[BUG] /ultrareview completes cloud-side but CLI reports 'failed' and /tasks does not list the run

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet (closest neighbours #50029 / #53599 / #49905 describe different failure modes — empty findings, explicit crash, real timeout)
• [x] This is a single bug report
• [x] I am using the latest version of Claude Code

Summary

/ultrareview completed successfully on the cloud side (web UI shows completion with N findings), but the CLI received a false-negative failed task-notification, and /tasks does not list the completed run. Findings are unreachable from the CLI session that initiated the review — the only way to view them is to manually navigate to the cloud session URL in a browser. The run was billed.

Per the docs ("Use /tasks to see running and completed reviews"), a completed cloud-side review should appear in /tasks. It doesn't.

Reproduction

1. Run /ultrareview against a branch with a substantial diff (in my case: 78 files, +4260 / -748 LOC).
2. CLI confirms launch and prints the cloud session tracking URL.
3. Wait for completion. The cloud-side claude.ai/code/session_<id> page shows the review completed with 7 findings.
4. CLI receives a task-notification with <status>failed</status> and <summary>Remote review failed: remote session exceeded 30 minutes</summary>.
5. Run /tasks. The run is not listed — neither as completed nor as failed.
6. There is no in-CLI path to retrieve the findings.

This happened on two separate ultrareview runs (a PR-scoped one and a local-diffs one), both billed, both completing cloud-side, neither surfacing to /tasks or to the CLI conversation.

Expected behaviour

• The CLI's task-notification status reflects the cloud-side outcome (completed, not failed).
• /tasks lists the completed review with a detail view that exposes the findings.
• Findings appear as a session notification with file:line locations, per the documented behaviour.

Actual behaviour

• Notification status: failed (with "remote session exceeded 30 minutes" as the reason — but the cloud session in fact completed before that 30-minute mark).
• /tasks: review absent.
• Findings: not visible from the CLI; only accessible via the authenticated cloud session URL the user has to manually keep track of.
• Billing: still consumed against the org's usage.

Evidence

• CLI launch echo from one of the runs:

Ultrareview launched for dois/aoh-6740 → develop (~5-10 min, runs in the cloud). Track: https://claude.ai/code/session_01Jt4zD4Yi3UsBCx3h9YpiSH

• Cloud-side completed review for the second run is at session ID session_01WjgLgfkLDwEKtfTa82BGtG. Notably, the cloud session ID where the review actually completed differs from the session ID returned at launch time — that's itself surprising and may be related to the notification-state mismatch.
• Failed task-notification payload received in the CLI:

``
<task-id>rznnwyw01</task-id>
<task-type>remote_agent</task-type>
<status>failed</status>
<summary>Remote review failed: remote session exceeded 30 minutes</summary>
``

Environment

• Claude Code: 2.1.119
• Node: v24.14.1
• macOS: 26.3.1 (a)
• Shell: zsh
• Platform: darwin

Impact

Paid (twice) for ultrareview runs whose findings I cannot access from the CLI. The findings exist only in cloud session URLs the user has to remember and manually navigate to, with no thread-back into the CLI conversation that initiated the review. This breaks the documented /tasks → review detail → findings workflow.