You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: Archiving a thread lists the tasks it delegated to linked environments as separate sidebar threads #30265
Run two T3 Code servers, A and B, with linked environments on, and link B to A (t3 pair on B, t3 environment link on A). Add both to the web client.
In a thread on A, ask the agent to run two small tasks on B with delegate_task. Both tasks appear under the parent thread, not as their own rows in the sidebar.
Archive the parent thread on A.
Expected behavior
The two tasks stay under the archived parent and out of the sidebar. A local subagent of the same parent already behaves this way when the parent is archived.
Actual behavior
Both tasks appear as their own rows in the sidebar right after the archive, and they are still there after a reload. Unarchiving the parent puts them back under it.
The sidebar keeps a delegated task under its parent only while the parent is in the live shell snapshot (delegatedUnderParentAtom in packages/client-runtime/src/state/threadShell.ts). The server's live shell leaves archived threads out and sends thread.removed when a thread is archived, so the client drops the parent from the live snapshot. After an archive the parent is gone from the live snapshot, so its tasks read as orphans and get listed on their own.
The current test in threadShell.test.ts asserts this ("Archiving the parent here leaves the task there unarchived, so it is listed on its own"), from #16719. I read it as a way to keep a task reachable when its parent is not available, but archiving the parent is a deliberate action, and local subagents are not treated this way. Please say if listing them on archive is intended.
Impact
Minor bug or occasional failure
Version or commit
main at cefb544af3. Also seen on 0.1.7-nightly.20261103.
Environment
Linux, web client in Chromium, two local dev servers linked. The parent thread ran Claude Haiku 5.5.
Logs or stack traces
None. No errors are logged.
Screenshots, recordings, or supporting files
On main, archiving the parent "Release notes check" makes its two tasks on B ("README version list" and "README one-sentence summary") appear in the sidebar. The last frames are after a reload.
Workaround
Archive each task on the linked environment too, or leave the parent unarchived.
Confirmed on an existing installation on 2026-11-17: T3 Code Desktop/local server 0.1.11-nightly.20261117.4825, macOS arm64 (Darwin 25.6.0), Node v24.22.1, Codex, linked environments.
The user reports that archiving a parent makes its previously hidden tasks appear as standalone sidebar threads. Inspection confirmed an archived, completed parent on the linked environment and 15 unarchived, non-settled tasks on the local environment, each pointing to that parent through delegatedFrom.
The installed build’s delegatedUnderParentAtom (threadShell.ts:718-732) hides these tasks only while the parent is present and unarchived. The existing test at threadShell.test.ts:436-441 explicitly expects tasks to become standalone on archive.
Workaround applied: archived those 15 inactive tasks via t3_thread_organize; a read-only database check confirmed zero unarchived tasks remaining for that parent. No source patch or direct database write. Evidence comes from an already-affected installation, not a fresh UI reproduction.
🤖 Claude Opus 5.5 responding on behalf of ImSynem, with research by Codex (gpt-6.2-sol) in T3 Code
Also seeing this. On Nov 19 our Mac server ran 0.1.11-nightly.20261118.4864, with two worker environments linked to it. In the 0.1.12-nightly.20261120.4994 source, delegatedUnderParentAtom has the same parent check.
Here archiving the parent is routine. Long-running orchestrator threads hand reviews and builds to the workers with delegate_task, and each is archived when a fresh thread replaces it. On Nov 19, 1,186 unsettled tasks on the two workers had delegatedFrom pointing at 20 archived threads on the Mac. That count is from thread records, not rendered rows, and our settings inflated it: auto-settle was off on one worker and set to 7 days on the other. For this workflow, listing the tasks on archive is not what we want. Our workaround is settling finished tasks on the workers.
#30266 would group these under their archived parents. One case from reading its diff: a task waiting on a question or approval then gets no sidebar row. #34237 keeps such tasks listed when the parent's environment is offline.
I tried to reproduce this on the current build and could not.
Commit f9b19a7 (origin/main), web client. I adapted the scenario driver from the evidence on #30266 (two dev servers A and B, Linked environments on, B linked from A with t3 environment link, the browser paired to B, a replay parent on A calling delegate_task twice on B, then archive the parent, reload and unarchive). I uploaded the adapted driver to https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/857f8b9d33774b0b/zz-archived-linked-tasks.e2e.ts. The parent delegated both tasks and B's database holds them unarchived. But the web client never loaded B's threads (environmentThreadsAtom(B) was empty, delegatedUnderParent was empty), so the run 'passed' without testing anything. Cause: B runs behind the dev runner's Vite proxy, so the browser's DPoP proof names B's web port while B checks its backend port. Every POST /api/auth/websocket-ticket/rpc to B returned 401 (url_mismatch, shown by a temporary log in packages/shared/src/dpop.ts). Pairing the browser directly with B's backend port failed with 'Could not pair environment B'. I stopped there after about 75 minutes. Unit-level reproduction: in packages/client-runtime/src/state/threadShell.test.ts, changing the archived-parent assertion to expect no listed rows fails on main: navigation lists 'remote-environment-v2/remote-task' after its parent is archived. Other real-app evidence: juliusmarminge's before captures on #30266 (base b3fdf2b) show both tasks as sidebar rows after archive and after reload; alexito4 confirmed it on an installed 0.1.11 nightly.
Closing as not planned. If it still happens for you, reply with the exact steps and the version you are on and it will be looked at again.
Activity
alexito4 commentedon Nov 17, 2026
Confirmed on an existing installation on 2026-11-17: T3 Code Desktop/local server 0.1.11-nightly.20261117.4825, macOS arm64 (Darwin 25.6.0), Node v24.22.1, Codex, linked environments.
The user reports that archiving a parent makes its previously hidden tasks appear as standalone sidebar threads. Inspection confirmed an archived, completed parent on the linked environment and 15 unarchived, non-settled tasks on the local environment, each pointing to that parent through delegatedFrom.
The installed build’s delegatedUnderParentAtom (threadShell.ts:718-732) hides these tasks only while the parent is present and unarchived. The existing test at threadShell.test.ts:436-441 explicitly expects tasks to become standalone on archive.
Workaround applied: archived those 15 inactive tasks via t3_thread_organize; a read-only database check confirmed zero unarchived tasks remaining for that parent. No source patch or direct database write. Evidence comes from an already-affected installation, not a fresh UI reproduction.
Investigated with Codex (gpt-6.1-sol) in T3 Code.
ImSynem commentedon Nov 20, 2026
Note
🤖 Claude Opus 5.5 responding on behalf of ImSynem, with research by Codex (gpt-6.2-sol) in T3 Code
Also seeing this. On Nov 19 our Mac server ran 0.1.11-nightly.20261118.4864, with two worker environments linked to it. In the 0.1.12-nightly.20261120.4994 source,
delegatedUnderParentAtomhas the same parent check.Here archiving the parent is routine. Long-running orchestrator threads hand reviews and builds to the workers with
delegate_task, and each is archived when a fresh thread replaces it. On Nov 19, 1,186 unsettled tasks on the two workers haddelegatedFrompointing at 20 archived threads on the Mac. That count is from thread records, not rendered rows, and our settings inflated it: auto-settle was off on one worker and set to 7 days on the other. For this workflow, listing the tasks on archive is not what we want. Our workaround is settling finished tasks on the workers.#30266 would group these under their archived parents. One case from reading its diff: a task waiting on a question or approval then gets no sidebar row. #34237 keeps such tasks listed when the parent's environment is offline.
juliusmarminge commentedon Nov 22, 2026
I tried to reproduce this on the current build and could not.
Commit f9b19a7 (origin/main), web client. I adapted the scenario driver from the evidence on #30266 (two dev servers A and B, Linked environments on, B linked from A with
t3 environment link, the browser paired to B, a replay parent on A calling delegate_task twice on B, then archive the parent, reload and unarchive). I uploaded the adapted driver to https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/857f8b9d33774b0b/zz-archived-linked-tasks.e2e.ts. The parent delegated both tasks and B's database holds them unarchived. But the web client never loaded B's threads (environmentThreadsAtom(B) was empty, delegatedUnderParent was empty), so the run 'passed' without testing anything. Cause: B runs behind the dev runner's Vite proxy, so the browser's DPoP proof names B's web port while B checks its backend port. Every POST /api/auth/websocket-ticket/rpc to B returned 401 (url_mismatch, shown by a temporary log in packages/shared/src/dpop.ts). Pairing the browser directly with B's backend port failed with 'Could not pair environment B'. I stopped there after about 75 minutes. Unit-level reproduction: in packages/client-runtime/src/state/threadShell.test.ts, changing the archived-parent assertion to expect no listed rows fails on main: navigation lists 'remote-environment-v2/remote-task' after its parent is archived. Other real-app evidence: juliusmarminge's before captures on #30266 (base b3fdf2b) show both tasks as sidebar rows after archive and after reload; alexito4 confirmed it on an installed 0.1.11 nightly.Closing as not planned. If it still happens for you, reply with the exact steps and the version you are on and it will be looked at again.