Claude Code 2.0.41-2.0.47 Release Notes

📋 Version Overview

This article summarizes the important updates in Claude Code versions 2.0.41 through 2.0.47, covering new features, improvements, and bug fixes across six releases.

🎯 Highlights

✨ PermissionRequest Hook: Automated permission approval mechanism
☁️ Azure AI Foundry Support: New cloud platform integration
🔧 Subagent Hooks Enhancement: More complete lifecycle control
🚀 Background Task Sending: Support for background execution on web
🐛 Extensive Bug Fixes: Improved stability and user experience

✨ New Features

1. PermissionRequest Hook (v2.0.45)

The new PermissionRequest Hook allows you to automatically approve or deny tool permission requests using custom logic, significantly improving automation workflow efficiency.

graph TD
    A[Tool Requests Permission] --> B{PermissionRequest Hook}
    B -->|Auto Approve| C[Execute Tool]
    B -->|Auto Deny| D[Stop Execution]
    B -->|No Hook Config| E[Ask User]

    style A fill:#e1f5fe
    style C fill:#c8e6c9
    style D fill:#ffcdd2
    style E fill:#fff3e0

Use Cases:

Auto-authorize specific tools or paths
Require manual confirmation for sensitive operations
Establish hierarchical permission management strategies

Configuration Example:

// .claude/hooks/permission-request.ts
export default async function({ tool, parameters }) {
  // Auto-approve read operations
  if (tool === 'Read' && parameters.file_path.startsWith('/safe/')) {
    return { approved: true };
  }

  // Deny writes to sensitive files
  if (tool === 'Write' && parameters.file_path.includes('secret')) {
    return { approved: false, reason: 'Sensitive files cannot be modified' };
  }

  // Let user decide for other cases
  return { approved: null };
}

💡 Further Reading: Hooks System and Migration Guide

2. Azure AI Foundry Support (v2.0.45)

Claude Code now supports the Azure AI Foundry platform! You can use Claude models in your Azure cloud environment.

Quick Start:

Set up Claude model endpoints in Azure AI Foundry
Configure Claude Code’s Azure credentials
Use via standard CLI or API

📘 Official Documentation: Azure AI Foundry Setup Guide

3. Subagent Hooks Enhancement (v2.0.42 & v2.0.43)

v2.0.42: New Fields in SubagentStop Hook

interface SubagentStopHookInput {
  agent_id: string;              // New: Agent ID
  agent_transcript_path: string; // New: Full conversation transcript path
  // ... other existing fields
}

v2.0.43: New SubagentStart Hook

// Triggered when subagent starts
export default async function(input: SubagentStartHookInput) {
  console.log(`Subagent ${input.subagent_type} started`);
  // Can be used for logging, monitoring, or initialization
}

Value:

Complete subagent lifecycle tracking (Start → Stop)
Access full conversation transcripts for analysis
Implement agent-level monitoring and logging

💡 Further Reading: Subagents and Plan Mode

4. Send Background Tasks to Web (v2.0.45)

Add the & symbol at the beginning of a message to send background tasks to Claude Code Web for execution.

# In CLI
& Run long-running data processing task

# Task will run in web background, CLI can continue other work

Suitable Scenarios:

Long-running tests or builds
Large codebase analysis
Tasks requiring continuous monitoring

🔧 Improvements

Hook System Improvements

v2.0.41: Prompt-based Stop Hooks Support Custom Model

# .claude/settings.json
{
  "hooks": {
    "stop": {
      "model": "haiku"  # New: Specify model for hook evaluation
    }
  }
}

v2.0.43: New tool_use_id Field

interface PreToolUseHookInput {
  tool_use_id: string;  // New: Unique tool call identifier
  // ...
}

v2.0.43: SDK Custom Timeout Support

// Custom hook execution timeout
{
  "hooks": {
    "timeout": 30000  // 30 seconds
  }
}

Skills System Improvements (v2.0.43)

New skills frontmatter field to auto-load specified skills for subagents:

# custom-agent.yaml
skills:
  - pdf-processor
  - data-analyzer

Benefits:

Reduce redundant configuration
Ensure subagents have necessary capabilities
Improve task execution efficiency

Permission and Security Improvements (v2.0.41)

Expanded list of safe git commands that don’t require approval, reducing unnecessary permission prompts:

# Following commands can run without user approval
git add
git rm
git submodule
# ... and other safe operations

Custom Agent Permission Mode (v2.0.43)

New permissionMode field allows setting permission policies for custom agents:

# custom-agent.yaml
permissionMode: "auto"  # Options: auto, ask, deny

🐛 Bug Fixes

v2.0.41

✅ Fixed slash commands from user settings being loaded twice, causing rendering issues
✅ Fixed incorrect labeling of user settings vs project settings in command descriptions
✅ Fixed crash when plugin command hooks timeout during execution
✅ Fixed Bedrock users seeing duplicate Opus entries with --model haiku
✅ Fixed broken security documentation links in trust dialogs and onboarding
✅ Fixed pressing ESC to close diff modal also interrupting the model
✅ ctrl-r history search landing on slash command no longer cancels search

v2.0.43

✅ Fixed nested CLAUDE.md files not loading when @-mentioning files
✅ Fixed duplicate rendering of some messages in UI
✅ Fixed some visual flickers
✅ Fixed NotebookEdit tool inserting cells at incorrect positions when cell IDs match cell-N pattern

v2.0.45

No new bug fixes (primarily a feature release)

v2.0.46

✅ Fixed image files being reported with incorrect media type when format cannot be detected from metadata

v2.0.47

✅ Improved error messages and validation for claude --teleport
✅ Improved error handling in /usage command
✅ Fixed race condition with history entry not getting logged at exit
✅ Fixed Vertex AI configuration not being applied from settings.json

💡 Best Practices

1. Leverage PermissionRequest Hook

Recommended Configuration Strategy:

// Hierarchical permission management
export default async function({ tool, parameters }) {
  // Layer 1: Fully trusted paths
  if (isTrustedPath(parameters.file_path)) {
    return { approved: true };
  }

  // Layer 2: Read-only operations
  if (isReadOnlyTool(tool)) {
    return { approved: true };
  }

  // Layer 3: Require manual confirmation
  return { approved: null };
}

2. Monitor Subagent Lifecycle

Implement complete monitoring using new hook fields:

// SubagentStart Hook
export default async function(input) {
  await logStart(input.subagent_type, input.agent_id);
}

// SubagentStop Hook
export default async function(input) {
  await logStop(input.agent_id);
  await analyzeTranscript(input.agent_transcript_path);
}

3. Upgrade Considerations

📌 Check if using deprecated Output Styles (refer to v2.0.30 Migration Guide)
📌 Update hook implementations to utilize new fields (agent_id, tool_use_id, etc.)
📌 If using Azure, refer to official documentation for setup
📌 Test PermissionRequest Hook to ensure permission logic is correct

Official Documentation

Article Series

📝 Summary

Versions 2.0.41-2.0.47 bring practical permission management automation (PermissionRequest Hook), cloud platform expansion (Azure AI Foundry), and extensive stability improvements. All users are recommended to upgrade to the latest version and explore new hook features to optimize workflows.

Upgrade Recommendation:

# Update to latest version
claude update

For any questions or suggestions, please refer to the official documentation or discuss in the community!