Skip to content
AI.info

The Pulse

Anthropic adds experimental Code Mode to Claude Code

Claude Code can now use JavaScript to call its tools through an experimental opt-in mode. The release also adds plugin version checks and fixes for permissions and session sign-ins.

Anthropic adds experimental Code Mode to Claude Code

Claude Code can now write JavaScript to call its tools when users opt in to an experimental feature called Code Mode. The feature arrives in version 2.1.323, according to Anthropic’s changelog, alongside changes to plugin compatibility and a broad set of fixes.

Organizations can also turn Code Mode off. The release notes point to documentation for enabling the feature and disabling it for an organization.

Plugin version checks

The update lets plugin authors specify a supported Claude Code version range in plugin.json using engines["claude-code"]. Plugins outside their stated range are not installed, updated or loaded; installation and update select the newest version that fits.

The plugin validation command now type-checks a plugin’s TypeScript hooks module. Each finding is reported as a warning, while the --strict option causes the command to fail on a finding. Validation also includes the $ nouns that a mod’s engine.create hook adds, withholds or replaces.

For plugin installations that return JSON, the result now includes unsetOptionCount, the number of user configuration options that have not yet been set. A new warning identifies a plugin’s language server when it cannot start and says what to check.

Permission and sandbox fixes

Anthropic also fixed a case in which the Bash sandbox could switch off during a session and another in which Bash permission checks missed part of a command. A separate fix addresses a case where the Write and Edit tools could write to a path covered by a Read deny rule.

The release fixes cases where Bash deny and ask rules missed commands run through wrappers such as sudo and env. It also fixes some cases in which Bash allow rules with an attached wildcard skipped a permission prompt.

Claude Code also stops at startup and identifies an invalid setting when sandbox-enabling settings fail validation in claude -p or the Agent SDK. Previously, those situations could result in Bash running without the sandbox.

Sessions and other changes

The release also fixes a background session sending a conversation under another sign-in when its named Anthropic profile was missing, changed or overridden. In that situation, the session now stops and says so. Another fix addresses background sessions moving between a profile and the stored login when an update restarted them.

For long-running hooks in claude -p and SDK sessions, the update adds a progress heartbeat every 30 seconds so a host such as Claude Desktop will not end the turn as hung. That behavior for Claude Desktop takes effect once the desktop app includes this version.

The update adds a menu with options to stop, switch to usage credits or upgrade when a model needs usage credits that are turned off or used up. It also adds .docx support for Artifact tool asset uploads and a startup warning for Bash allow rules that put a wildcard in a variable’s value.

Sources

Related stories

Explore

More articles