Version 2.0.45

What: Claude Code now supports connecting to Claude models deployed on Azure AI Foundry with Azure AD token authentication

How to use:

# Enable Foundry mode
export CLAUDE_CODE_USE_FOUNDRY=true

# Option 1: Use API key authentication
export ANTHROPIC_FOUNDRY_API_KEY=your-api-key
export ANTHROPIC_FOUNDRY_BASE_URL=https://your-resource.services.ai.azure.com/anthropic/

# Option 2: Use resource name (automatically constructs URL)
export ANTHROPIC_FOUNDRY_RESOURCE=your-resource-name

# Option 3: Use Azure AD authentication (no API key needed)
# Azure AD token provider is automatically configured

Details:

• Supports both API key and Azure AD token provider authentication methods
• Can specify either a full baseURL or a resource name (which constructs the Azure services URL automatically)
• Azure AD authentication uses the https://cognitiveservices.azure.com/.default scope
• For testing/debugging, you can skip authentication with CLAUDE_CODE_SKIP_FOUNDRY_AUTH=true
• Foundry deployments are now recognized as a first-party environment alongside standard Anthropic deployments
• Evidence: dsA class at line 202345, Foundry client initialization in xw() at line 243262, environment detection via j3() at line 86382

What: New internal function to check if experimental beta features should be enabled based on deployment environment

Details:

• Experimental features are automatically enabled for firstParty and foundry deployments
• Can be disabled with CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=true environment variable
• Evidence: mq4() at line 86382

What: Replaced optional native image processor module with pure Sharp-based implementation for more reliable and consistent image handling

Details:

• Previous versions attempted to use a compiled native module (image-processor.node) with Sharp as a fallback, often showing warnings like "Native image processor not available, falling back to sharp"
• New implementation uses Sharp exclusively with a sophisticated multi-stage compression strategy:

1. Multi-scale resizing (100%, 75%, 50%, 25% of original dimensions)
2. PNG-specific palette compression (64 colors at 800x800)
3. Progressive JPEG quality reduction (starting at quality 50 for 600x600)
4. Final fallback (400x400 at quality 20)

• Eliminates dependency on native compilation, improving reliability across different platforms
• More predictable behavior with deterministic compression strategies
• Evidence: Removed UqQ() at line 280470, $qQ at line 280527, qqQ() at line 280537 in v2.0.44; Added K5A() at line 156002, _P8() at line 156053, xP8() at line 156081, vP8() at line 156089, bP8() at line 156097 in v2.0.45

What: Terminal cursor positioning now uses semantic event types instead of raw ANSI control sequences

Details:

• Replaced raw stdout strings (like \r for carriage return, \b for wrap handling) with semantic event types
• New event types: resolvePendingWrap, carriageReturn, cursorMove
• Supports returning multiple terminal events per transaction for more complex operations
• Improves code maintainability and testability while producing identical terminal output
• Evidence: Replaced Hk0() at line 77408 with Ii0() at line 77184, updated supporting class from zk0 to Yi0

What: Model name display logic now recognizes Foundry deployments and adjusts accordingly

Details:

• Foundry deployments skip the friendly model name display to show raw model identifiers
• Ensures consistent behavior across different deployment environments
• Evidence: Y6Q() at line 86439 includes foundry environment check

• Language Server Protocol support remains fully functional with comprehensive code reorganization
• Main manager function renamed and restructured for improved maintainability
• All LSP features (initialization, diagnostics, workspace configuration) preserved
• Evidence: _MQ() at line 288094 renamed to C62() at line 314954

• Semantic versioning regex patterns module moved to a different location in the bundled code
• All regex patterns remain functionally identical
• Pure internal refactoring with no user-facing impact
• Evidence: s0A at line 80963 in v2.0.44 moved to Z5A at line 150649 in v2.0.45

• Updated @azure/msal-node to v3.8.1
• Updated @azure/msal-common to v15.13.1
• Added support for Azure identity authentication flows
• Internal version identifier updated from 0.68.0 to 0.70.0

• Added graceful shutdown handler for telemetry operations
• Evidence: wj1() at line 148844

• Structural similarity between v2.0.44 and v2.0.45: 89.7%
• Total changes: 1,069 additions, 52 deletions, 24 modifications
• Significant internal variable renaming (9,301 renames) due to minification changes
• No breaking changes to public APIs or command-line interface