fix(client-runtime): stop resubscribing threads the server reports missing

[lnieuwenhuis]

Every foreground wakeup re-issues subscribeThread for a thread the server already reported missing, and the debounced persistence writer can resurrect the thread right after cache deletion.

Return a dedicated missing-thread error and stop the durable subscription across reconnects and foreground wakeups. Serialize cache saves and removal so queued or in-flight writes cannot restore a deleted thread. Other snapshot failures still retry.

Reimplements the intent of #8192 (which only drained the inner stream) on the current client-runtime layout, with regressions for single-attempt termination, no foreground resubscribe, and no persistence resurrection.

Built with muse-spark-1.3-contributor via OpenCode in T3 Code. Review follow-ups by GPT-6 via Codex.

---

Note

Medium Risk
Changes subscribeThread wire errors and client thread subscription, cache deletion, and persistence ordering; incorrect classification could stop sync early or leave stale cache.

Overview
Introduces OrchestrationThreadNotFoundError on the subscribeThread RPC and has the server emit it when a thread snapshot is unavailable (no replay fallback), instead of overloading OrchestrationGetSnapshotError.

On the client, subscribeDynamic gains a terminalFailure path that halts session- and wakeup-driven resubscribes after a classified failure. Thread sync treats not-found as terminal: it sets a tombstone latch, marks the thread deleted, removes cache, and does not retry on foreground wakeups or session replacement. Persistence is tightened with a lock, skipping writes after deletion, polling the debounced queue before removeThread, so stale snapshots cannot resurrect a deleted thread.

Other snapshot failures still use the existing retry and resubscribe behavior. Tests cover RPC decoding, terminal vs retriable failures, and deletion parity with thread.deleted.

^{Reviewed by Cursor Bugbot for commit 74ce349. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Stop resubscribing threads the server reports as missing

• Adds OrchestrationThreadNotFoundError to the contracts layer and the subscribeThread RPC error union; the server WebSocket layer now returns this typed error with the missing threadId instead of a generic snapshot error
• Introduces a terminalFailure classifier in subscribeDynamic that halts all session-driven resubscription when every error in a cause matches the classifier, then runs a terminal handler once
• makeEnvironmentThreadState classifies OrchestrationThreadNotFoundError as terminal, marks the thread deleted via setDeleted, and blocks foreground, probe, and session-replacement resubscriptions; generic snapshot errors remain retryable
• setDeleted now drains the pending persistence queue and waits for an in-flight cache save before removing the cache entry; the persistence worker skips snapshots when the thread is already deleted
• Risk: setDeleted in threads.ts acquires a persistence semaphore that serializes cache removal with saves — if persist blocks indefinitely on a cache write, deletion will also block; any caller that previously retried on a generic not-found message now needs to handle the typed error or it will fall into the ordinary retry path

^{Macroscope summarized 74ce349.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This change modifies an existing production subscription path across the server protocol, client runtime, thread state, and cache persistence. It introduces terminal stream halting and concurrency-sensitive deletion ordering, so the runtime and wire-level effects are broader than a small isolated bug fix.

Not approved because:

• Monthly spending limit reached (workspace setting). Approvability relies on correctness review in order to determine eligibility

^{Review your spending limits in Billing settings. You can add or adjust custom eligibility rules. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 0895455. Configure here.}