Claude Code 2.1.328 adds auto mode for Opus 4.6 and Sonnet 4.6 across several providers, alongside a wide-ranging set of fixes for coding sessions, tools and self-hosted runners. For developers, the release combines a new way to run sessions with practical improvements to how they behave when things go wrong.
Anthropic Watch analysis
What happened
Anthropic’s Claude Code 2.1.328 release notes list auto mode for Opus 4.6 and Sonnet 4.6 on Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry and the Claude apps gateway. Interactive sessions with no permission mode now start in auto mode.
The release also adds an onFailure: block option for MCP tool hooks, so a failed hook can stop the action, as command and HTTP hooks already can. Among the many fixes are repairs for Code Mode scripts, MCP tool handling on third-party providers, cloud-session recovery, scheduled tasks and terminal input. Self-hosted runner updates include new session-queue metrics and labels that can be passed to runner hooks.
Why it matters
This is a substantial update to a coding assistant used across different providers and session types. The auto-mode change extends a capability across more hosting environments, while the hook option gives administrators a clearer way to prevent an action when an MCP check fails. The runner metrics also give operators more detail about queue delays and sessions that fail to start.
The unusually long fix list reaches into everyday friction, from lost terminal input to sessions and tools behaving inconsistently. Release notes do not tell us how often each problem affected users, but they do give developers concrete behaviours to check against their own setups.
Our read
There is real utility here, not just a version number with a fresh coat of paint. If you rely on auto mode, MCP hooks, Code Mode or self-hosted runners, the changes are worth checking against your configuration. For everyone else, the quieter fixes may be the ones that make a long coding session feel less like a test of patience.
What to watch
- How auto mode behaves across the newly listed providers.
- Whether onFailure: block gives teams the control they need for MCP tool hooks.
- Whether developers see the listed session, Code Mode and runner fixes in practice.
Discussion spark: For AI coding tools, should teams prioritise new ways to run sessions, or spend more release effort making permissions, hooks and long-running workflows dependable?
Sources and evidence
- v2.1.328 (19 November 2026, 01:14 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.