/ultrareview crashes deterministically on large PR (108 files / 9400 lines)

Resolved 💬 2 comments Opened May 23, 2026 by DonaldMurillo Closed May 26, 2026

Environment

  • Claude Code CLI
  • Max \$200 plan
  • macOS Darwin 25.3.0

What happened

Ran \/ultrareview 15\ twice against the same PR (https://github.com/DonaldMurillo/gofastr/pull/15). Both runs returned identical failure with no findings:

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

Both runs deducted from the 3-run free tier. So far: 2 of 3 burned with zero output.

Session URLs

  1. https://claude.ai/code/session_01UDDtuEb3TWhNjjJk4URUCj
  2. https://claude.ai/code/session_01J7DL21XRHT4rpiExLwCnGR

PR under review

https://github.com/DonaldMurillo/gofastr/pull/15

  • 108 files changed
  • ~9,400 line additions
  • 5 commits, all clean per local \scripts/test-all.sh\
  • Repository is public

Hypothesis

Repeat crash with identical error string suggests a deterministic ceiling (payload size or analysis timeout) rather than a transient failure. PR is on the large side; smaller PRs in the same repo have run ultrareview successfully in the past.

Expected

Either: review completes with findings (preferred), OR the launch is rejected up-front with a clear "PR exceeds size limit" message AND the run is not deducted from the free tier.

Actual

Run starts, crashes silently mid-analysis, charges the credit, leaves only "see session logs" — which the user cannot access from CLI.

Suggested fixes

  1. Pre-flight size check that refuses to start (or warns + asks for confirm) above whatever limit triggers the crash.
  2. Don't deduct credits for runs that crash before producing any findings.
  3. Surface a slice of the session log in the CLI error so users can self-triage.

Reproducer

Open any PR in the 100+ file / ~10k line range against a public repo, then \/ultrareview <PR#>\.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗