🤖 refactor: replace the bash-monitor wake outbox with a level-triggered reconciler - #4029
Conversation
_Generated with `xum` • Model: `openai:gpt-5.6-sol` • Thinking: `high` • Cost: `$5.98`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=high costs=5.98 -->
|
@codex review |
This comment has been minimized.
This comment has been minimized.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Security Review
Here are some automated security review suggestions for this pull request.
Reviewed commit: f20d1fa739
ℹ️ About Codex security reviews in GitHub
This is an experimental Codex feature. Security reviews are triggered when:
- You comment "@codex security review"
- A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review
Once complete, Codex will leave suggestions, or a comment if no findings are found.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f20d1fa739
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".
|
@codex review |
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a75438b78f
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".
|
@codex review |
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 247619523e
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".
|
@codex review |
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b234d7379f
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".
|
@codex review |
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 433f3c1b39
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".
|
@codex review |
|
Codex Review: Didn't find any major issues. Another round soon, please! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
This comment has been minimized.
This comment has been minimized.
1 participant
Summary
Replaces the edge-triggered bash-monitor wake outbox (
bashMonitorWakeStore.ts, 3.5k LOC production + 5.5k LOC tests, plus ~1.9k wake lines insideworkspaceService.ts) with a level-triggeredBashMonitorWakeReconciler. One durable watermark comparison replaces five separate failure-recovery mechanisms; process events only poke the reconciler and are never the correctness mechanism. Net −13,596 LOC across production and tests.Background
Card 1 of the 2026-08-31 architecture review. The wake store persisted every matched-line/settlement event as a durable wake record and kept exact-delivery correct through staged 2PC clears with tombstones, staged-clear heartbeat refreshes, deferred temp-file recovery timers, quarantine dirs with stranded-prune recovery, and snapshot rollback/restore tokens, with three lock maps and clear-commit retry timers leaking into
workspaceService.ts. Edge-triggered wake-delivery hardening was already rejected once in favor of level-triggered reconciliation (#3990,terminalAttentionStore); this module was the remaining instance of that pattern.Implementation
Durable state shrinks to two small things:
bashMonitorRegistryStore.ts, existing): lifetime extended from "until monitor retires" to "until all signals delivered"; rows gain an optional terminal summary (written at settlement) and a lost summary (written on runtime monitor failure), so a restart can deliver a real settlement/runtime-failure notice instead of a generic monitor-lost.matchedThroughOffsetdelivered, terminal delivered, lost delivered.reconcile(workspaceId)(single-flight per workspace) derives outstanding signals from liveBackgroundProcessManagerstate past the watermark: matched lines beyondmax(watermark, shownThroughOffset), undelivered settlements (wakeOnExit), and registry rows without a live monitor. It composes one synthetic turn (prompt format unchanged), sends when the owner is idle, advances the watermark on acceptance, then deletes fully-delivered dead rows. All events (monitor:match/armed/stopped,output:shown, startup, history clear) reduce toscheduleReconcilepokes. Dispatches reportin-flightvsdeferredso a busy-owner deferral clears dispatch dedupe state and the post-idle re-poke redelivers.History clear becomes a watermark bump (pre-bump, clear, post-bump; fails toward cleared); truncations no longer touch wake state.
pendingWakeKindfor the UI derives from the same reconciler snapshot. Legacy wake-record files are deleted best-effort once per workspace, never parsed.Intentional behavior change: matched-line content that was never delivered before an app shutdown is not replayed after restart; the workspace still wakes with a settlement-summary/monitor-lost notice naming the process and filter.
Validation
make static-check, typecheck, and the four touched suites green on this head (reconciler 12, backgroundProcessManager 127, registry 8, workspaceService 401).workspaceService.ts:12890); the guarded invariant (clear consumes pending signals durably; post-clear matches wake) is covered by reconciler unit tests and a workspaceService wiring test.Risks
Wake delivery is intricate timing-sensitive logic feeding agent turns; the regression surface is background-bash monitor wakes (line-match, settlement, monitor-lost), the pending-wake UI indicator, and history-clear interaction. Mitigations: the behavioral invariants are pinned by tests on the
reconcile()surface (busy deferral redelivery, watermark dedup across reconstruction, cancellation silence, maxEvents retirement, generation reuse, clear pre/post consumption, tail frontier filtering), and two UAT rounds exercised the real UI end to end. The restart-content change above is the one deliberate semantic difference.Generated with
xum• Model:anthropic:claude-fable-5• Thinking:xhigh• Cost:$120.95