fix(server): a delegated child's answer after its monitor ends reaches the parent by juliusmarminge · Pull Request #31856 · pingdotgg/t3code

juliusmarminge · 2026-11-06T06:58:30Z

Problem

A delegated child can arm a Monitor, skip t3_background_wait, and end its turn. Unawaited monitors don't hold completion, so the task finalizes with its "monitor armed" text; that part is by design. When the Monitor later ends, Claude wakes the child and it answers, but that answer never reaches the parent. The same wake from a background command ending does reach it (https://git.cafe/pingdotgg/t3code/pulls/23542).

To reproduce on main (5c004fd), with a dev server on a fresh database: a Claude coordinator delegates an async Claude worker with the task "use only the Monitor tool with sleep 60; echo doneD | tee …, don't call t3_background_wait, end your turn, then reply with what it reported".

The task finalizes about 0.5 s after the Monitor is armed.
Sixty seconds later the child is woken with {source:{kind:"monitor"}, outcome:"completed"} and replies doneD.
The parent gets no subagent.updated and no delivery.

Change

mayStartSubagentFollowup (apps/server/src/orchestration-v2/Orchestrator.ts) blocked every monitor-kind notification from starting a subagent follow-up. It now blocks only monitor notifications with outcome "updated", which are per-event progress. A monitor that ends (completed, failed, cancelled) now starts a follow-up the same way a background command's ending wake does, and its result is delivered to the parent.

The parent-lock decision at dispatch uses the same predicate, so it follows automatically. Nothing else changes: the awaited/hold policy (backgroundWorkHoldsCompletion), the prompts, and the contracts.

PR-watch notifications also use the monitor kind, but a subagent cannot hold a watch. watch_pull_request is hidden from subagents and rejected by the handler, and the reactor ends any watch on a subagent thread.

Scope and approval

Fixes #31843

Verification

vp test run apps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.ts: 40/40 pass.
- The new case delivers the run a monitor's ending wake starts after the original result fails on main with the run woken by the child's monitor ending should be tracked as a follow-up delegated task for the parent: expected undefined to not equal undefined.
- The background-command case still passes.
- The existing delivers a %s follow-up independently of the original task case still asserts that an "updated" monitor notification starts no follow-up.
vp run check:changed passes after merging main: format, lint, knip, graphql, typecheck (3 packages), and 5 test files with 60 tests.
Runtime check with the real dev server, a fresh database, and the same coordinator and worker prompt on each side:

Before, main 5c004fd: worker D's monitor-end answer never reaches the parent

# seq|time UTC|event|thread|status|item|text|pendingBackgroundTasks
745|06:38:28.805|provider-thread.updated|child |active|||monitor:b1f0w07xb:aw=-
756|06:38:30.513|run.updated|child |waiting|||
764|06:38:31.059|subagent.updated|parent|completed||The Monitor is armed (task `b1f0w07xb`). It runs `sleep 60; echo doneD | tee …` and …
815|06:39:28.969|message.updated|child |||Monitor "Waiting for doneD marker after 60s sleep" stream ended   (notification {source:monitor, outcome:completed})
826|06:39:30.519|message.updated|child |||doneD
837|06:39:30.562|run.updated|child |completed|||
(no further subagent.updated or delivery on the parent)

After, fix/issue-31843 405e1db: the monitor-end answer is delivered and wakes the parent

# seq|time UTC|event|thread|status|item|text
124|06:52:13.596|subagent.updated|parent|completed||The Monitor is armed (task `b9c3z2yxp`) and will emit `doneD` after about 60 seconds. …
169|06:53:11.584|message.updated|child |||Monitor "Wait 60s then emit doneD marker" stream ended
185|06:53:12.959|message.updated|child |||doneD
198|06:53:13.554|subagent.updated|parent|completed||doneD
219|06:53:15.624|message.updated|parent|||Worker D's second result arrived and it is `doneD`, …

Not checked: live Grok or ACP monitors. From the code, their monitor wakes carry only terminal outcomes.

AI-assisted; verified with the tests and tools listed above.

A delegated child woken by a monitor's terminal notification (completed, failed, cancelled, unknown) answered without starting a subagent follow-up, so its new result never reached the parent. Monitor notifications now gate on outcome: only "updated" progress events skip the follow-up, matching how a background command's ending wake works. AI-assisted; verified with the tests and tools listed above.

macroscopeapp · 2026-11-06T07:00:25Z

Approvability

Verdict: Approved at a99108b

Macroscope's review found this PR approvable — Monitor-end notifications now let a delegated child’s subsequent answer reach its parent through existing follow-up handling. Progress updates remain excluded, and existing task guards and parent locking constrain the change.

No code changes detected at a61fc32. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

github-actions · 2026-11-06T07:01:04Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: d290c1f · PR result: a61fc32 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 108.5 KiB
Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

coderabbitai · 2026-11-06T07:01:48Z

No actionable comments were generated in the recent review.

Walkthrough

The orchestrator now allows monitor notifications whose outcome is not "updated" to start delegated follow-up work. Tests cover command and monitor notifications through follow-up completion, claiming, and a distinct completion wake for the parent.

Changes: Delegated follow-up

Layer / File(s)	Summary
Monitor notification follow-up rule `apps/server/src/orchestration-v2/Orchestrator.ts`	`mayStartSubagentFollowup` now allows monitor notifications unless their outcome is `"updated"`.
Follow-up completion and parent wake `apps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.ts`	Tests cover command and monitor notifications, verify that the follow-up task completes and is claimed, and check that recovery requests a distinct completion wake for the parent.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Medium

Suggested reviewers: yash-singh1

Merge Risk: Minimal · up to a61fc

Terminal monitor notifications can start the intended follow-up and deliver its result to the parent. No actionable merge-blocking risk remains after normal checks.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly describes the fix: a delegated child’s answer after its monitor ends reaches the parent.
Description check		The description includes the required Problem, Change, Scope and approval, and Verification sections. It gives detailed reproduction and test results and links the issue. It says AI-assisted but does …
Linked Issues check		[`#31843`] The predicate change lets terminal Monitor notifications start a delegated follow-up while keeping `updated` notifications excluded. The delivery test checks follow-up tracking, result delive…
Out of Scope Changes check		The Orchestrator predicate change and delegated-delivery test both support [`#31843`]. The test retains background-command coverage and checks the Monitor behavior. The change summary identifies no unre…

Full details: Description check

Explanation

The description includes the required Problem, Change, Scope and approval, and Verification sections. It gives detailed reproduction and test results and links the issue. It says AI-assisted but does not name the model and harness, as the template requests when an agent was used.

Full details: Linked Issues check

Explanation

[#31843] The predicate change lets terminal Monitor notifications start a delegated follow-up while keeping updated notifications excluded. The delivery test checks follow-up tracking, result delivery, and a parent wake. The reported runtime check confirms the child’s later answer reaches the parent. This implements the issue’s later-notification approach without changing the awaited-work policy.

Full details: Out of Scope Changes check

Explanation

The Orchestrator predicate change and delegated-delivery test both support [#31843]. The test retains background-command coverage and checks the Monitor behavior. The change summary identifies no unrelated changes.

Finishing Touches

Generate docstrings

Commit to this branch
Create a new PR

Generate unit tests (beta)

Commit to this branch
Create a new PR

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Usage-based review receipt

Mode: Continue automatically
Reviewed files: 2
Charged: $0.50
View usage details

Note

This review was completed with usage-based billing: files reviewed beyond your plan's included limits are billed at $0.25/file. View usage-based billing.

Recent review info

Run configuration

Source mode: Auto

Repository: pingdotgg/t3code

Branch: fix/issue-31843

File: .coderabbit.config.ts

Commit: a61fc32

Layers: Repository TS config

Review profile: CHILL

Plan: Team

Run ID: `9de0152d-c728-4c65-a6ee-5669251f7b41`

Commits

Reviewing files that changed from the base of the PR and between `a99108b` and `a61fc32`.

Files selected for processing: 2

`apps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.ts`

`apps/server/src/orchestration-v2/Orchestrator.ts`

Limit details: You’ve used all 10 included reviews currently available.

juliusmarminge · 2026-11-10T21:59:45Z

Evidence

Run the same coordinator prompt on both commits: delegate an asynchronous local Claude worker using only Monitor with sleep 60; echo MONITOR_DONE, omit t3_background_wait, finish initially with WAITING, and answer FINAL when Monitor ends. Let the parent finish; inspect the worker and then the parent after Monitor completes.

Both captures: real development web client in Chromium on Linux, dark theme, 1280×800 CSS viewport, separate fresh databases, identical synthetic workspace and driver. Setup trimmed; private runtime metadata cropped or labeled as redacted; remaining footage plays in real time.

Before	After
d290c1f1b6	a61fc3286e
Before recording	After recording
Worker answers `FINAL`; parent remains at `WAITING`.	Separate delegated `FINAL` result reaches the parent, which answers `FINAL`.

Driver · Seed · Observation script · Rendering script · Environment and instructions

Supporting persisted observations: base, head. No tests used as evidence.

AI-assisted; verified with the tests and tools listed above.

Not shown:

Monitor progress updates and unsuccessful endings were not exercised; this reproduction covers a successful Monitor stream end.
Other providers, remote delegation, desktop, and mobile were not exercised; verification followed the requested local Claude flow through the web client.

juliusmarminge · 2026-11-13T21:29:56Z

Thanks for this. Main has the change since #36339.

Main now delivers a delegated child's answer when its Monitor ends while suppressing Monitor progress follow-ups. I compared this PR's complete diff with the merge: mayStartSubagentFollowup has the identical production predicate, and main's existing delivery test now covers monitor/completed; the remaining differences restructure tests and clarify a comment.

Closing as superseded.

juliusmarminge added 3 commits November 6, 2026 06:43

test(server): reproduce #31843 monitor-end wake not delivered to parent

1dc0530

Merge remote-tracking branch 'origin/main' into fix/issue-31843

a99108b

github-actions Bot added vouch:trusted Author is trusted by repo permissions or the VOUCHED list. size:XS 0-9 changed lines (additions + deletions). labels Nov 6, 2026

macroscopeapp Bot approved these changes Nov 6, 2026

View reviewed changes

juliusmarminge added bugfix Fix PR opened by an issue lead; enters the PR loop pr:gate-failed A land gate failed labels Nov 6, 2026

Merge remote-tracking branch 'origin/main' into fix/issue-31843

a61fc32

This was referenced Nov 13, 2026

[Bug]: Claude's per-line Monitor wakes are tagged background_task, so each one wakes a delegated child's parent #36113

Closed

App-owned follow-up completes while nested work runs, then loses the final result #21837

Closed

juliusmarminge mentioned this pull request Nov 13, 2026

fix(server): a Claude Monitor line no longer wakes a delegated child's parent #36339

Merged

juliusmarminge closed this Nov 13, 2026

Navigation Menu

Repository navigation

fix(server): a delegated child's answer after its monitor ends reaches the parent - #31856

fix(server): a delegated child's answer after its monitor ends reaches the parent#31856
juliusmarminge wants to merge 4 commits into
mainfrom
fix/issue-31843

juliusmarminge commented Nov 6, 2026

Uh oh!

macroscopeapp Bot commented Nov 6, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Nov 6, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Nov 6, 2026 •

edited

Loading

Uh oh!

juliusmarminge commented Nov 10, 2026

Uh oh!

juliusmarminge commented Nov 13, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

juliusmarminge commented Nov 6, 2026

Problem

Change

Scope and approval

Verification

Uh oh!

macroscopeapp Bot commented Nov 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Approvability

Uh oh!

github-actions Bot commented Nov 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Thread transfer impact

Uh oh!

coderabbitai Bot commented Nov 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Usage-based review receipt

Uh oh!

juliusmarminge commented Nov 10, 2026

Evidence

Uh oh!

juliusmarminge commented Nov 13, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

macroscopeapp Bot commented Nov 6, 2026 •

edited

Loading

github-actions Bot commented Nov 6, 2026 •

edited

Loading

coderabbitai Bot commented Nov 6, 2026 •

edited

Loading