feat(mobile): show delegated work and retry actions - #327
leoisadev1 wants to merge 20 commits into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
ff6a743 to
08a2122
Compare
08a2122 to
886efc6
Compare
|
886efc6 to
c7ffe10
Compare
c7ffe10 to
0fab063
Compare
0fab063 to
b4233ad
Compare
b4233ad to
7b1f6f2
Compare
7b1f6f2 to
6576c1f
Compare
| { | ||
| botId: original.childBotId, | ||
| task: original.task, | ||
| expectedResult: original.expectedResult, | ||
| ...(deadline ? { deadline } : {}), | ||
| allowedToolIds: original.access.allowedToolIds, | ||
| memoryScopes: original.access.memoryScopes, | ||
| mcpServerIds: original.access.enabledMcpServerIds, | ||
| sandbox: original.access.sandbox, | ||
| runtimeMode: original.access.runtimeMode, | ||
| approvalCeiling: original.access.approvalCeiling, | ||
| keep: original.keep, | ||
| }, | ||
| { | ||
| trigger: original.trigger, | ||
| retryOfDelegationId: original.delegationId, | ||
| anchorMessageId: original.anchorMessageId, | ||
| }, |
There was a problem hiding this comment.
🔍 Retries omit contextual background
A retry reuses the stored task and access grant, but not the original context, which the record does not retain. The docs disclose this limit; review whether the action needs an in-product warning for context-dependent work.
Was this helpful? React with 👍 or 👎 to provide feedback.
| A delegation that ends while the server is down, or before `routine.run.start` | ||
| records the run's `threadRef`, does not settle the run. Recovering those runs after a restart is | ||
| follow-up work. |
There was a problem hiding this comment.
🔍 Documented gap in scheduled recovery
The new scheduled-work path has no restart reconciliation for a delegation that finishes while the server is down. The internal docs acknowledge this gap; track it before relying on routine status across restarts.
Was this helpful? React with 👍 or 👎 to provide feedback.
…r matrix (LEO-609, LEO-610, LEO-612) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…firm Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…le card actions (LEO-611, LEO-613) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Compare the offered card actions when memoized feed rows decide whether to render. Cover the retry case where a later delegation changes the earlier card without changing its record. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Apply the repository formatter to files changed on stack branch 23. Add an empty changeset for this CI-only formatting fix. GPT-6 Sol via T3 Code Codex harness.
The decider refuses to cancel a run that already ended, and a late cancel from settled bot work leaves that outcome in place. Canceled scheduled work settles its claim as canceled, including when the user cancels the run. A run whose start fails for any reason other than having ended is blocked with that reason and its bot work is canceled. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Scheduled work now gets its own turn id instead of borrowing the chat's current turn, so later replies in that turn no longer pull the card down. Without an anchor message, a scheduled card sits at its creation time rather than the end of the chat. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When a new delegation cannot record its running state or start its child turn, the record now fails with the reason instead of staying open with no watcher, so Try again reappears. The runtime also refuses to retry work that already has a successor, covering two retries accepted before either started. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
plainText now removes single-star and underscore emphasis without touching snake_case words, and turns star or plus list bullets into plain dashes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A remembered chat path without a thread shell no longer enables routine creation that would fail on submit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The chat guide still called mobile work cards read-only after they gained Let it finish, Cancel, and Try again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Settings now keeps the selected channel tab in its store rather than reading the URL hash, so the roster sheet's update-credentials action sets that tab before it opens Settings. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
🔍 Channel recovery lacks user guidance
A channel can now recover in the attempted project after reassignment fails. The repository requires user-visible behavior changes in user documentation; this fallback is undocumented.
(Refers to this code)
Was this helpful? React with 👍 or 👎 to provide feedback.
Part 23 of 29. Stacked on #326.
Mobile chats needed the same visible delegated work and reverse actions as web. This cut adds mobile work cards, result and speaker labels, retry-once and routine delegation actions for LEO-607, LEO-608, LEO-609, LEO-610, LEO-611, LEO-612 and LEO-613. Child work stays on cards instead of appearing as ordinary chats; channel failures keep their reason and assignment.
The cards use upstream #300's cached feed and compare offered actions so an earlier card updates after retry. Web rows retain #299 memoization. A frozen install and all five package typechecks passed with zero errors and warnings. No physical phone was tested.
Screenshots
Captures use fictional fixture data on web at 1440x900.
Failed delegated work now offers a Try again action beside the Mira work card.
Created with Claude Opus 5.5 and GPT-6 Sol in T3 Code.