Best possible solution:

Keep the current main implementation and close this PR as redundant. If maintainers still want the exact extra iMessage regression test from the branch, that should be a small follow-up on top of main rather than merging an already-implemented duplicate patch.

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

Yes. The reproduction path is a closed or errored imsg rpc stdin while IMessageRpcClient.request() is pending; current main can be verified by the stdin error listener and write callback that reject pending requests instead of leaving an uncaught EPIPE.

Is this the best way to solve the issue?

Yes for the underlying issue. Handling this in the iMessage-owned RPC client is the narrowest maintainable fix, and adjacent generic stdin write sites were handled separately in the merged #75602 work.

Security review:

Security review cleared: The PR diff only changes the iMessage client, adds a focused test, and updates the changelog; it does not alter dependencies, CI, packaging, secrets, or downloaded code paths.

What I checked:

• current implementation: stdin error listener: Current main attaches an error listener to child.stdin and routes stream errors through failAll, preventing an unhandled stdin stream error from escaping to the gateway-level uncaught exception path. (extensions/imessage/src/client.ts:111, 0bb52118e6c8)
• current implementation: write callback rejection: Current main writes RPC request payloads with a callback and rejects the matching pending request when the write callback receives an error. (extensions/imessage/src/client.ts:189, 0bb52118e6c8)
• current changelog documents the iMessage fix: The active changelog records the iMessage IMessageRpcClient stdin write callback and error listener fix for IMessageRpcClient.request() does not handle async stdin write errors → uncaughtException → gateway crash #75438. (CHANGELOG.md:99, 0bb52118e6c8)
• release provenance: git branch --contains shows the proof commit on main; git tag --contains returned no tags, so this is implemented on current main but not proven shipped in a release. The latest release in the provided context is v2026.4.29 at a448042, before the proof commit. (5c7362fe9d69)
• dependency contract: Node v22 stream docs confirm writable write callbacks receive write errors before the stream error event, so the current callback-plus-listener shape matches the upstream stream contract for this failure mode.
• related merged stdin work: The provided GitHub context shows fix: handle EPIPE errors on child process stdin writes #75602 was merged for adjacent child-process stdin EPIPE handling in MCP stdio and process exec paths, while current main separately contains the iMessage-specific implementation. (22e9656f54d1)

Likely related people:

• steipete: The current-main proof commit and blame for the iMessage stdin listener/write-callback lines point to Peter Steinberger, and git shortlog --all -- extensions/imessage shows the largest share of iMessage-area history. (role: recent maintainer and proof-commit author; confidence: medium; commits: 5c7362fe9d69; files: extensions/imessage/src/client.ts, CHANGELOG.md)
• vincentkoc: Recent iMessage monitor/type-check/refactor history includes multiple commits by Vincent Koc, and the area is adjacent to the RPC lifecycle that uses IMessageRpcClient. (role: recent iMessage maintainer; confidence: medium; commits: ea71a59127, 35a784c165, 503b43f43f; files: extensions/imessage/src/monitor/monitor-provider.ts, extensions/imessage/src/client.ts)
• amknight: The related merged fix: handle EPIPE errors on child process stdin writes #75602 in the provided context handled the same child-process stdin EPIPE class in sibling MCP stdio and process exec paths. (role: adjacent stdin EPIPE fix owner; confidence: medium; commits: 22e9656f54d1; files: src/agents/mcp-stdio-transport.ts, src/process/exec.ts)

Codex review notes: model gpt-5.5, reasoning high; reviewed against 0bb52118e6c8; fix evidence: commit 5c7362fe9d69, main fix timestamp 2026-05-01T21:08:43Z.