/ultrareview crashes silently — opaque "Review crashed" error with no root cause or mitigation

Summary

/ultrareview <PR#> consistently crashes during the cloud review run
and surfaces only an opaque error to the Claude Code CLI:

{"error":"Review crashed before producing findings. See session logs for details."}

The web session shows an equally-uninformative line:

[heartbeat] 06:44:27
<remote-review>{"error":"Review crashed before producing findings. See session logs for details."}</remote-review>

There is no:

• specific root cause (timeout? OOM? tool error? rate limit?)
• file or path that triggered the crash
• suggested mitigation
• way to retry from the failed step

Repro

1. Create a stack of 4 PRs against your own repo.

PR 1: ~6 commits, ~80 files (initial scaffold).
PR 2: 42 commits.
PR 3: 24 commits.
PR 4: 14 commits.

1. Run /ultrareview <N> against each.
2. All four crash with the same opaque error. Three were the free

tier; the fourth was billed as Extra Usage and also crashed —
user paid for nothing.

Expected behavior

• /ultrareview should complete regardless of PR size, or fail with

a clear cause when it cannot.

• Any failure should surface, in the CLI:
• specific root cause ("PR diff exceeded N tokens", "tool X

timed out at file Y", "rate limited by upstream", etc.)

• mitigation suggestion ("split into smaller PRs", "retry in

M minutes", "run /ultrareview --shallow", etc.)

• Billed runs that crash before producing findings should not be

charged, or should be auto-retried once.

Environment

• Claude Code 2.1.119
• Linux 6.17.0-20-generic x86_64
• Repo: private, ~100k LOC
• Session URLs (for log correlation, available to Anthropic):
• session_017tzr7T9oYcnSENzsjNxmtD
• session_01HYSQ37zdsDMMg2DL3U2Yap
• session_01Mh223mfcVKB3mkaebWj7vM
• session_01Gz5DPnqz42hy4JEij4MPsZ