Repository navigation
feat(web): resume a paused queue from the queue itself - #14617
juliusmarminge wants to merge 810 commits into
Conversation
…#12514) Co-authored-by: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Build run-control callbacks outside the startup scope and release the initial handoff callback after turn acceptance or failure. Background event workers retain live routing identities without pinning completed thread history. Add isolated GC regressions for regular turns, handoffs, compaction, and failed startup.
Read lineage, delivery ownership, queue eligibility, and checkpoint capture targets without decoding unrelated tool history. Preserve full handoff and delegated cohort state when a queued run or delegated delivery actually needs it.
Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
A conflict from ProjectService.create means another project owns the claimed folder, so createNamedProject no longer removes it then. Other rejections remove it only after checking that no project owns the path. Cleanup before the create is dispatched now runs on any non-success exit, an interrupt included, so cancelled creates no longer leave folders behind and use up the -2...-100 names. After the create is dispatched an interrupt still keeps the folder, because the create may have committed. t3_project_create without workspaceRoot rejects defaultModelSelection like scripts. Project creation records no model default (only an update does), so it would otherwise be dropped silently. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…efactor Brings in #14546 (stop stranded thinking after a steer), a client-only change to four web and mobile files this pair does not touch. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
After stopping a run the queue is paused, and the composer's only action continued the stopped run with "Continue where you left off." There was no way on web to send the paused queued message itself. The queue header now reads "Queue paused" with a Resume queue action that sends the next queued message. The composer's Resume keeps continuing the stopped run first. Mobile's existing Resume queue label no longer claims the hold came from a restart. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Thread transfer impact✅ Thread transfer remains within every enforced ceiling.
Baseline: unavailable · PR result: Scenario and decoded snapshot size10 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.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed. |
ApprovabilityVerdict: Not approved Macroscope's review found this PR not approvable — The requested comparison is a large multi-area branch rather than an isolated queue enhancement, with new runtime workflows, authentication-sensitive changes, production deployment/default changes, and added diagnostic suppressions. An unresolved High-severity finding also identifies publicly dispatchable server lifecycle callbacks that require human attention. Not approved because:
Adjust the Minimum Blocking Severity for this repo — including turning it Off — in Settings. You can add or adjust custom eligibility rules. Learn more. |
Resuming mid-edit would send the saved text instead of the draft being edited. The action is disabled during an edit, with a tooltip saying why, and the handler rejects it too. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The button stayed enabled during thread detail loading and checkpoint reverts, where onResume returns without doing anything. It now uses the same conditions. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…s the queue The server rejects queue.resume while a usage-limited run blocks the queue, so the action failed and replaced the limit error. It is disabled in that state, with a tooltip pointing to the composer's Resume, which continues the limited run and then the queue. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
| Schema.Struct({ type: Schema.Literal("start_immediately") }), | ||
| ]), | ||
| }), | ||
| Schema.Struct({ |
There was a problem hiding this comment.
🟠 High src/orchestrationV2.ts:2666
Authenticated clients can dispatch prepared-run.release, prepared-run.progress, and prepared-run.fail, allowing them to start or terminate a run before worktree/setup completes. These server-only lifecycle callbacks are currently part of the public OrchestrationV2Command schema; move them to OrchestrationV2InternalCommand so dispatchCommand cannot accept them.
🚀 Reply "fix it for me" or copy this AI Prompt for your agent:
In file @packages/contracts/src/orchestrationV2.ts around line 2666:
Authenticated clients can dispatch `prepared-run.release`, `prepared-run.progress`, and `prepared-run.fail`, allowing them to start or terminate a run before worktree/setup completes. These server-only lifecycle callbacks are currently part of the public `OrchestrationV2Command` schema; move them to `OrchestrationV2InternalCommand` so `dispatchCommand` cannot accept them.
|
Deterministic server evidence supporting the explicit-resume behavior in this PR:
Four synthetic runtime cases on current main pass on These cases reproduce the confusing held-queue behavior, but do not reproduce a missing completion wake after handoff. They support making the paused state and explicit Resume visible as this PR does. They introduce no implicit resumption or product-policy change. The test-only patch is available if useful here. Focused command: Model: GPT-6.1-Sol. Harness: Codex in T3 Code. |
|
Has conflicts. Can you rebase? |
…age limit Resume queue sent the saved text while a queued message was being edited, and the server refused it when a usage limit held the queue. - Web and mobile disable Resume queue while a queued message is edited. - Web routes Resume queue through the composer's Resume when a usage limit holds the queue: it continues the limited run, then the queue. Mobile has no way to continue a limited run, so it disables Resume queue and says why. - The shared queue workflow state reports heldByUsageLimit with the server's own rule (usageLimitBlockedRun). Fork guards until upstream pingdotgg#14617 lands; replace with upstream when it merges.
|
Thanks for the earlier work on resuming a stopped queue. #22790 (96f09c6558) now lets users send the first or any selected held message, releasing the queue without continuing the stopped run. The queue UI has since moved into the timeline; if a separate paused indicator or resume control still matters, a smaller PR focused on that would be welcome. |
Problem
After Stop, the queue is paused (by design, and it stays that way). On web the only way forward was the composer's play button, which continues the stopped run with "Continue where you left off." There was no way to send the paused queued message itself. That mattered most when steering was failing too: the queued message couldn't be steered and couldn't be sent.
Fix
queue.resume).docs/user/composer.mdnow explains pausing on Stop and both resume paths.Independent of, and complementary to, the steer fix and the paused-queue status fix (separate PRs).
Verification
QueuedRunsControl.test.tsx,session-logic.test.ts: 48/48 pass. Web and mobile typecheck clean.No screenshots: the T3 Browser panel was unavailable for capture, and the maintainer waived UI images for this PR. The behavioural tests above are the evidence.
🤖 Generated with Claude Code