claude-code - 💡(How to fix) Fix ultrareview crashed before producing findings + rate limit error — 2nd occurrence same account, all 3/3 free quota now consumed [1 comments, 2 participants]

Summary

Second /ultrareview crash on the same account today, this time on a different repository (monorepo with proper SessionStart hook for dependency install). All 3/3 free runs now consumed with 0 findings produced.

Repro

What happened

CLI launched the cloud session successfully. After the usual wait the task completed with this exact payload:

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

Concurrently the CLI surfaced:

API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

(Status page https://status.claude.com showed all systems operational at the time, no public incident posted.)

What I had done to avoid the known crash bug

Read the docs and prior issues (#55031, #54871, #53455, #54812, #49905) before launching. Specifically prepared:

1. PR-mode review (not branch bundle) — clone-from-GitHub path is more reliable than bundle upload per docs

2. SessionStart hook committed to repo at .claude/settings.json:

   {
     "hooks": {
       "SessionStart": [{
         "matcher": "startup|resume",
         "hooks": [{
           "type": "command",
           "command": "if [ \"$CLAUDE_CODE_REMOTE\" = \"true\" ]; then pnpm install --frozen-lockfile; fi"
         }]
       }]
     }
   }

3. Repo state clean: pnpm type-check 11/11 PASS, pnpm test 85 tests PASS, pnpm build PASS locally before push

4. Constructed a dedicated review base branch to keep PR scope tight (excluded data/docs/.autoflow noise)

→ Setup-script-missing is not the cause here. The crash is in the review pipeline itself.

Same-account, same-day pattern

Earlier today I had the exact same crash on a different repo (dhscfh/stgfolio PR #4), already documented:

• Standalone issue: #56202
• Cross-comment on the master tracking issue: #55031 (first reply)

That crash also consumed a free run. Plus this one. Plus an earlier one. = 3/3 free runs consumed today, 0 findings produced, ~30 minutes of cloud sandbox time wasted.

The free runs expire today (May 5, 2026) per the docs, so I have no recourse to retry — these crashes effectively converted my entire trial allotment into nothing.

Asks

1. Refund all 3 free runs to my account — none of them produced any output. Per the docs "a run counts once the remote session starts," but there are now ~9 open issues (#55031, #54871, #53455, #54812, #54516, #53301, #52780, #49905, #56202) where users have hit this same crash class and asked the same. The pattern is too consistent to be edge-case.
2. Investigate the rate-limit + crash co-occurrence — the simultaneous "Server is temporarily limiting requests" + "Review crashed" suggests the cloud sandbox provisioner is failing under load and surfacing a generic crash to the user, not a real review-pipeline bug. Worth distinguishing in telemetry.
3. Make crash failures NOT consume quota — if the review never produced findings, charging it as a successful run is asymmetrical. This has been requested across most of the linked issues.

Environment

• Claude Code v2.1.x (latest stable as of 2026-05-05)
• macOS 25.4.0 (Darwin)
• Pro subscriber
• No proxy / VPN / IP allowlist on the org

Related

• #55031 (master tracking, also commented there from this account)
• #56202 (standalone issue I opened earlier today for the stgfolio crash)
• #54871, #53455, #54812, #54516, #53301, #52780, #49905 (similar reports)