Claude Code 2.1.325

Claude Code 2.1.325 fixed a security issue in how zstd-compressed responses are decoded. The release notes don't give more detail on the issue itself.

Several fixes touch subagents and model handling:

• Subagents with their own model no longer get moved to the plan-mode model when your model setting is opusplan or haiku.
• Subagents on a different model now size file reads and MCP results for their own model, not the main session's.
• Sessions on many third-party gateways and Foundry no longer stop with a context length error on every turn. They compact instead.

Other fixes:

• /security-review and skills' inline git commands no longer stop with an error in worktrees of bare repositories, --separate-git-dir checkouts, and Android repo checkouts.
• Recent turns no longer sometimes go missing from a resumed session after Claude Code stopped without a clean exit.
• Shell commands running in the launch directory were also fixed.

A few small additions: when your organization's policy blocks a request, the error now says how to continue (run /rewind and change the message, or run /clear). In VS Code and similar integrated terminals, a spinner tip about re-running a notebook cell appears once Claude has read or edited a Jupyter notebook. Claude also gets a note when a sandboxed git checkout, git stash or autostash pull can't update protected files such as those in .claude.