Anthropic: Claude Code v2.1.193 — auto-mode classifier for shell commands and OpenTelemetry logging

Anthropic releases Claude Code v2.1.193 with focus on enterprise security and observability

On June 25, 2026, at 21:45 UTC, Anthropic released Claude Code v2.1.193 — an agentic coding CLI that, since v2.1.191, has been evolving toward greater control and visibility over executed actions. While the previous v2.1.191 brought ergonomic improvements such as the /rewind command and a 37% reduction in streaming CPU usage, v2.1.193 focuses on shell execution security and integration with enterprise monitoring infrastructure.

What is autoMode.classifyAllShell and why does it change the game?

The new autoMode.classifyAllShell setting forces Claude Code to route every Bash or PowerShell command — without exception — through the auto-mode classifier before execution. The auto-mode classifier is an internal component that assesses the risk of an action and decides whether to request user approval or whether the command is safe enough for automatic execution.

Previously, some routine commands bypassed this check, which could lead to unpredictable behavior in automated CI/CD pipelines. With classifyAllShell, every shell operation goes through the same decision path, delivering a more consistent and auditable security policy — especially important in team or multi-agent environments.

Denial reasons now visible in the interface

Alongside the classifier, v2.1.193 introduces the display of denial reasons — explanations for why a particular command or action was rejected. These reasons appear in the transcript view, toast notifications, and the permissions panel. Instead of a generic rejection message, the user now sees a concrete reason, which significantly speeds up diagnosis and adjustment of prompts or permission policies.

OpenTelemetry: enterprise observability for model responses

The third addition is a new claude_code.assistant_response log event for OpenTelemetry — an open industry standard for collecting telemetry data (logs, metrics, and traces) used by Datadog, Grafana, Honeycomb, and other enterprise tools. The event records every model response within the OTel system and is redacted by default (content is not sent). It is activated explicitly by setting the variable OTEL_LOG_ASSISTANT_RESPONSES=1.

This change allows IT and security teams to integrate Claude Code into their existing observability infrastructure without additional tooling and monitor agent behavior at the level of individual responses — a step that was unavailable in v2.1.191.

Continuity of development

Version 2.1.193 continues Anthropic’s daily development rhythm: v2.1.191 and v2.1.193 were released one day apart, each with a clearly scoped set of changes. The trend is evident — Claude Code is increasingly becoming an enterprise-grade tool, with emphasis on security control, decision transparency, and integration with industry-standard monitoring systems.