Claude Code 2.1.328: Auto Mode Added for Opus 4.6 and Sonnet 4.6 on Bedrock, Foundry and Other Platforms

Anthropic's changelog shows Claude Code 2.1.328 adds auto mode for Opus 4.6 and Sonnet 4.6 on third-party platforms. This release also changes how hooks, /code-review, the sandbox and self-hosted runners behave.

News date: About 7 min read

Share on LINE
Contents
  1. 1. What does 2.1.328 change for permissions and auto mode?
  2. 2. Which environments and tools are affected?
  3. 3. How does this affect everyday development and administration workflows?
  4. 4. What new settings do self-hosted runners and embedded apps get?

Anthropic's Claude Code changelog lists version 2.1.328 (dated 2026-11-19). This release adds auto mode, a Claude Code permission mode, for Opus 4.6 and Sonnet 4.6 on third-party platforms. It also changes the behavior of hooks (custom handlers that run automatically at specific moments), code review, the sandbox (an isolated environment that limits what commands can do) and self-hosted runners. If you use Claude Code through these cloud platforms, the default permission-prompt flow and the way hooks are handled will differ after you upgrade.

What does 2.1.328 change for permissions and auto mode?

According to Anthropic's changelog, 2.1.328 adds auto mode for Claude Opus 4.6 and Sonnet 4.6 on the following platforms: Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry and the Claude apps gateway. Interactive sessions with no permission mode set start in this mode. On third-party providers, plan mode and Bash prompts for these two models now also use and offer auto mode, the same as for newer models. Setting disableAutoMode turns off both.

In addition, mcp_tool hooks gain "onFailure": "block", working the same way as command and HTTP hooks: if the hook call fails, the action is blocked. The changelog also adds a warning: messages sent by a session that skips permission prompts may be held back and dropped by cloud sessions.

Which environments and tools are affected?

Selected 2.1.328 changes (source: Anthropic changelog)
ItemWhat changedScope (per changelog)
auto modeAdded for Opus 4.6 and Sonnet 4.6; interactive sessions with no permission mode start in it by defaultAmazon Bedrock, Google Cloud Agent Platform, Microsoft Foundry, Claude apps gateway
/code-reviewReports only bugs and CLAUDE.md violations; labels findings Important, Pre-existing or Nit; each summary ends with a suggested fixMost models
Managed PostToolUse hooksNo longer fire for calls denied before they runManaged hooks
Code ModeNo longer runs UserPromptSubmit hooks on script outputCode Mode
SandboxWhen it cannot start, the next sandboxed Bash command is denied with a message that the sandbox is offWhen sandbox.failIfUnavailable is not set
Self-hosted runnerAdds --repository-folder, --hook-label, --metadata and two /metrics metricsSelf-hosted runners

How does this affect everyday development and administration workflows?

For code review, /code-review on most models now reports only bugs and CLAUDE.md violations; for reuse, simplification and efficiency cleanup, use /simplify instead. Each finding is labeled Important, Pre-existing or Nit, and each summary ends with a suggested fix. When there are no problems, it shows "Code review found no issues." /bug now asks Yes or No before sending, so repeatedly pressing Enter no longer submits a report or opens GitHub.

For hooks, Code Mode no longer runs UserPromptSubmit hooks on script output, managed hooks included. To redact that output, use a PostToolUse hook with updatedToolOutput. Managed PostToolUse hooks also no longer fire for calls denied before they run: no logging, no rewriting, no blocking and no stopping. PostToolBatch mostly receives those calls, but does not mark them as denied. Entries in deniedMcpServers of the form "claude.ai plus connector name" now also apply to that connector in cloud and self-hosted runner sessions.

For subagents (agents the main session dispatches to carry out subtasks), when CLAUDE_CODE_FORK_SUBAGENT=0 is set, subagents in claude -p, the Agent SDK, the desktop app and VS Code wait for their own subagents by default. Only for claude -p with that setting, a background subagent and the subagents it waits on are counted together within a single 10-minute wait. For longer chains, raise CLAUDE_CODE_PRINT_BG_WAIT_CEILING_MS.

For the sandbox, when sandbox.failIfUnavailable is not set, the way a sandbox startup failure is reported has changed: the next sandboxed Bash command is denied with a message saying the sandbox is off for this session. In addition, if a PermissionRequest hook rewrites a shell command into one the sandbox excludes, that command now runs outside the sandbox.

What new settings do self-hosted runners and embedded apps get?

A self-hosted runner is a program that provides the execution environment for Claude Code sessions on your own machines, and this release adds several settings for it. With --capacity 1, --repository-folder lets you specify the folder a session uses for a repository, replacing its path under the base folder; this folder cannot be a symlink. claude self-hosted-runner orchestrator gains --hook-label, and claude --environment gains --metadata key=value; both pass labels to the spawn-runner hook. The orchestrator's /metrics gains two metrics: claude_code_self_hosted_orchestrator_session_queue_wait_by_reason_seconds and claude_code_self_hosted_orchestrator_sessions_given_up_total.

Where settings protect Claude's GitHub tools by name, the built-in gh api is now off by default and can be turned on with --use-anthropic-gh true. With --capacity 1, the runner now rejects a session earlier if the path of the pushed repository is a symlink. A warning is also shown at startup when the host's settings.json sets a default permission mode that sessions will ignore. CLAUDE_RUNNER_SESSION_UUID and CLAUDE_CODE_REMOTE_SESSION_UUID are now both set for the session and for its checkout, post-session and spawn-runner hooks.

For apps that embed Claude Code, the changelog says they can turn on the policy plugin's allowManagedModsOnly option, though the administrator's managed settings still take precedence. CLAUDE_CODE_DISABLE_DEFERRED_TOOL_RESUME is also new, letting a host prevent claude -p --resume and --continue from running tool calls that a hook deferred.

Frequently asked questions

Which models and platforms does auto mode now support?

Per Anthropic's changelog, it covers Claude Opus 4.6 and Sonnet 4.6 on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry and the Claude apps gateway.

What if I do not want to use auto mode?

The changelog states that setting disableAutoMode turns off the auto mode that plan mode and Bash prompts now use and offer.

How do I include reuse or efficiency suggestions in a review?

The changelog says /code-review on most models reports only bugs and CLAUDE.md violations; for reuse, simplification and efficiency cleanup, run /simplify.

What happens when the sandbox fails to start?

When sandbox.failIfUnavailable is not set, the next sandboxed Bash command is denied with a message saying the sandbox is off for this session.

Can a self-hosted runner still use gh api?

Per Anthropic's changelog, where settings protect Claude's GitHub tools by name, the built-in gh api is now off by default and can be turned on with --use-anthropic-gh true.

Browse the latest news in this topic

Found this guide helpful? Share it with someone who needs it

Share on LINE

Sources

Compiled from the references and verification dates listed above; fares, schedules, product features, and terms can change, so please confirm the latest details on the relevant official website before travelling, booking, or purchasing.

Lifestyle