Bun segfault on image attachment crashes session (Claude Code 2.1.142, Windows)
Resolved 💬 4 comments Opened May 21, 2026 by EF0216 Closed Jun 20, 2026
## Summary
Attaching an image to a message reliably crashes the local session on Windows. The Bun runtime panics with a segmentation fault. No image attachment ever succeeds; the same operation reportedly works on other machines.
Environment
- Claude Code: 2.1.142
- Desktop app: Claude 1.8089.1.0 (Microsoft Store install)
- Runtime: Bun v1.3.14 (d2989145) Windows x64 (baseline)
- OS: Windows 10 Pro 10.0.19045
- CPU features: sse42, avx, avx2
- Machine RAM: 16.79 GB
Steps to reproduce
- Start a local session in Claude Code 2.1.142.
- Attach an image to the next user message (text content can be empty).
- Send.
Expected
Image is processed and included in the message.
Actual
The Claude Code CLI process (claude.exe, Bun-based) crashes with a segfault immediately after sendMessage is invoked with imageCount=1. The session enters an unhealthy cycle (reason=no_response) and the message never reaches the model. Reproduces 3+ times back-to-back in the log.
Log excerpt (%APPDATA%\Claude\logs\main.log)
LocalSessions.sendMessage: ... messageLength=0, imageCount=1, toolStates=0
panic(main thread): Segmentation fault at address 0x0
oh no: Bun has crashed. This indicates a bug in Bun, not your code.
https://bun.report/1.3.14/e_1d298914mgkgkEuhogCCcKERNELBASE.dll6tmgBCSastMC.dllk6nMCSastMC.dllqt3UCSastMC.dll4i5U0wm40B__kkkm2CA2AA
[CCD CycleHealth] unhealthy cycle for local_<id> (Xs, hadFirstResponse=false, reason=no_response)
Crash modules in trace: KERNELBASE.dll, MC.dll (Windows locale / message-compiler boundary).
Notes
- Started after a recent auto-update to 2.1.142. Prior versions handled image attachment correctly on the same machine.
- Working directory contained a non-ASCII path, but the issue also reproduces from ASCII-only working directories (per user), so the path is unlikely to be the trigger.
- The bundled Bun is 1.3.14 baseline — pinning a newer or "modern" Bun build may be worth testing on the maintainer side.
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗