[Bug] Cache eviction triggered during active nested subagent execution

Status Open
Reported on v2.1.237
Maintainer reply None cached
Activity 0 comments · opened Aug 20, 2026

Bug Description
a nested subagent was long running (and was still running) and the main thread cached (I've not seen that happen before) it said to clear to save 250k tokens - however it was my main orchestration thread mid-orchestration. I had to resume it and spend the 250k tokens. I've only ever seen it cache before after a subagent truly died out without notification or other truly idle states.

Environment Info

  • Platform: win32
  • Terminal: null
  • Version: 2.1.237
  • Feedback ID: 4fbb13e6-36c1-4f1a-af1e-412589696f82

Errors

[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=214 messages=213 range=[171,214))\n    at lGE (B:/~BUN/root/cli:23198:33898)\n    at i8h (B:/~BUN/root/cli:23198:26870)\n    at vi (B:/~BUN/root/cli:2140:21439)\n    at Lm (B:/~BUN/root/cli:2140:40550)\n    at Iy (B:/~BUN/root/cli:2140:51489)\n    at ii (B:/~BUN/root/cli:2140:89194)\n    at co (B:/~BUN/root/cli:2140:88149)\n    at $t (B:/~BUN/root/cli:2140:87970)\n    at kUe (B:/~BUN/root/cli:2140:84225)\n    at gt (B:/~BUN/root/cli:2140:6694)","timestamp":"2026-08-20T23:32:28.748Z"}]

View original on GitHub ↗