[BUG] Claude Code long running 30+ agents for 18h crashes bash

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?

Summary: After running Claude Code for 18+ hours with 30+ parallel agents, the Bash tool becomes permanently broken. All
commands return exit code 1 with zero stdout/stderr output. Even echo test fails. Only exit 0 "succeeds" (by
immediately terminating). Starting a new Claude Code session does not fix it — a full machine reboot was required last
time.

Reproduction:

1. Run Claude Code with many parallel agents (30+) for an extended period (18+ hours)
2. At some point, Bash tool stops returning output
3. New Claude Code sessions inherit the broken state
4. File-based tools (Read, Glob, Grep, Write) continue working normally

Observed behavior:

• echo hello → Exit code 1, no output
• true → Exit code 1, no output
• /bin/true → Exit code 1, no output
• env -i /bin/bash --noprofile --norc -c 'echo test' → Exit code 1, no output
• exit 0 → Success, no output (only command that "works")
• Write tool → ENOSPC: no space left on device (initially, disk was 100% full at 3.3T/3.6T)
• After freeing disk space (106 GB free, 97%), Bash still broken

Environment:

• Linux 6.14.0-37-generic
• Claude Code CLI Version: 2.1.84
• 30+ concurrent agents spawned over 18 hours
• Disk filled to 100% during session (3.6T drive)

Likely root cause: The Bash tool's underlying shell process or PTY gets corrupted — possibly due to resource exhaustion
(file descriptors, PIDs, PTY allocation) from running many parallel agents. Freeing disk space alone does not recover
it. Previous occurrence required a full system reboot.

Workaround: Reboot the machine, then start a new Claude Code session.

What Should Happen?

It should continue to work even over 18+ hours running claude agent teams.

Error Messages/Logs

See bug

Steps to Reproduce

Reproduction:

1. Run Claude Code with many parallel agents (30+) for an extended period (18+ hours)
2. At some point, Bash tool stops returning output
3. New Claude Code sessions inherit the broken state
4. File-based tools (Read, Glob, Grep, Write) continue working normally

Observed behavior:

• echo hello → Exit code 1, no output
• true → Exit code 1, no output
• /bin/true → Exit code 1, no output
• env -i /bin/bash --noprofile --norc -c 'echo test' → Exit code 1, no output
• exit 0 → Success, no output (only command that "works")
• Write tool → ENOSPC: no space left on device (initially, disk was 100% full at 3.3T/3.6T)
• After freeing disk space (106 GB free, 97%), Bash still broken

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.84

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Xterm

Additional Information

_No response_