Claude Code v2.1.192 to v2.1.193 Major Updates

This is Ishikawa from the Cloud Business Division. Claude Code v2.1.193 (released June 25, 2026) has been released. This update focuses on enhanced safety and visibility for auto-mode (automatic execution mode) and bug fixes around background agents. Note that v2.1.192, which precedes this article, is a missing number not published in either npm or CHANGELOG, so we cover the existing v2.1.193 here.

Update Summary

v2.1.193 includes 15 changes. New features include the autoMode.classifyAllShell setting and automatic memory reclamation for background shells, while fixes resolve multiple bugs around backgrounding. Additionally, new log events have been added to OpenTelemetry, and care should be taken as upgrading with existing settings may change behavior.

Notable Updates

• New Feature: Live file path completion has been added to bash mode (!). Additionally, a startup notification now appears when an MCP server requires authentication.
• Bug Fixes: Multiple bugs related to backgrounding have been fixed together. This includes resolving an issue where "phantom sub-agents" were generated that re-executed conversations when the main turn was backgrounded.
• Security: The autoMode.classifyAllShell setting has been added to route all Bash/PowerShell commands through the auto-mode classifier. Additionally, a log event containing model response text has been added to OpenTelemetry (see "Breaking Changes & Deprecations" below).

Update Details

New Features

• Added autoMode.classifyAllShell setting: All Bash/PowerShell commands will now be routed through the auto-mode classifier. Previously, only "arbitrary-code-execution patterns" were subject to classification, but enabling this setting gates all commands.
• Visibility of auto-mode rejection reasons: Auto-mode rejection reasons are now displayed in the transcript, rejection toast, and the recent rejections list in /permissions. You can now trace why something was rejected.
• Live file path completion in bash mode: In bash mode (!), file paths can now be live-completed while typing commands.
• Startup notification for MCP authentication: When an MCP server requires authentication, a startup notification now directs users to /mcp.
• Automatic memory reclamation for background shells: Idle background shell commands are now automatically reaped under memory pressure. This can be disabled with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1.

Improvements

• Background agent launch results: Launch results no longer instruct Claude to "end your response," allowing other tasks to continue while agents are running.
• MCP headersHelper auto re-authentication: When a tool call returns 401/403, headersHelper now automatically re-runs and reconnects.
• Automatic plugin renaming: The marketplace renames map is now automatically followed, updating settings to new names.
• /add-dir message: The message when the specified directory is already the working directory has been improved.

Fixes

• Fixed phantom sub-agent generation when backgrounding the main turn: Resolved an issue where backgrounding the main turn would generate a sub-agent called "general-purpose (resumed)" that re-executed the main conversation.
• Fixed backgrounding being incorrectly cancelled: Resolved an issue where backgrounding (←←) would incorrectly display "N background tasks would be abandoned" and cancel, even though all running tasks were being handed off to a new session.
• Fixed re-prompting of pinned background agents: Resolved an issue where pinned background agents would be prompted again with "Continue from where you left off" on every automatic refresh.
• Additionally, fixes have been made to the display of UI elements that depend on client data (such as /model immediately after /login), and an issue where the agent panel would hide sibling agents when displaying sub-agents.

Breaking Changes & Deprecations

A claude_code.assistant_response log event containing model response text has been added to OpenTelemetry. By default it is redacted, but if OTEL_LOG_ASSISTANT_RESPONSES is not set, it follows the OTEL_LOG_USER_PROMPTS setting.

Therefore, deployments that are already logging prompt content with OTEL_LOG_USER_PROMPTS=1 will begin logging response content as well after upgrading. If you want to keep logging only prompts as before, explicitly set OTEL_LOG_ASSISTANT_RESPONSES=0.

Since logs may contain sensitive information, we recommend verifying behavior before upgrading in environments where OpenTelemetry is enabled.

In Closing

v2.1.193 is an update focused on enhanced safety and visibility for auto-mode and bug fixes around background agents. For those who frequently use backgrounding, multiple experience improvements can be expected. On the other hand, please be aware that in environments running OpenTelemetry, the logging behavior for response text will change upon upgrading. Why not update and give it a try?

References

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

https://code.claude.com/docs/en/changelog