/ultrareview returns empty findings paired with rate-limit error across 3 consecutive runs

/ultrareview returns empty findings paired with rate-limit error across 3 consecutive runs

Summary

/ultrareview returned findings: [] on three consecutive runs across two different scopes today, while the terminal showed "API Error: Server is temporarily limiting requests (not your usage limit) - Rate limited" during each run. This burned all three free runs without producing any actionable review output. The pattern across very different diff sizes (7-commit slice and full-repo 50-commit/22k-line scope) suggests the cloud reviewers failed to start or returned errors that got squashed into an empty findings array.

Environment

• Claude Code CLI (running on Linux, Drupal 11 / Node 20 project)
• Account: yabdikad@gmail.com
• Date: 2026-05-09 (UTC)
• Repo has master as working branch, no main branch (passed explicit base commits to /ultrareview)

Reproduction

Three separate runs, all on the same master branch HEAD = 90abf51c:

Run 1 (free 1 of 3) - narrow scope, May 7-9 slice

• Command: /ultrareview be74f702
• Launch output: "Free ultrareview 1 of 3. Scope: 1710 files changed, 1875 insertions(+), 1395679 deletions(-)"
• Tracking URL: https://claude.ai/code/session_01UhsDC5AzexSidJ29RsQJ5z?from=cli
• task-notification status: completed
• Findings: []
• Concurrent terminal error: API Error: Server is temporarily limiting requests (not your usage limit) - Rate limited

Run 2 (free 2 of 3) - full project from initial commit

• Command: /ultrareview 116ff891
• Launch output: "Free ultrareview 2 of 3. Scope: 1980 files changed, 44441 insertions(+), 1398592 deletions(-)"
• Tracking URL: https://claude.ai/code/session_01He4WqSpHg712gyRRdh3VQa?from=cli
• task-notification status: completed
• Findings: []
• Concurrent terminal error: same rate-limit message

Run 3 (free 3 of 3) - same full-project scope as run 2

• Command: /ultrareview 116ff891
• Launch output: "Free ultrareview 3 of 3. Scope: 1980 files changed, 44441 insertions(+), 1398592 deletions(-)"
• Tracking URL: https://claude.ai/code/session_01PkFQWiF5gWLrJWgjMoRuoD?from=cli
• task-notification status: completed
• Findings: []
• Concurrent terminal error: same rate-limit message

Expected vs actual

Expected: at least some findings on a 22,013-line custom-code diff covering three months of development (Drupal modules, Cloudflare workers, theme, scripts). Or a clear failure message explaining no review was produced.

Actual: silent empty findings array with no surface-level signal that the review failed. The "completed" status on the task-notification implies success, but the rate-limit error in the terminal suggests the cloud reviewers never produced any actual output.

Why I think this is a backend issue, not a clean codebase

• The same project was reviewed locally earlier in the same session with three specialized agents (drupal-reviewer, accessibility-auditor, cloudflare-auditor) which produced a substantive punch list (account-takeover risk in OAuth callback, open-redirect via protocol-relative URL, em-dashes in templates, etc.). Those findings were real and shipped fixes for them.
• /ultrareview run 1 was scoped to include those just-shipped fixes, so the diff being clean there is plausible. But /ultrareview runs 2 and 3 covered the entire 22k-line custom codebase including the older modules the local reviewers did NOT touch. Empty findings on that scope is implausible.
• The "Server is temporarily limiting requests" terminal error appeared concurrent with each run.

Request

1. Treat findings: [] as success only when the cloud reviewers genuinely produced no output, not when the request was rate-limited or the reviewers errored. Surface the underlying error in the task-notification when one occurred.
2. Refund the three free runs since none produced actionable output.
3. Consider rate-limit-aware retry, or at least clearer signaling when a run was degraded.

Happy to provide more detail or repro on demand.