[Bug] Subagent completion race conditions cause orchestrator hallucinations of integrated results

[kenxle]

Bug Description
Whenever I spawn a lot of subagents, I get this phenomenon where The orchestrating agent will say that it has analyzed their output. And then it will get a notification a few seconds later that the agent has completed its work. it will say oh I've already integrated the work from that agent you can ignore this late returning call, but there's no evidence anywhere of what the agent was doing, whether the agent had actually completed its task when the orchestrator went and got information from it. It's completely opaque as to whether the agent is just hallucinating that the subagent was complete. And so it's quite confusing. to see and it happens consistently over and over again

Environment Info

• Platform: darwin
• Terminal: ghostty
• Version: 2.1.34
• Feedback ID: b9f65b4f-86d3-4154-bbb0-5915a948cbe3

Errors

[{"error":"Error: {\"message\":\"Failed to export 73 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\",\"originalLine\":\"1861\",\"originalColumn\":\"1409\",\"line\":\"1861\",\"column\":\"1409\",\"sourceURL\":\"/$bunfs/root/claude\",\"stack\":\"Error: Failed to export 73 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\\n    at doExport (/$bunfs/root/claude:1861:1409)\\n    at processTicksAndRejections (native:7:39)\",\"name\":\"Error\"}\n    at error (/$bunfs/root/claude:1904:25028)\n    at <anonymous> (/$bunfs/root/claude:1856:21308)\n    at gy7 (/$bunfs/root/claude:1856:21852)\n    at <anonymous> (/$bunfs/root/claude:1857:17366)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:02.748Z"},{"error":"Error: 1P event logging: 8 events failed to export (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\n    at queueFailedEvents (/$bunfs/root/claude:1861:2350)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:08.061Z"},{"error":"Error: {\"message\":\"Failed to export 8 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\",\"originalLine\":\"1861\",\"originalColumn\":\"1409\",\"line\":\"1861\",\"column\":\"1409\",\"sourceURL\":\"/$bunfs/root/claude\",\"stack\":\"Error: Failed to export 8 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\\n    at doExport (/$bunfs/root/claude:1861:1409)\\n    at processTicksAndRejections (native:7:39)\",\"name\":\"Error\"}\n    at error (/$bunfs/root/claude:1904:25028)\n    at <anonymous> (/$bunfs/root/claude:1856:21308)\n    at gy7 (/$bunfs/root/claude:1856:21852)\n    at <anonymous> (/$bunfs/root/claude:1857:17366)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:08.061Z"},{"error":"Error: 1P event logging: 61 events failed to export (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\n    at queueFailedEvents (/$bunfs/root/claude:1861:2350)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:13.951Z"},{"error":"Error: {\"message\":\"Failed to export 61 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\",\"originalLine\":\"1861\",\"originalColumn\":\"1409\",\"line\":\"1861\",\"column\":\"1409\",\"sourceURL\":\"/$bunfs/root/claude\",\"stack\":\"Error: Failed to export 61 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\\n    at doExport (/$bunfs/root/claude:1861:1409)\\n    at processTicksAndRejections (native:7:39)\",\"name\":\"Error\"}\n    at error (/$bunfs/root/claude:1904:25028)\n    at <anonymous> (/$bunfs/root/claude:1856:21308)\n    at gy7 (/$bunfs/root/claude:1856:21852)\n    at <anonymous> (/$bunfs/root/claude:1857:17366)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:13.951Z"},{"error":"Error: 1P event logging: 28 events failed to export (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\n    at queueFailedEvents (/$bunfs/root/claude:1861:2350)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-03T12:46:20.316Z"},{"error":"Error: {\"message\":\"Failed to export 28 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\",\"originalLine\":\"1861\",\"originalColumn\":\"1409\",\"line\":\"1861\",\"column\":\"1409\",\"sourceURL\":\"/$bunfs/root/claude\",\"stack\":\"Error: Failed to export 28 events (status=429, code=ERR_BAD_REQUEST, Request failed with status code 429)\\n    at doExport (/$bunfs/root/claude:1861:1409)\\n    at processTicksAndRejections (native:7:39)\",\"name\":\"Error\"}\n    at error (/…

Note: Content was truncated.

[github-actions]

Found 3 possible duplicate issues:

1. Background Task Notifications Not Delivered When Multiple Agents Complete Simultaneously (v2.1.19) #20754
2. [BUG] Model emits end_turn while parallel background Task agents still running, returning to prompt prematurely #23806
3. [BUG] Task tool: Main thread ends prematurely while subagents run in background (VSCode/Cursor) #23909

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[kenxle]

This appears to be different from the 3 issues listed. This is particularly about the main agent reporting that sub-agents have completed, and taking their "work" to integrate it, then receiving "late returning notifications" from the sub-agents which it has to report on to the user.

This issue primarily happens when I am doing research reports, where the main agent will spawn sub-agents for different components of the research, and the main agent is waiting for their research to come back to integrate into a full research report including the info from all the agents. Frequently there will be notes from the main agents saying "Oh that was a late returning notification, we can ignore that because I've already integrated their research into the report." But of course I don't know if it somehow accessed the sub-agents work in a partial state, and then when the sub-agent reported completion, ignored the rest of the work that was done by the sub-agent. It may very well be the case that it's doing things correctly and the issue really is late returning agents, but I have no way of knowing if it's hallucinating that or if it's correct. And in the case of these research reports, they could be skipping valuable information that I want included.

[github-actions]

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

[github-actions]

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.