autoMode.classifyAllShell and Automatic Memory Reclamation for Background Shells
Claude Code v2.1.193 landed 15 changes, and the two most consequential are both about auto mode. The new autoMode.classifyAllShell setting extends auto mode's command-risk classification to every shell command it runs, not just a pre-defined subset — closing a gap where commands outside the classifier's known list previously fell through to a less scrutinised default path. Separately, background shells now get automatic memory reclamation: long-running or forgotten background processes launched during a session no longer silently accumulate memory until something else notices.
// settings.json
{
"autoMode": {
"classifyAllShell": true
}
}
Why the classification gap mattered
Auto mode's safety model depends on correctly categorising what a command will do before running it unattended. A command that fell outside the classifier's known patterns previously got treated more permissively than it should have — not necessarily unsafe, but inconsistent with the rest of the safety model. Making classification apply universally removes that inconsistency rather than patching a specific exploit.
If you use Claude Code for extended unattended runs — overnight test suites, long refactors, background monitoring scripts — both changes in this release are worth adopting immediately: classifyAllShell tightens the safety net, and automatic background-shell cleanup prevents the slow memory creep that used to require a manual restart on very long sessions.