fix(events): cancel orphaned IPC renders, live-socket ownership, streaming UTF-8, ms timeout precision, fresh diagnostic

[ScriptedAlchemy]

Summary

Fixes the five event-IPC findings from #180 review:

• ipc.ts:186 (P1): the per-connection handler now receives an AbortSignal tied to socket teardown (close/end/error), and the connection fiber itself is interrupted — a client disconnect or timeout cancels the in-flight shared render instead of letting it run to completion invisibly. mcp-server-runtime threads the signal into dispatch.
• ipc.ts:223 (P1): startup no longer unconditionally unlinks the endpoint. It probes the existing socket first: a live server fails startup with a typed error, only a stale (ECONNREFUSED) file is removed. Shutdown removes the endpoint only when this instance still owns it (device/inode identity check), so a replacement bound later is never deleted.
• ipc.ts:117 (P2): request decoding uses a per-connection StringDecoder with byte-accurate limit accounting and newline-delimited framing — multibyte code points split across chunks decode correctly.
• normalize.ts:447 (P2): the canonical hook model now carries timeoutMs end-to-end; conversion to host seconds happens only at native host projection (docs/tests updated). Sub-second event-route timeouts survive to the generated wrapper.
• validate.ts:1722 (P2): the missing-shared-runtime condition gets fresh code AB4817 (AB4816 stays retired per the no-reuse rule); docs/diagnostics.md registry updated.

Test plan

• New transport regressions: disconnect interrupts the in-flight handler; split-multibyte decode; live-server rejection without disturbing the owner; stale-socket replacement (ported to the effect-rstest harness from test(effect): adopt effect-rstest helpers #192)
• Timeout precision fixture (1250 ms preserved in the model, host doc projects ceil-seconds); AB4817 assertions
• Scoped suites (event-ipc, hooks, route-graph, target-hook-contract, cursor-adapter), pnpm typecheck, pnpm lint all green after rebase onto current main

[changeset-bot]

🦋 Changeset detected

Latest commit: ea7b0b4

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
agent-bundle	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-01T21:11:40.277742Z	ea7b0b4	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ea7b0b4c04

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]

Recheck stale endpoint ownership before unlinking

When two runtime processes start concurrently while a stale endpoint exists, both can observe endpointState === 'stale'; after the first process removes the stale file and successfully binds, the second can execute this unconditional rm(endpoint) and unlink the first process's live socket. The first server remains running but becomes unreachable to new hook clients, so stale cleanup must be tied to the probed file identity or otherwise serialized/rechecked immediately before removal.

Useful? React with 👍 / 👎.

[ScriptedAlchemy]

Fixed in #216. Stale-endpoint cleanup is now performed only while holding an exclusive endpoint claim (<endpoint>.lock created with open('wx')), with a fresh probe re-run under the lock: a re-probed live endpoint fails with the existing live-server error instead of being removed, so a concurrent winner's socket can never be unlinked. A held claim is awaited with bounded retries, and an orphaned claim from a crashed process fails closed with a typed error rather than being stolen (stealing would recreate the same unlink race). Claim release is dev/inode identity-checked, mirroring removeOwnedEndpoint. Deterministic regression test freezes two servers in the historical probe window and asserts the loser errors while the winner keeps serving on its original socket.