Claude Code 2.1.325: Experimental Code Mode, Safer Bash

Claude Code v2.1.323, released November 12, 2026, introduced Code Mode, an experimental opt-in feature that enables Claude to write JavaScript to coordinate multiple tool calls rather than invoking them individually. Version 2.1.325, released November 13, now requires explicit approval for sandboxed Bash commands by default, tightening permissions that previously bypassed the normal prompt flow.
Experimental Code Mode Changes How Claude Coordinates Tools
Without Code Mode, Claude requests tool calls such as Bash, Read, or Edit directly and reads each result before deciding what to do next. With Code Mode, Claude writes JavaScript that makes the tool calls, and Claude Code runs that code in the session. The feature is designed to batch operations rather than handling them one at a time.

Anthropic labels Code Mode as experimental and off by default. On Claude Code v2.1.323 or later, users can start Claude with CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 in their shell, or place it in the env block of ~/.claude/settings.json, then confirm activation with “Yes, turn on Code Mode.”
Code Mode only works in an interactive Claude session in a terminal; in claude -p, v2.1.325 prints a warning that Code Mode is not available and starts with the ordinary tools. This means scripts and continuous integration workflows that pick up the environment variable continue to function without the new mode.
Stricter Permission Requirements for Sandboxed Commands
Version 2.1.325 changed sandboxed Bash commands to follow the normal permission flow by default. Previously, sandboxed operations could skip prompts, but now users must approve each command unless they set sandbox.autoAllowBashIfSandboxed to true to restore the previous no-prompt behavior.

The 2.1.325 update also fixed Code Mode scripts ignoring a worktree entered with EnterWorktree, writing ~ paths into the project, and leaving /mcp showing an MCP server as connected after its sign-in expired. Additional fixes addressed recent turns sometimes going missing from a resumed session after Claude Code stopped without a clean exit.
Service Incidents Affect Claude Code and Models
Claude Code and the broader Claude platform experienced service disruptions during the week of November 11. A major incident affected Claude Code and the Claude Agent SDK while Claude Sonnet 5.5 and Claude Fable 5.2 were experiencing elevated errors. Anthropic resolved an incident that disrupted Claude.ai, the Claude API, Claude Code, and Claude Cowork from November 15 into November 16, deploying a fix and monitoring recovery.
Developer Response and Community Tools
By the weekend following the November 12 release, developers on X were calling Code Mode the end of chatty agents, and community-created configuration tools began to emerge. Third-party community tools such as claude-code-templates offered configuration commands, though these are not official software or documentation.
Claude Code version 2.1.325 had accumulated 1,562,814 downloads in the seven days following its release, reflecting strong adoption of the terminal-based coding agent.

Broader Enterprise Adoption Concerns
While Anthropic expands Claude Code features, some financial institutions are restricting access. Goldman Sachs reportedly restricted most employees from using Anthropic’s Claude Code and Microsoft’s GitHub coding tools after previously allowing experimentation, reflecting growing enterprise scrutiny around the security and governance of AI coding agents.

Anthropic’s newsroom published a detailed account of how Claude Code evolved from an internal command-line tool into a broader coding agent, including perspectives from researchers, engineers, and early users.
Key Facts
- Code Mode release: Version 2.1.323 on November 12, 2026 introduced the experimental opt-in feature
- Latest version: 2.1.325 released November 13, 2026 with stricter sandbox permissions
- Activation requirement: CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 environment variable and user confirmation
- Terminal only: Code Mode works exclusively in interactive terminal sessions, not in claude -p or CI workflows
- Downloads: Over 1.5 million downloads in the week following the 2.1.325 release
- Service disruption: Platform incidents affected Claude Code and core models November 15-16
Sources
- Vibecoding.cz: Claude Code Weekly 2026 W46
- Anthropic Newsroom
- Technobezz: Claude Service Incident Resolved
- Neomanex: Changelog 2026 W46
Sources
- Claude Code Changelog (November 2026)
- Claude Code’s Experimental Code Mode Batches Tool Calls in JavaScript. The Viral 37% Figure Is From Elsewhere
- Claude Code: Anthropic’s Terminal AI Coding Agent
- Claude Code Code Mode: How to Turn It On and What It Changes
- nick on X: “claude code has a code mode now where it writes one script to run a bunch of tool calls at once. 37% fewer tokens. my max plan is gonna last way longer based on all the amazing conversations from AIE Code this past week, the bigger savings in the future will likely be sending the tiny… / X
- @anthropic-ai/claude-code – npm







