Best possible solution:

Keep the useful shard isolation and log-record lookup changes, but drain the expected ws abort error during the open-timeout close path before merge.

Do we have a high-confidence way to reproduce the issue?

Do we have a high-confidence way to reproduce the issue? Source-reproducible, but not live-reproduced in this read-only pass: current main shows the non-isolated shard and first-line log assumption, and upstream ws source confirms the PR's timeout path can emit an unhandled error.

Is this the best way to solve the issue?

Is this the best way to solve the issue? No, not as submitted. The selected stabilization surfaces are narrow, but the websocket open timeout must keep or install an error listener while closing a CONNECTING socket.

Full review comments:

• [P2] Drain abort errors when timing out websocket opens — src/gateway/shared-auth.test-helpers.ts:32-33
  When the open timeout fires, cleanup() removes the error listener before ws.close(). In ws 8.20.0, closing a CONNECTING socket aborts the handshake and emits an error on the next tick, so this can crash the Vitest worker instead of producing the bounded timeout failure this helper is meant to add.
  Confidence: 0.9

Overall correctness: patch is incorrect
Overall confidence: 0.86

Acceptance criteria:

• pnpm exec oxfmt --check --threads=1 src/gateway/shared-auth.test-helpers.ts
• OPENCLAW_VITEST_MAX_WORKERS=2 OPENCLAW_VITEST_NO_OUTPUT_TIMEOUT_MS=60000 pnpm exec node scripts/run-vitest.mjs run --config test/vitest/vitest.gateway-server.config.ts --reporter=dot
• pnpm check:changed

What I checked:

• Protected PR state: The PR is open, ready for review, and labeled maintainer, so cleanup policy keeps it open even aside from review findings. (49c1506e8be1)
• Gateway shard isolation source: Current main's gateway-server config does not pass isolate, while createScopedVitestConfig resolves isolation from env and uses the non-isolated runner when isolation is false. (test/vitest/vitest.gateway-server.config.ts:11, e8d0cf75ea0e)
• Request-trace flake source: Current main parses only the first log line before asserting trace fields, so an unrelated startup record can make the test inspect the wrong JSON record. (src/gateway/server-http.request-trace.test.ts:91, e8d0cf75ea0e)
• Introduced websocket timeout bug: The PR timeout path calls cleanup() before ws.close(), removing the error listener before aborting a CONNECTING websocket. (src/gateway/shared-auth.test-helpers.ts:32, 49c1506e8be1)
• ws dependency behavior: ws 8.20.0 close() on a CONNECTING socket calls abortHandshake; that path schedules emitErrorAndClose, which emits error, so the PR must keep or install an error listener while closing. (49c1506e8be1)
• Routing history: GitHub commit history for the touched paths points to Vincent Koc for recent gateway-server shard/request-trace work and Peter Steinberger for the shared-auth helper and network-runtime e2e.

Likely related people:

• vincentkoc: Recent history shows Vincent Koc authored gateway-server shard stabilization and the request trace propagation test area affected by this PR. (role: recent gateway-server shard and request-trace maintainer; confidence: high; commits: 833a42c25338, 3ad0ad994af2, 3ae6f01d619e; files: test/vitest/vitest.gateway-server.config.ts, test/vitest/vitest.scoped-config.ts, src/gateway/server-http.request-trace.test.ts)
• steipete: Peter Steinberger introduced the shared-auth websocket helper, has recent shared-auth flake fixes, and authored the gateway env proxy dispatcher e2e surface. (role: shared-auth and gateway network-runtime maintainer; confidence: high; commits: ee54a8d298ce, 15f2cebc3ee2, d513dc71469f; files: src/gateway/shared-auth.test-helpers.ts, src/gateway/server.shared-auth-rotation.test.ts, src/gateway/server-network-runtime.e2e.test.ts)
• ngutman: Nimrod Gutman has prior merged work reducing shared auth rotation teardown flakes in the same helper/test family. (role: adjacent shared-auth flake contributor; confidence: medium; commits: 6474795890be; files: src/gateway/server.shared-auth-rotation.test.ts)

Remaining risk / open question:

• No live Vitest reproduction was run in this read-only review; the positive flake assessment rests on source inspection and the PR author's reported validation.
• Current main has advanced past the PR base, so changed-gate proof should be refreshed after any rebase or repair.

Codex review notes: model gpt-5.5, reasoning high; reviewed against e8d0cf75ea0e.