Claude Code v2.1.171 to v2.1.172 Major Updates

Claude Code v2.1.171 to v2.1.172 Major Updates

Claude Code v2.1.172 has been released. In addition to multi-level hierarchization of sub-agents and improved Amazon Bedrock region resolution, numerous bugs related to 1M context and model selection have been fixed. Here are the notable update highlights.
2026.06.11

This page has been translated by machine translation. View original

This is Ishikawa from the Cloud Business Division. Claude Code v2.1.172 (released June 10, 2026) has been released. In addition to new features such as multi-level sub-agent nesting and improved Amazon Bedrock region resolution, numerous bug fixes have been made around 1M context and model selection.

Note that of the specified version range v2.1.171–v2.1.172, v2.1.171 was not published to npm (updated directly from v2.1.170 to v2.1.172), so the effective target is v2.1.172 alone. This article also covers v2.1.172.

Update Summary

v2.1.172 includes 30 changes. In addition to new features and performance improvements, the release is centered around numerous bug fixes. Fixes related to session stability (1M context, image processing), the model selection UI, and permission rule behavior are particularly notable. For Amazon Bedrock users, the key practical point is that region resolution has been aligned with the same priority order as the AWS SDK. There are no breaking changes or deprecations.

Notable Updates

  • New feature: Sub-agents can now launch their own sub-agents (up to 5 levels deep).
  • Bug fix: An issue where sessions using 1M context without usage credits would get permanently stuck has been fixed, and sessions will now return within the standard context limit via auto-compact.
  • Security: Fixed an issue where background agents on pre-warmed workers could load configuration (.mcp.json approvals/trust) from a different directory, as well as issues where wildcard domain rules in WebFetch and file permission rules containing wildcards mid-pattern were not functioning correctly.

Update Details

New Features

  • Multi-level sub-agent nesting: Sub-agents can now launch their own sub-agents, allowing nesting up to 5 levels deep. This makes it easier to decompose complex tasks into deeper hierarchies for processing.
  • Added plugin search bar to /plugin: A search bar has been added when browsing marketplace plugins, making it easier to find the desired plugin.
  • Added model attribute to OTEL metrics: A model attribute has been added to the claude_code.lines_of_code.count metric, enabling aggregation by model.

Improvements

  • Amazon Bedrock region resolution: When AWS_REGION is not set, the region will now be read from the ~/.aws configuration file. This aligns with AWS SDK priority behavior, and the source of the region can be confirmed via /status. If you are using Claude Code via Bedrock, standard AWS profile configuration will now be reflected without relying on environment variables.
  • Improved usage policy rejection message: When rejected due to a Usage Policy in non-interactive mode, the message now suggests starting a new session or changing the model.
  • ultra option display in /code-review: The ultra option continues to be displayed even when not signed in to claude.ai, with an explanation that a claude.ai account is required for cloud review.
  • Simplified remote control display: The footer indicator has been shortened to "/rc active" and will be hidden in narrow terminals.
  • Suppressed /loop guidance in remote sessions: In remote sessions where pending loops do not keep the container alive, /loop guidance is no longer displayed.

Performance

  • Performance improvement for long conversations: Processing in long conversations has been lightened by removing redundant message normalization and avoiding full message history conversion when tool usage state has not changed during streaming.
  • Reduced CPU usage when idle: The /goal status chip no longer redraws the terminal at 5Hz when idle, and UI redraws during parallel sub-agent execution have also been reduced.
  • Improved tool loading for Claude in Chrome: Browser tools are now loaded with a single batch call instead of one call per tool.

Security

  • Fixed background agent configuration loading: Fixed an issue where background agents dispatched to pre-warmed workers could load project configuration (.mcp.json approvals/trust) from a different directory. This is a particularly important fix for environments handling multiple projects, as there was a risk of unintended directory approval/trust settings being applied.
  • Fixed wildcard handling in permission rules: Fixed an issue where wildcard domain rules such as WebFetch(domain:*.example.com) were not matching subdomains in allow, deny, or ask positions, and an issue where file permission rules containing wildcards mid-pattern (e.g., Read(secrets-*/config.json)) were being rejected at startup. There were also cases where deny rules were not working as intended, so if you use permission settings, it is worth checking the behavior.

Fixes

A selection of major fixes:

  • Fixed session stuck with 1M context: Fixed an issue where sessions using 1M context without usage credits would get permanently stuck. Sessions will now be automatically compacted and return within the standard context limit.
  • Fixed repeated errors with multiple images: Fixed an issue where the error "an image in the conversation could not be processed and was removed" was repeatedly displayed when a conversation contained multiple images.
  • Fixed lingering spinner in agent view: Fixed an issue where the busy spinner kept spinning in Working state for up to 30 seconds after the worker responded.
  • Fixed inconsistency in Bedrock model selection: Fixed an issue in Bedrock where the /model picker displayed models not offered by the provider, and selecting them would silently switch the session's model with selection markers lighting up on multiple lines.
  • Fixed team shared memory not referenced in remote: Fixed an issue where memory calls could not find mounted team-shared memory stores (CLAUDE_MEMORY_STORES) in remote sessions.
  • Fixed incorrect rejection in workflow validation: Fixed an issue where workflow scripts that merely mentioned Date.now() / Math.random() in prompt strings or comments were being rejected by validation.
  • [VSCode] Fixed PowerShell tool display: Fixed an issue where PowerShell tool calls were displayed as raw JSON instead of the correct command display and permission dialog, and removed ANSI escape codes from displayed shell output.
  • In addition, numerous minor bugs have been fixed, including background session attachment (EAUTH), availableModels restrictions/allowlists, 1M context in opusplan plan mode, /model candidate display, plugin browser cursor/tab restoration, and Windows console mouse tracking.

Closing

v2.1.172 is a release that includes the functional expansion of multi-level sub-agent nesting, while overall being centered around stability, model selection, and permission-related bug fixes. There are no breaking changes or deprecations. In particular, the fix for sessions getting stuck with 1M context and the improvements to Amazon Bedrock region resolution and model selection are significant points for those using these features in affected environments. Why not update and give it a try?

Recent Updates

https://dev.classmethod.jp/articles/20260602-cc-updates-v2-1-160/

https://dev.classmethod.jp/articles/20260603-cc-updates-v2-1-161/

https://dev.classmethod.jp/articles/20260603-cc-updates-v2-1-165/

https://dev.classmethod.jp/articles/20260608-cc-updates-v2-1-168/

https://dev.classmethod.jp/articles/20260609-cc-updates-v2-1-169/

https://dev.classmethod.jp/articles/20260609-cc-updates-v2-1-170/

References

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

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


Claudeならクラスメソッドにお任せください

クラスメソッドは、Anthropic社とリセラー契約を締結しています。各種製品ガイドから、業種別の活用法、フェーズごとのお悩み解決などサービス支援ページにまとめております。まずはご覧いただき、お気軽にご相談ください。

サービス詳細を見る

Share this article

AWSのお困り事はクラスメソッドへ