Recent updates to Claude Code introduce several new features and improvements for users. A new restricted mode enhances security by limiting tool access and file operations. Agents now support per-agent prompt caching, and diagnostics for server-managed settings have been improved.
Key Points
- A new
--restrictedmode (orCLAUDE_CODE_RESTRICTED=1) removes built-in tools that run commands or code and WebFetch, unless explicitly named in--tools. - The restricted mode keeps file tools within the working directory, refuses
bypassPermissions, and ignores user, project, and local settings files. experimental.cacheTtl(e.g., "5m" or "1h") has been added to agent frontmatter for per-agent prompt cache time-to-live when no subagent TTL is configured.claude self-hosted-runner --client-label <label>(orSELF_HOSTED_RUNNER_CLIENT_LABEL) allows overriding the label a runner registers with.- Server-managed settings diagnostics now provide a startup warning for load failures and
/doctorand/statuslines explaining load issues or why settings were not fetched. - A warning in
/web-setupindicates when the GitHub CLI token lacks the workflow scope. /usage-creditsis available for Enterprise organizations billed via AWS Marketplace, self-serve Enterprise, and Enterprise trials, allowing members to request higher usage limits.
Context
According to the Claude Code changelog, these updates aim to provide users with more control over security, performance, and system diagnostics. The introduction of a restricted mode allows for tighter control over the execution environment, while caching options can help optimize agent performance.
Why It Matters
These updates offer builders enhanced security controls and performance tuning options for their Claude Code environments. The improved diagnostics can also streamline troubleshooting for server-managed settings, reducing operational friction.
What To Do
- Review the
--restrictedmode documentation to understand its implications for tool access and file operations. - Experiment with
experimental.cacheTtlin agent frontmatter to optimize prompt caching for specific agents. - Note the new
claude self-hosted-runner --client-labeloption for custom runner labeling. - Check
/doctorand/statusfor server-managed settings diagnostics if encountering load failures.
