Claude Code v2.1.193: Auto Mode Gets Granular Shell Controls and OpenTelemetry Logs Model Responses

Claude Code v2.1.193 landed on June 25 at 21:45 UTC. Most of the changes are in auto mode, observability, and background agent handling.

Shell Classification Gets Wider

Auto mode currently routes commands through a classifier when they look like arbitrary code execution patterns. The new autoMode.classifyAllShell setting removes that scoping: when enabled, every Bash and PowerShell command goes through the classifier, not just the ones that match specific ACE patterns.

This gives you more consistent and predictable auto-mode behavior. If you’re running Claude Code in automated pipelines where shell command approval needs to be deterministic, this is the setting to look at.

Auto-mode denial reasons are also more visible now. They appear in the transcript, in the denial toast notification, and in the /permissions recent denials tab. Previously you’d see that something was denied without much context on why.

OpenTelemetry: Model Response Logging

A new claude_code.assistant_response event has been added to the OpenTelemetry output. It contains the model’s response text, but it’s redacted by default. Set OTEL_LOG_ASSISTANT_RESPONSES=1 to enable it. The setting follows OTEL_LOG_USER_PROMPTS behavior when unset, so you get consistent treatment across both.

This fills an observability gap for teams running Claude Code at scale. You could already log user messages and tool calls, but not what the model was actually saying back.

Live Path Autocomplete in Bash Mode

Bash mode (triggered with !) now supports live file path autocomplete. Small improvement, but it’s the kind of thing you notice daily once it’s there.

MCP Auth Notices

Claude Code now shows a startup notice when an MCP server requires authentication, with a pointer to the /mcp menu. Separately, the headersHelper auth flow now automatically re-runs and reconnects when a tool call returns a 401 or 403.

Background shell processes get a new memory-pressure reaping mechanism. Idle shell commands that have been sitting in the background are now reaped to reduce memory bloat in long sessions. You can disable this with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1.

Background Agent Bug Fixes

Several background agent issues are fixed in this release:

• Backgrounding a session with running tasks no longer shows a spurious “N background tasks would be abandoned” warning when all tasks are carrying over.
• Pinned background agents no longer get re-prompted with “Continue from where you left off” after every auto-update.
• Sending the main turn to the background no longer spawns a phantom “general-purpose (resumed)” subagent that re-runs the conversation.
• The agent panel was hiding sibling agents when you opened a subagent view. That’s fixed too.
• /model and other client-data-gated UI no longer shows stale or empty state immediately after /login.

---

Source: Claude Code v2.1.193 on GitHub