Claude Code 2.1.325 stops sandboxed Bash from running without prompts by default
Anthropic's Claude Code 2.1.325 makes sandboxed Bash commands follow the normal permission flow unless sandbox.autoAllowBashIfSandboxed is set to true.
Anthropic's Claude Code 2.1.325, dated November 13, 2026, makes sandboxed Bash commands follow the normal permission flow by default. Users who want those commands to keep running without prompts must set sandbox.autoAllowBashIfSandboxed to true. Put simply, a command inside the sandbox no longer skips the permission check just because it is sandboxed. The setting is the opt-out, and it has to be written into configuration to take effect.
For organizations whose managed settings turn the sandbox on, the notes give a more specific instruction: set sandbox.autoAllowBashIfSandboxed to true in your highest-ranking managed source. That wording matters, since managed settings can come from several places, and the notes say the flag belongs in the one that ranks highest. The release also changes what the sandbox will write. It now refuses sandboxed writes to .claude settings, skills, agents, commands and hooks in added directories. The launch folder already had this protection before this release.
Several other permission changes ship in the same release. Read-only git commands now ask first in a repository that isn't the project's or one the user has trusted. To run them without prompts in such a repository, the notes say to name it with --add-dir rather than the additionalDirectories setting. Scheduled tasks and /loop now ask before running a skill that declares allowed-tools or hooks. A user can add Skill(name) to permissions.allow to let it run unattended. In Desktop, IDE and SDK hosts, the 'Always allow' option offers a rule only when saving it would allow that command again. The notes point out the consequence: a scheduled task that relied on such a rule can now wait for the user.
The release also lists sandbox and permission fixes among its 88 bug fixes. These include a case where the Bash sandbox did not fully protect a worktree's settings and hooks, another where it did not fully protect a linked worktree's shared git folder, a gap in Bash permission checks that let some compound commands run without asking, and a case where a Bash allow rule could run a different program than the one it names. The notes give one line for each and do not say how many users or setups were affected. A separate change has the sandbox's own helper programs, such as its network bridge, run without Claude Code's sign-in tokens and API keys in their environment.
What changed
Sandboxed Bash commands now follow the normal permission flow by default instead of running without prompts, and users must set sandbox.autoAllowBashIfSandboxed to true to keep the old behaviour.
Type: Breaking change · Takes effect: Friday, November 13, 2026 · Applies to: Claude Code 2.1.325
Why it matters
Teams that relied on the sandbox to skip approvals may find unattended or scripted runs pausing for input after upgrading. Managed rollouts need the new flag set explicitly.
What you can do
Before rolling 2.1.325 out to unattended runs, check managed and project settings for sandbox.autoAllowBashIfSandboxed.
Original source
Where we saw it
Claude Code changelog
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works