[BUG] Repeated Bun panic "switch on corrupt value" on Windows 11 x64 during long sessions (v2.1.50 / Bun 1.3.10)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Description

Claude Code crashes with panic(main thread): switch on corrupt value during extended sessions. This has happened twice on the same machine, both times during sessions involving multiple subagent spawns (worktree orchestration with 5-7 parallel agents). The crash is non-deterministic — it occurs at different points in the session.

This appears related to #26763, #26590, #26531, #27086, and #27095. Filing separately to add another data point with full crash output and session context.

Environment

| Detail | Value |
|--------|-------|
| Claude Code version | 2.1.50 |
| Bun version | 1.3.10 (1423d3c8) |
| OS | Windows 11 Pro 10.0.26200 (x64) |
| CPU | x64 baseline (sse42, avx, avx2) |
| RAM | 33.78GB (reported by Bun — actual is 32GB) |
| Shell | Git Bash |

Session context at time of crash

| Metric | Value |
|--------|-------|
| Elapsed time | ~37 minutes (2,255,367ms) |
| Peak RSS | 1.56GB |
| Page faults | 7,700,941 |
| Session activity | Orchestrating 5-7 parallel worktree agents via Task tool |

Reproduction

Not deterministic, but both crashes shared these conditions:

1. Windows 11 x64, Claude Code 2.1.50
2. Session running 20-40 minutes
3. Heavy use of Task tool with isolation: "worktree" spawning multiple parallel subagents (5-7 concurrent)
4. Each subagent performing file reads, writes, git operations, and bash commands
5. High I/O churn — creating branches, editing files across worktrees, running linters

The crash has not occurred during simpler sessions (single-agent, no worktrees, shorter duration).

Expected behaviour

Session should remain stable regardless of duration or number of subagents.

Actual behaviour

Bun panics with switch on corrupt value after ~20-40 minutes of heavy multi-agent use. All unsaved session state is lost. Work in progress across worktrees is left in an inconsistent state (some branches created, some not merged).

Impact

• Session state is lost on crash (no auto-recovery)
• Worktree branches may be left in partial states requiring manual cleanup
• Forces users to limit session length and agent concurrency as a workaround
• Particularly impacts orchestration workflows that rely on parallel agents

Notes

• RAM is not exhausted at crash time (0.84GB RSS, 1.56GB peak, 33.78GB machine)
• Bun reports machine RAM as 33.78GB when actual is 32GB — possible GC threshold miscalculation (noted in oven-sh/bun#27099)
• The root cause appears to be in Bun's string deinitialization during event loop cleanup (string.zig:885 per oven-sh/bun#27099 stack trace)
• Since Anthropic now owns Bun, this may be fixable in a future Claude Code release without waiting for an upstream Bun release

Potential Similar issues

• anthropics/claude-code#26763 — switch on corrupt value on Windows 10
• anthropics/claude-code#26590 — Recurring Bun crash on Windows
• anthropics/claude-code#26531 — Bun panic during Bash tool execution (v2.1.45)
• anthropics/claude-code#27086 — Bun panic during Explore agents on Windows 11
• anthropics/claude-code#27095 — BUN crash switch on corrupt value
• oven-sh/bun#27138 — switch on corrupt value, Claude Code on Windows x64
• oven-sh/bun#27099 — switch on corrupt value in string.zig (includes stack trace)

What Should Happen?

Runs without any errors.

Error Messages/Logs

### Crash output
Bun v1.3.10 (1423d3c8) Windows x64 (baseline)
Windows v.win11_dt
CPU: sse42 avx avx2
Args: "claude"
Features: Bun.stderr(2) Bun.stdin(2) Bun.stdout(2) abort_signal(286) fetch(1250) jsc spawn(310) standalone_executable process_dlopen(2) yaml_parse(35)
Builtins: "bun:main" "node:assert" "node:async_hooks" "node:buffer" "node:child_process" "node:constants" "node:crypto" "node:events" "node:fs" "node:fs/promises" "node:http" "node:https" "node:module" "node:net" "node:os" "node:path" "node:path/posix" "node:path/win32" "node:perf_hooks" "node:process" "node:stream" "node:timers/promises" "node:tls" "node:tty" "node:url" "node:util" "node:zlib" "node:worker_threads" "undici" "ws"
Elapsed: 2255367ms | User: 223093ms | Sys: 53968ms
RSS: 0.84GB | Peak: 1.56GB | Commit: 1.77GB | Faults: 7700941 | Machine: 33.78GB
panic(main thread): switch on corrupt value
oh no: Bun has crashed. This indicates a bug in Bun, not your code.

Steps to Reproduce

Run Claude with worktree agents > 5 with run time longer than 30 mins.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v2.1.50

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

_No response_