Claude Code release notes: 2.1.206-2.1.207 trim instructions and expand Auto mode

Claude Code 2.1.206 and 2.1.207 improve two everyday parts of the product: project instructions can get shorter, and Auto mode becomes available without an opt-in on more enterprise platforms. That combination makes sense. More autonomy works better when the instructions are easy to understand.

Claude Code 2.1.206 helps trim CLAUDE.md

CLAUDE.md is the persistent project instruction file that Claude Code reads when a session starts. Version 2.1.206 added a /doctor check that proposes trimming checked-in CLAUDE.md files by removing content Claude can derive from the codebase.

Anthropic's documentation recommends concise, specific instructions and a target of fewer than 200 lines per file. Procedures that matter only sometimes are better placed in skills or path-scoped rules. This saves context and keeps important rules from getting buried under facts already visible in the code.

Sources: Claude Code 2.1.206; How Claude remembers your project

Claude Code 2.1.207 expands cloud Auto mode

Auto mode is Claude Code's mode for taking more tool actions without asking at every step, within the rules and settings set by the organization. In 2.1.207, the mode is available without the CLAUDE_CODE_ENABLE_AUTO_MODE environment variable on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. It can be disabled with disableAutoMode in settings.

The same release stops reading autoMode from the repo-local .claude/settings.local.json and uses the user-level ~/.claude/settings.json instead. Anthropic also changed the default model to Claude Opus 4.8 on Bedrock, Vertex, and Claude Platform on AWS.

Teams running Claude Code through a cloud provider should do a quick review before the next rollout: which model is active, where is Auto mode configured, and which instructions truly need to follow every session?

Sources: Claude Code 2.1.207; Claude Code settings

Two smaller fixes that matter in real workflows

Version 2.1.206 also fixed timeouts for MCP servers configured in .mcp.json or through --mcp-config. MCP, or Model Context Protocol, is the standard that connects Claude to external tools and data sources. Long tool calls should now respect each server's request_timeout_ms in fresh sessions.

Version 2.1.207 fixed terminal freezes during very long responses, along with several Remote Control, AWS authentication, and background-session problems after network interruptions. These changes do not create a new workflow, but they remove friction when Claude Code runs inside a longer team process.

Sources: Claude Code 2.1.206; Claude Code 2.1.207

Try this prompt this week

Human step: Confirm that Claude Code is running 2.1.207 and open /doctor. Review its proposal before changing project files or user settings.

Source: Claude Code 2.1.206

Read this project's CLAUDE.md, relevant .claude rules, and settings scopes.
Flag anything that can be derived directly from the codebase.
Propose a shorter version that keeps project-specific commands and decisions.
List the Auto mode and model settings the team should verify separately.
Do not edit files until I approve the proposal.

Good output should show:

• which lines duplicate information already visible in the code,
• which project-specific instructions must stay,
• where user, project, and managed settings apply,
• a short review packet before any file changes.

For teams connecting Claude Code to real tools, this is a practical Tool Forge task: keep instructions understandable, store secrets in environment variables or a secret manager, and use scoped permissions, approval gates, and run logs where needed.