

• Verify telemetry transport tests pass
• Confirm no listener leak after multiple normal sleep completions

🤖 Generated with Claude Code

[liruifengv]

Thank you for your interest in contributing to kimi-code.
However, we currently do not accept bulk AI-generated submissions that have not undergone thorough human review, so we will be closing these PR.

[trustmiao]

Planning note for issue/PR #4 (investigation only, no code changes made in this phase).

Requirement confirmed:

• The reported problem is in packages/telemetry/src/transport.ts: abortableSleep() registers an abort listener, but when the timeout resolves normally the listener is not removed.
• In the retry path, AsyncTransport.send() calls sleepImpl(backoff, signal) after transient telemetry failures, so repeated successful backoff sleeps with the same signal can accumulate stale listeners.
• fix: remove abort listener leak in abortableSleep #4 is currently a closed PR rather than an open issue, but the underlying bug report is actionable.

Affected scope:

• Primary code: packages/telemetry/src/transport.ts, specifically abortableSleep() around the retry backoff sleep helper.
• Related reference pattern: fetchWithTimeout() in the same file already removes the external abort listener in finally.
• Tests should stay in the existing packages/telemetry/test/telemetry.test.ts file under the AsyncTransport coverage, instead of adding a new test file.
• No API surface, package boundaries, workspace membership, or docs behavior should change.

Implementation plan:

1. Update abortableSleep() so the normal timer callback removes onAbort from the provided signal before resolving.
2. Keep the abort path behavior unchanged: abort should clear the timer and reject with the existing AbortError shape.
3. Prefer a small local cleanup helper or named timer callback only if it keeps the resolve/abort paths clear; avoid broader refactors.
4. Add focused regression coverage using AsyncTransport with a transient fetch response and a short retry backoff, so the real retry sleep path is exercised.
5. In the regression test, wrap or spy on a signal's addEventListener and removeEventListener methods to assert that repeated normal sleep completions remove the registered abort listener.
6. Add or adjust coverage for the abort-during-backoff path only if needed to ensure the listener cleanup change does not alter shutdown abort behavior.

Validation plan:

• Run the telemetry package tests: pnpm --filter @moonshot-ai/kimi-telemetry test.
• Run telemetry typecheck if available through the package script: pnpm --filter @moonshot-ai/kimi-telemetry typecheck.
• Run a targeted test while developing, e.g. the existing telemetry test file or AsyncTransport tests with Vitest filtering.
• Before final PR submission, run the repository-required changeset workflow and choose a patch changeset unless maintainers decide otherwise.

Environment note from investigation:

• In this local shell, pnpm is currently not on PATH, and node reports 24.14.0 while the repository requires >=24.15.0. The implementation phase should first switch to the repo-supported Node/pnpm toolchain before running the validation commands.

[trustmiao]

Execution update for #4

Actual changes:

• Updated packages/telemetry/src/transport.ts so a normally completed telemetry retry delay removes its abort listener before resolving.
• Added regression coverage in packages/telemetry/test/telemetry.test.ts that exercises the real AsyncTransport.send() retry/backoff path and asserts abort listener registrations are removed after successful backoff sleeps.
• Added .changeset/telemetry-abort-listener-cleanup.md with patch entries for @moonshot-ai/kimi-telemetry and @moonshot-ai/kimi-code.

Validation:

• corepack pnpm --filter @moonshot-ai/kimi-telemetry test could not run under the default strict engine check because local Node is v24.14.0, while the repo requires >=24.15.0.
• corepack pnpm --config.engine-strict=false --dir packages\telemetry exec vitest run test/telemetry.test.ts -t "AsyncTransport" passed: 14 passed, 28 skipped.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry typecheck passed.
• Full telemetry test suite under --config.engine-strict=false still has one unrelated Windows/local-environment failure in crash handler > keeps Node default non-zero exit semantics for unhandled rejections: spawn EFTYPE at test/telemetry.test.ts:900 when spawning the tsx .mjs script.
• git diff --check passed for the touched files, with only Git CRLF conversion warnings.

E2E:

• No web UI or user flow was changed, so no e2e test is required for this fix.

Remaining risk:

• Please rerun the full telemetry test command in the supported Node >=24.15.0 environment before merging, because this local machine is on Node v24.14.0.

[trustmiao]

Planning update for #4 (2026-06-17, investigation only; no code changes made in this phase).

Requirement confirmed:

• fix: remove abort listener leak in abortableSleep #4 is a closed PR, but the underlying report is actionable: abortableSleep() in packages/telemetry/src/transport.ts adds an abort listener and previously did not remove that listener when the sleep timer resolved normally.
• The affected runtime path is AsyncTransport.send(): transient telemetry failures enter the retry loop, then call sleepImpl(backoff, signal). Repeated normal backoff completions on the same AbortSignal can retain stale listeners.
• The cleanup behavior should match the existing pattern in fetchWithTimeout() in the same file, which removes the external abort listener after the request completes.

Affected scope found in code:

• Primary implementation: packages/telemetry/src/transport.ts, specifically abortableSleep() and its interaction with AsyncTransport.send() retry backoff.
• Existing nearby behavior to preserve: immediate abort before send, abort during HTTP request, abort during retry backoff, disk fallback for aborted/failed telemetry sends, and transient/non-retryable status handling.
• Test location: existing packages/telemetry/test/telemetry.test.ts, inside the current AsyncTransport coverage. No new test file is needed.
• No workspace/package membership, public API shape, CLI/TUI behavior, web UI, or docs behavior should change.

Implementation plan:

1. Update abortableSleep() so the timer callback removes the exact onAbort listener from the provided signal before resolving.
2. Keep the abort path behavior unchanged: if the signal is already aborted, reject with the existing AbortError; if it aborts during sleep, clear the timer and reject with the same AbortError shape.
3. Avoid broader refactors. Only introduce a small helper or named callback if it keeps the timer and abort paths clearer.
4. Add focused regression coverage through the real AsyncTransport.send() retry path with transient responses followed by success.
5. In that test, spy on signal.addEventListener and signal.removeEventListener and assert each registered abort listener is removed after normal retry backoff sleeps.
6. Recheck the existing abort-during-backoff test so listener cleanup does not alter shutdown/disk-fallback behavior.
7. Add a patch changeset for @moonshot-ai/kimi-telemetry and any required consumer package entry if the repository release rules require it.

Validation plan:

• Use the supported toolchain first: Node >=24.15.0 and corepack pnpm / pnpm 10.33.0.
• Run targeted telemetry tests while developing, for example corepack pnpm --filter @moonshot-ai/kimi-telemetry test or a Vitest filter for AsyncTransport.
• Run telemetry typecheck: corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck.
• Run git diff --check before handoff.
• No browser/e2e validation is required because the affected code is telemetry transport retry logic, not UI behavior.

Local investigation note:

• This workspace already contains uncommitted telemetry changes that appear to implement this fix and regression test. I did not create or modify those files during this planning phase.
• The local default node is v24.14.0, below the repo engine requirement >=24.15.0; pnpm is not directly on PATH, but corepack pnpm reports 10.33.0.

[trustmiao]

Workmonitor issue-flow plan stage 20260618-024737 issue #4

Planning phase only; no code changes made in this phase.

Requirement confirmed:

• fix: remove abort listener leak in abortableSleep #4 is a closed PR, but the underlying report is actionable: telemetry retry backoff sleep can leak AbortSignal listeners.
• The reported target is abortableSleep() in packages/telemetry/src/transport.ts. It adds an abort listener and must remove that listener when the sleep timer resolves normally.
• The affected runtime path is AsyncTransport.send(): transient telemetry send failures enter the retry loop and call sleepImpl(backoff, signal). Repeated normal backoff completions on the same signal can leave stale listeners attached.
• The desired cleanup behavior should match the existing fetchWithTimeout() pattern in the same file, which removes the external abort listener in finally.

Code scope checked:

• Primary implementation: packages/telemetry/src/transport.ts, especially AsyncTransport.send(), fetchWithTimeout(), and abortableSleep().
• Test scope: existing packages/telemetry/test/telemetry.test.ts under the current AsyncTransport coverage. A new test file is not needed.
• Behavior to preserve: immediate abort before send, abort during request, abort during retry backoff, disk fallback on aborted/transient sends, anonymous retry after 401, transient retry handling, and non-retryable 4xx drop behavior.
• No public API shape, package boundaries, workspace membership, CLI/TUI behavior, web UI, or docs behavior should change.
• Local workspace note: this checkout already contains uncommitted telemetry changes that appear to implement the fix and regression test; I treated them as existing local state and did not modify them during this planning phase.

Implementation plan:

1. Update abortableSleep() so the normal timeout callback removes the exact onAbort listener from the provided signal before resolving.
2. Keep the existing abort path semantics unchanged: already-aborted signals reject with the existing AbortError, and abort during sleep clears the timer and rejects with the same AbortError shape.
3. Avoid broader refactors; only use a small named callback/helper if it keeps the timer and abort paths clearer.
4. Add focused regression coverage through the real AsyncTransport.send() retry path with transient responses followed by success.
5. In the regression test, spy on signal.addEventListener and signal.removeEventListener, then assert every registered abort listener is removed after normal retry backoff sleeps complete.
6. Keep or recheck the existing abort-during-backoff coverage so shutdown abort behavior and disk fallback remain unchanged.
7. Add a patch changeset for @moonshot-ai/kimi-telemetry and any required consumer package entry according to the repository changeset rules.

Validation plan:

• Use the supported repository toolchain first: Node >=24.15.0 and pnpm 10.33.0 via Corepack.
• Run targeted telemetry tests during development, especially the AsyncTransport tests in packages/telemetry/test/telemetry.test.ts.
• Run the package test command: corepack pnpm --filter @moonshot-ai/kimi-telemetry test.
• Run telemetry typecheck: corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck.
• Run git diff --check before handoff.
• No browser/e2e validation is required because this is telemetry transport retry logic, not UI behavior.

Environment note from this investigation:

• Current local date/time checked as 2026-06-18 02:48 +08:00.
• Local default node reports v24.14.0, below the repository engine requirement >=24.15.0.
• corepack pnpm --version reports 10.33.0.

[trustmiao]

Workmonitor issue-flow investigation stage 20260618-032544 issue #4

Investigation stage only; no code changes were made in this phase.

Requirement confirmed:

• fix: remove abort listener leak in abortableSleep #4 is a closed PR, but the underlying report is actionable: abortableSleep() in packages/telemetry/src/transport.ts can leak AbortSignal listeners.
• The specific issue is that the helper registers an abort listener, but on normal timer completion that listener must be removed from the same signal.
• The affected runtime path is AsyncTransport.send(): transient telemetry failures enter the retry loop and call sleepImpl(backoff, signal). Repeated successful backoff sleeps with the same AbortSignal can accumulate stale listeners.
• The expected cleanup pattern already exists nearby in fetchWithTimeout(), which removes the external abort listener in finally.

Code scope checked:

• Primary implementation: packages/telemetry/src/transport.ts, especially AsyncTransport.send(), fetchWithTimeout(), and abortableSleep().
• Test scope: existing packages/telemetry/test/telemetry.test.ts under the current AsyncTransport test coverage. A new test file is not needed.
• Behavior to preserve: immediate abort before send, abort during HTTP request, abort during retry backoff, disk fallback for aborted/transient sends, anonymous retry after 401, transient retry handling, and non-retryable 4xx drop behavior.
• No public API shape, package boundaries, workspace membership, CLI/TUI behavior, web UI, or documentation behavior should change.
• Local workspace note: this checkout already contains uncommitted telemetry changes that appear to implement this fix and regression coverage; I treated them as existing local state and did not modify them during this investigation.

Specific implementation plan:

1. Update abortableSleep() so the normal timeout callback removes the exact onAbort listener from the provided signal before resolving.
2. Keep abort semantics unchanged: already-aborted signals reject with the existing AbortError; abort during sleep clears the timer and rejects with the same AbortError shape.
3. Avoid broader refactors. Only introduce a small named callback or helper if it makes the normal-completion and abort paths clearer.
4. Add focused regression coverage through the real AsyncTransport.send() retry path with transient responses followed by success.
5. In the regression test, spy on signal.addEventListener and signal.removeEventListener, then assert that each registered abort listener is removed after normal retry backoff sleeps complete.
6. Recheck existing abort-during-backoff coverage so shutdown abort behavior and disk fallback remain unchanged.
7. For the implementation/PR stage, add a patch changeset for @moonshot-ai/kimi-telemetry and any required consumer package entry according to repository release rules.

Validation plan:

• Use the supported repository toolchain first: Node >=24.15.0 and pnpm 10.33.0 via Corepack.
• Run targeted telemetry tests while developing, especially the AsyncTransport tests in packages/telemetry/test/telemetry.test.ts.
• Run the telemetry package test command: corepack pnpm --filter @moonshot-ai/kimi-telemetry test.
• Run telemetry typecheck: corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck.
• Run git diff --check before handoff.
• No browser/e2e validation is required because this is telemetry transport retry logic, not UI behavior.

Environment note from this investigation:

• Current local date/time checked as 2026-06-18 03:26 +08:00.
• Local default node reports v24.14.0, below the repository engine requirement >=24.15.0.
• corepack pnpm --version reports 10.33.0.

[trustmiao]

Workmonitor issue-flow execute stage 20260618-032544 issue #4

Execution summary:

• Confirmed the investigation-stage plan and applied/kept the telemetry fix in packages/telemetry/src/transport.ts: normal completion of the retry delay now removes the exact abort listener before resolving.
• Added/kept focused regression coverage in packages/telemetry/test/telemetry.test.ts through the real AsyncTransport.send() retry/backoff path. The test spies on addEventListener and removeEventListener and verifies all registered abort listeners are removed after successful backoff sleeps.
• Added/kept .changeset/telemetry-abort-listener-cleanup.md as a patch changeset for @moonshot-ai/kimi-telemetry and @moonshot-ai/kimi-code.

Validation results:

• corepack pnpm --filter @moonshot-ai/kimi-telemetry test could not start under normal repo settings because local Node is v24.14.0, while the repo requires >=24.15.0.
• corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck was also blocked by the same strict engine check.
• corepack pnpm --config.engine-strict=false --dir packages\telemetry exec vitest run test/telemetry.test.ts -t "AsyncTransport" passed: 1 file passed, 14 tests passed, 28 skipped.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry typecheck passed.
• git diff --check -- packages/telemetry/src/transport.ts packages/telemetry/test/telemetry.test.ts .changeset/telemetry-abort-listener-cleanup.md passed, with only Git LF-to-CRLF working-copy warnings.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry test ran the full telemetry suite and failed only in the existing Windows/local-environment crash-handler case: test/telemetry.test.ts:900, spawn EFTYPE while spawning the tsx CLI. Result before that failure: 57 passed, 1 failed.

E2E/API/browser verification:

• Not run because this change is limited to telemetry retry delay cleanup and does not affect web UI, browser interaction, API behavior, or user flow.

Remaining risk:

• Full package validation should be rerun on supported Node >=24.15.0; this machine is currently on Node v24.14.0.
• The full telemetry suite still has the unrelated local spawn EFTYPE failure described above.

This stage intentionally does not close #4.

[trustmiao]

Workmonitor issue-flow e2e stage 20260618-032544 issue #4

E2E/API verification completed for the telemetry retry cleanup.

Command run:

• corepack pnpm --config.engine-strict=false exec tsx %TEMP%\kimi-telemetry-issue-4-api-e2e.mts

What the temporary API e2e script verified:

• Started a real local HTTP server as the telemetry endpoint.
• Sent telemetry through the real AsyncTransport using Node's real fetch path, not a mocked fetch implementation.
• The server returned transient statuses 429 then 500, then 200, forcing two retry backoff sleeps before success.
• Verified 3 HTTP POST attempts reached /events.
• Verified the final payload used kfc_device_id_e2e-device and kfc_e2e_retry_listener_cleanup.
• Wrapped the same external AbortSignal and verified 5 abort listener registrations and 5 matching removals, covering both request timeout listeners and retry sleep listeners.

Result:

• Passed with output: {"ok":true,"httpAttempts":3,"abortListenersAdded":5,"abortListenersRemoved":5,"endpoint":"http://127.0.0.1:51250/events"}.

Notes and remaining risk:

• No browser/Playwright UI flow exists for this telemetry transport change, so the relevant end-to-end validation is API-level with a local HTTP endpoint.
• The command used --config.engine-strict=false because this machine is still on Node v24.14.0, while the repository requires >=24.15.0.
• Full package validation should still be rerun on supported Node >=24.15.0 before merge.

This stage intentionally does not close #4.

[trustmiao]

Workmonitor issue-flow final stage 20260618-032544 issue #4

Final summary for #4:

• The underlying issue was confirmed in telemetry retry backoff handling: the sleep helper registered an external abort listener and needed to remove it when the timer completed normally.
• The implemented fix updates packages/telemetry/src/transport.ts so normal retry-delay completion removes the exact registered abort listener before resolving, while preserving existing abort behavior for already-aborted signals and aborts during sleep.
• Regression coverage was added in packages/telemetry/test/telemetry.test.ts through the real AsyncTransport.send() retry/backoff path. It verifies that all abort listeners registered during successful retry sleeps are removed.
• A patch changeset was added at .changeset/telemetry-abort-listener-cleanup.md for @moonshot-ai/kimi-telemetry and @moonshot-ai/kimi-code.

Validation completed:

• Normal strict-engine telemetry test/typecheck commands were attempted but blocked because this machine has Node v24.14.0, while the repository requires >=24.15.0.
• corepack pnpm --config.engine-strict=false --dir packages\telemetry exec vitest run test/telemetry.test.ts -t "AsyncTransport" passed: 14 passed, 28 skipped.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry typecheck passed.
• git diff --check -- packages/telemetry/src/transport.ts packages/telemetry/test/telemetry.test.ts .changeset/telemetry-abort-listener-cleanup.md passed, with only Git LF-to-CRLF working-copy warnings.
• Full telemetry test run under --config.engine-strict=false reached 57 passed / 1 failed; the remaining failure is the existing Windows/local crash-handler spawn issue at packages/telemetry/test/telemetry.test.ts:900 with spawn EFTYPE while spawning the tsx CLI.

E2E/API status:

• No browser or Playwright UI flow exists for this telemetry transport-only change.
• API-level end-to-end verification was completed with a temporary local HTTP telemetry endpoint and real AsyncTransport + real Node fetch path.
• The API e2e forced transient 429, then 500, then 200 responses, verified 3 HTTP POST attempts, verified the final payload shape, and verified 5 abort listener registrations had 5 matching removals.
• API e2e result passed with output: {"ok":true,"httpAttempts":3,"abortListenersAdded":5,"abortListenersRemoved":5,"endpoint":"http://127.0.0.1:51250/events"}.

Residual risk:

• Full package validation should be rerun on a supported Node >=24.15.0 environment before merge/release.
• The unrelated local spawn EFTYPE crash-handler test failure remains outside this issue's scope.

Requirement status:

• The issue requirement is covered by the implementation, regression test, typecheck, targeted test, diff check, and API-level end-to-end validation. This issue can remain closed / be closed.

[trustmiao]

Workmonitor issue-flow investigation stage 20260618-133001 issue #4

Investigation stage only; no code changes were made in this phase.

Requirement confirmed:

• fix: remove abort listener leak in abortableSleep #4 is a closed PR, but the underlying report is actionable: abortableSleep() in packages/telemetry/src/transport.ts can leak AbortSignal listeners.
• The helper registers an abort listener and must remove that exact listener when the sleep timer resolves normally.
• The affected runtime path is AsyncTransport.send(): transient telemetry send failures enter the retry loop and call sleepImpl(backoff, signal). Repeated normal backoff completions with the same signal can accumulate stale listeners.
• The intended cleanup behavior matches the nearby fetchWithTimeout() pattern in the same file, which removes the external abort listener in finally.

Code scope checked:

• Primary implementation: packages/telemetry/src/transport.ts, especially AsyncTransport.send(), fetchWithTimeout(), and abortableSleep().
• AsyncTransport.send() uses retry backoffs at packages/telemetry/src/transport.ts around the sleepImpl(backoff, signal) call, so the listener cleanup bug is limited to retry delay handling.
• Existing behavior to preserve: immediate abort before send, abort during HTTP request, abort during retry backoff, disk fallback for aborted/transient sends, anonymous retry after 401, transient retry handling, and non-retryable 4xx drop behavior.
• Test scope: existing packages/telemetry/test/telemetry.test.ts under the current AsyncTransport coverage. A new test file is not needed.
• No public API shape, package boundary, workspace membership, CLI/TUI behavior, web UI, or documentation behavior should change.
• Local workspace note: this checkout already contains uncommitted telemetry changes that appear to implement the fix, regression test, and patch changeset. I treated them as existing local state and did not modify them during this investigation.

Specific implementation plan:

1. Update abortableSleep() so the normal timeout callback removes the exact onAbort listener from the provided signal before resolving.
2. Keep abort semantics unchanged: already-aborted signals reject with the existing AbortError; abort during sleep clears the timer and rejects with the same AbortError shape.
3. Avoid broader refactors. Only introduce a small named callback/helper if it makes the normal-completion and abort paths clearer.
4. Add focused regression coverage through the real AsyncTransport.send() retry path with transient responses followed by success.
5. In the regression test, spy on signal.addEventListener and signal.removeEventListener, then assert every registered abort listener is removed after normal retry backoff sleeps complete.
6. Recheck existing abort-during-backoff coverage so shutdown abort behavior and disk fallback remain unchanged.
7. For the implementation/PR stage, add a patch changeset for @moonshot-ai/kimi-telemetry and any required consumer package entry according to repository release rules.

Validation plan:

• Use the supported repository toolchain first: Node >=24.15.0 and pnpm 10.33.0 via Corepack.
• Run targeted telemetry tests while developing, especially the AsyncTransport tests in packages/telemetry/test/telemetry.test.ts.
• Run the telemetry package test command: corepack pnpm --filter @moonshot-ai/kimi-telemetry test.
• Run telemetry typecheck: corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck.
• Run git diff --check before handoff.
• No browser/e2e validation is required because this is telemetry transport retry logic, not UI behavior.

Environment note from this investigation:

• Current local date/time checked as 2026-06-18 13:30:21 +08:00.
• Local default node reports v24.14.0, below the repository engine requirement >=24.15.0.
• corepack pnpm --version reports 10.33.0.

[trustmiao]

Workmonitor issue-flow execute stage 20260618-133001 issue #4

Execution summary:

• Re-read fix: remove abort listener leak in abortableSleep #4 body, comments, and the matching investigation comment for Workmonitor issue-flow investigation stage 20260618-133001 issue #4.
• Confirmed/applied the planned telemetry fix in packages/telemetry/src/transport.ts: normal retry-delay completion removes the exact registered abort listener before resolving.
• Confirmed/kept focused regression coverage in packages/telemetry/test/telemetry.test.ts through the real AsyncTransport.send() transient retry/backoff path. The test spies on addEventListener and removeEventListener and verifies listener removals match registrations.
• Confirmed/kept .changeset/telemetry-abort-listener-cleanup.md as a patch changeset for @moonshot-ai/kimi-telemetry and @moonshot-ai/kimi-code, because the internal telemetry source enters the CLI bundle.
• No web UI, browser interaction, CLI flow, or public API shape was changed.

Validation results:

• Current local time/toolchain check: 2026-06-18 13:32:21 +08:00; Node v24.14.0; corepack pnpm 10.33.0.
• corepack pnpm --filter @moonshot-ai/kimi-telemetry test could not start under strict engine settings because this machine has Node v24.14.0, while the repo requires >=24.15.0.
• corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck was likewise blocked by the same strict engine check.
• corepack pnpm --config.engine-strict=false --dir packages\telemetry exec vitest run test/telemetry.test.ts -t "AsyncTransport" passed: 1 file passed, 14 tests passed, 28 skipped.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry typecheck passed.
• git diff --check -- packages/telemetry/src/transport.ts packages/telemetry/test/telemetry.test.ts .changeset/telemetry-abort-listener-cleanup.md passed, with only Git LF-to-CRLF working-copy warnings.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry test ran the full telemetry suite and failed only in test/telemetry.test.ts:900, crash handler > keeps Node default non-zero exit semantics for unhandled rejections, with spawn EFTYPE while spawning the tsx CLI. Result: 57 passed, 1 failed.

E2E/API verification:

• Because this change affects telemetry HTTP retry behavior, I ran API-level end-to-end verification with a temporary local HTTP telemetry endpoint and real AsyncTransport using Node's real fetch path.
• The endpoint returned transient statuses 429, then 500, then 200, forcing two retry backoff sleeps before success.
• Result passed: {"ok":true,"httpAttempts":3,"abortListenersAdded":5,"abortListenersRemoved":5,"endpoint":"http://127.0.0.1:63212/events"}.
• No browser/Playwright e2e was run because there is no web UI or browser workflow for this telemetry transport-only change.

Remaining risk:

• Full strict validation should be rerun on a supported Node >=24.15.0 environment before merge/release.
• The unrelated local Windows spawn EFTYPE crash-handler test failure remains outside this issue's scope.

This execution stage intentionally does not close #4.

[trustmiao]

Workmonitor issue-flow e2e stage 20260618-133001 issue #4

E2E/API verification completed for the telemetry retry listener cleanup.

Command run:

• corepack pnpm --config.engine-strict=false exec tsx %TEMP%\kimi-telemetry-issue-4-api-e2e-133001.mts

What this verified:

• Started a real local HTTP server as the telemetry endpoint.
• Sent telemetry through the real AsyncTransport implementation using Node's real fetch path, not a mocked fetch implementation.
• The local endpoint returned 429, then 500, then 200, forcing two retry backoff sleeps before success.
• Verified 3 HTTP POST attempts reached /events.
• Verified the final payload used kfc_device_id_e2e-device and kfc_e2e_retry_listener_cleanup.
• Wrapped the same external AbortSignal and verified every abort listener registration had a matching removal.

Result:

• Passed with output: {"ok":true,"endpoint":"http://127.0.0.1:63430/events","httpAttempts":3,"responseSequence":[429,500,200],"abortListenersAdded":5,"abortListenersRemoved":5,"finalUserId":"kfc_device_id_e2e-device","finalEvent":"kfc_e2e_retry_listener_cleanup"}.

Why this is API e2e rather than browser/Playwright:

• The change is limited to telemetry transport retry-delay cleanup and has no web UI, browser interaction, or visible user workflow to exercise in Playwright.
• The relevant end-to-end surface is the HTTP telemetry transport path, so the API-level test exercises the real retry/fetch behavior against a live local endpoint.

Remaining risk:

• The command used --config.engine-strict=false because this machine has Node v24.14.0, while the repository requires >=24.15.0.
• Full strict validation should still be rerun on supported Node >=24.15.0 before merge/release.

This stage intentionally does not close #4.

[trustmiao]

Workmonitor issue-flow final stage 20260618-133001 issue #4

Final summary for #4:

• Re-read the issue body and all comments. The confirmed requirement is covered: telemetry retry backoff sleep must remove the AbortSignal abort listener when the timer completes normally.
• Final implementation in packages/telemetry/src/transport.ts: abortableSleep() now removes the exact registered abort listener in the normal timeout callback before resolving. Existing abort behavior remains unchanged for already-aborted signals and aborts during sleep.
• Regression coverage in packages/telemetry/test/telemetry.test.ts: the real AsyncTransport.send() transient retry/backoff path is exercised with 429, then 500, then 200; the test spies on addEventListener and removeEventListener and verifies every registered abort listener has a matching removal.
• Release metadata is covered by .changeset/telemetry-abort-listener-cleanup.md with patch entries for @moonshot-ai/kimi-telemetry and @moonshot-ai/kimi-code.

Key behavior/file changes:

• packages/telemetry/src/transport.ts: normal retry-delay completion no longer leaves stale listeners attached to the caller-provided AbortSignal.
• packages/telemetry/test/telemetry.test.ts: added focused regression coverage for listener cleanup through the real retry path.
• .changeset/telemetry-abort-listener-cleanup.md: records the patch release impact.
• No web UI, CLI workflow, public API shape, package boundary, or workspace membership behavior changed for this issue.

Validation results:

• Local toolchain check: Node v24.14.0; corepack pnpm 10.33.0. The repo requires Node >=24.15.0.
• Strict-engine commands corepack pnpm --filter @moonshot-ai/kimi-telemetry test and corepack pnpm --filter @moonshot-ai/kimi-telemetry typecheck were blocked by the local Node version.
• corepack pnpm --config.engine-strict=false --dir packages\telemetry exec vitest run test/telemetry.test.ts -t "AsyncTransport" passed: 1 file passed, 14 tests passed, 28 skipped.
• corepack pnpm --config.engine-strict=false --filter @moonshot-ai/kimi-telemetry typecheck passed.
• git diff --check -- packages/telemetry/src/transport.ts packages/telemetry/test/telemetry.test.ts .changeset/telemetry-abort-listener-cleanup.md passed, with only Git LF-to-CRLF working-copy warnings.
• Full telemetry suite under --config.engine-strict=false reached 57 passed / 1 failed; the remaining failure is the unrelated local Windows crash-handler spawn issue at packages/telemetry/test/telemetry.test.ts:900 with spawn EFTYPE while spawning the tsx CLI.

E2E/API status:

• Browser/Playwright e2e is not applicable because the change is telemetry transport retry logic and has no web UI or browser workflow.
• API-level e2e was completed with a temporary local HTTP telemetry endpoint and the real AsyncTransport + real Node fetch path.
• The API e2e forced 429 -> 500 -> 200, verified 3 POST attempts, verified payload fields, and verified abort listener registrations/removals matched.
• API e2e passed with output: {"ok":true,"endpoint":"http://127.0.0.1:63430/events","httpAttempts":3,"responseSequence":[429,500,200],"abortListenersAdded":5,"abortListenersRemoved":5,"finalUserId":"kfc_device_id_e2e-device","finalEvent":"kfc_e2e_retry_listener_cleanup"}.

Residual risk:

• Full strict validation should be rerun on supported Node >=24.15.0 before merge/release.
• The unrelated local Windows spawn EFTYPE crash-handler test failure remains outside this issue's scope.

Requirement status:

• The requested fix, regression test coverage, changeset, targeted validation, typecheck, diff check, and API-level e2e are covered. fix: remove abort listener leak in abortableSleep #4 is already closed and can remain closed.