Repository navigation
Let delegate_task run a child in its own project and worktree #18373
Closed
Goshujinsama
started this conversation in
Ideas
Replies: 1 comment
|
Note Grok responding on behalf of Julius. Thanks for the detailed write-up. This is the same proposal as #17415 (optional Related: #16460 (per-child worktrees) and #16719 ( |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Problem
delegate_taskchildren always run in their parent's project, branch and worktree. That fits fanning out slices of one task, but not a change that spans repositories, which is a common layout: an application repo, the repo that builds its runtime image, and an infrastructure repo (Ansible/Terraform/Helm) that deploys it. One logical change touches all three, each with its own branch and PR.Today an agent can get a worktree in another project only through
t3_thread_launch, but that thread is top-level: it has no parent, so its completion never wakes the agent that started it, and there's notask_statusfor it. So an agent coordinating such a change either polls threads or the human does the coordination by watching the sidebar.Related reports from people working this way: #17103 and #9435 (threads in a non-git parent workspace with PRs in other repos), #13390 (threads linked to repos other than their project's), #1453 (multi-repo projects).
Proposal
Let
delegate_tasktake an optionalprojectIdandworkspaceStrategy(the same shapest3_thread_launchalready accepts:worktree/existing_worktree/root).defer_start, worktree creation, setup script, release), and the task is delivered once preparation finishes. A preparation failure ends the task as failed and wakes the parent with the reason.task_statusandtask_cancelare unchanged, so the parent is woken as for any child.existing_worktreemust belong to the target project.No new UI: the child appears in its project in the sidebar and as a sub-agent of its parent.
Status
We're building this in a fork (a T3 Code fork for a multi-machine setup) and plan to run it for a while. Would a maintainer confirm whether this direction and scope fit T3 Code before we propose a PR?
All reactions