Enhanced Linux Sandbox Security with Seccomp Filtering

What: Linux sandboxing now blocks Unix domain socket creation using Berkeley Packet Filter (BPF) seccomp filters, preventing sandboxed processes from bypassing network restrictions via local sockets.

How it works: When network sandboxing is enabled on Linux, Claude Code automatically applies pre-compiled seccomp filters to block socket(AF_UNIX, ...) syscalls.

Details:

• Architecture-specific (x64/arm64) pre-compiled BPF filters in vendor/seccomp/ directories
• apply-seccomp binary applies the filters to child processes
• Functions: Q$Q() detects CPU architecture (x64/arm64), Ox1() locates BPF filter files, NiA() finds the apply-seccomp binary
• I$Q() retrieves the appropriate seccomp filter for the platform
• Automatic cleanup via dP6() and Rx1() on process exit
• 32-bit x86 (ia32) explicitly not supported due to socketcall() syscall complexity (line 1507)
• Fallback: allowAllUnixSockets configuration option to disable seccomp filtering when needed (line 1693)
• Evidence: Q$Q() at line ~1535, Ox1() for BPF filter location, NiA() for apply-seccomp binary discovery, seccomp integration at lines 1686-1693