/ultrareview crashes twice on same PR — "Review crashed before producing findings"

Summary

/ultrareview <PR#> crashed twice in a row on the same target PR, consuming 2 of 3 free reviews with zero findings delivered. Both attempts returned the identical opaque error.

Sessions to investigate

• session_013KSjLYXJiXrU55LsjTZqHq
• session_019eUQ5uYQj2WKg5h2q3rzBw

Both completed (status: completed) but the task-notification payload was:

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

Reproduction

• Target: a GitHub PR
• Diff size: 51 files changed, +2937 / -1126 (might be relevant if there's a payload-size threshold)
• Invocation 1: /ultrareview origin/develop — rejected by the local CLI as not-a-valid-branch (separate, minor UX issue — origin/<branch> should either be accepted or the error should suggest the bare-branch form)
• Invocation 2: /ultrareview 557 — ran in the cloud, crashed
• Invocation 3: /ultrareview 557 again — same crash

Asks

1. Refund the 2 free reviews that produced no output.
2. Surface the underlying error in the task-notification payload (or the tracking URL) instead of "See session logs for details" — users have no way to access session logs from the CLI.
3. Pre-flight the crash condition if it's deterministic (e.g., diff too large, missing base-ref, auth issue). Better to fail fast at invocation time than burn quota on a guaranteed crash.

Environment

• Claude Code CLI
• Branch was active and well-formed (verified by user via gh pr view)
• Tests passing, lint clean — the PR itself wasn't malformed