Claude Code v2.1.325 fixes security issue and adjusts permission flow

Anthropic released Claude Code v2.1.325, which fixes a security issue in decoding zstd-compressed responses and resolves defects including subagent models being switched incorrectly, third-party gateways and Foundry sessions reporting context length errors every turn, and resumed sessions losing recent conversation. This version also adjusts permission behavior: Bash commands inside the sandbox follow the normal permission flow by default, and sandbox.autoAllowBashIfSandboxed: true must be enabled in settings for them to run without prompting; when the sandbox is enabled by managed settings, this item must likewise be set in the highest-priority managed source. Scheduled tasks and /loop will ask before running skills that declare allowed-tools or hooks; adding Skill(name) to permissions.allow allows them to run without asking. In integrated terminals such as VS Code, this version adds prompt suggestions; press Tab to accept the suggested next prompt, and set promptSuggestionEnabled: false in settings.json to disable them.

AI overview based on source reports · Updated 16 hr ago