Source Intelligence

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

All releases Home olderv2.0.47 v2.0.50newer

Claude Code v2.0.49

9 entries read diff v2.0.47 → v2.0.49 Markdown

Version 2.0.49 introduces significant infrastructure improvements including a new GrowthBook feature flag system for dynamic feature control, unrestricted model access for Foundry users, and enhanced terminal rendering for emoji and Unicode characters.

Find
Pick an entry · j / k steps through
9 entries

Changesopen

GrowthBook Feature Flag Integration#

What: Complete integration of GrowthBook SDK (v1.6.1) for remote feature management and A/B testing How to use:

# Enable GrowthBook integration
export CLAUDE_CODE_ENABLE_GROWTHBOOK=true
claude

Details:

  • Enables Anthropic to dynamically control features without CLI updates
  • Supports real-time feature flag updates via Server-Sent Events (SSE)
  • Implements intelligent caching with 60-second stale TTL and 4-hour max age
  • Includes automatic retry logic with exponential backoff
  • Features pause when browser tab is hidden to save resources
  • Evidence: bn2() at line 447290, z31 class at line 448817, SDK key CO0 at line 35311

GitHub Repository Path Tracking#

What: New system to track which local directories correspond to which GitHub repositories How to use: This feature works automatically when you work with GitHub repositories. It enables repository-aware context switching and settings. Details:

  • Automatically tracks repo-to-directory mappings in settings
  • Enables better multi-repository workflow support
  • Stored in githubRepoPaths field in user settings
  • Evidence: lX9() at line 507265, iX9() at line 507276

Unrestricted Model Access for Foundry Users#

What: Foundry deployment users now have access to all Claude models without version restrictions Details: Previously, Foundry users had the same model restrictions as first-party users (excluding claude-3-* models). Now Foundry users can access all available models including legacy versions.

  • Evidence: Ez4() at line 86189 vs dC4() at line 86088 in v2.0.47

Enhanced Terminal Text Rendering for Emoji and Unicode#

What: Significantly improved display width calculation for emoji, flag characters, and complex Unicode scripts Details:

  • Adds fast-path optimization for ASCII strings (performance improvement)
  • Correctly handles Regional Indicator flag emojis (🇺🇸, 🇬🇧)
  • Proper width calculation for keycap sequences (1️⃣, 2️⃣)
  • Better support for zero-width characters in Indic, Arabic, and Hebrew scripts
  • Prevents terminal misalignment issues with emoji-containing content
  • Evidence: N94() at line 61894, q94() at line 61905, _g0() at line 61921

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

Improved HTTPS Proxy Compatibility#

What: Enhanced proxy agent with DNS address family normalization Details:

  • Adds custom DNS lookup callback for better proxy compatibility
  • Normalizes IPv4/IPv6 address family specifications
  • Prevents connection failures with proxies sensitive to address family format
  • Maintains existing mTLS certificate support
  • Evidence: WhQ() at line 146672, N$8() at line 146623

Enhanced Settings File Security Isolation#

What: Improved protection against accidental modification of parent workspace settings Details: When working in a subdirectory, Claude Code now denies write access to parent workspace .claude/settings.json and .claude/settings.local.json files. This prevents unintended configuration changes when working in nested directories.

  • Evidence: oA2() function sandbox configuration (diff line 5527-5531)

Improved Tool Schema Validation#

What: VerifyResult tool now includes JSON Schema in addition to Zod schema Details:

  • Provides dual validation formats for better compatibility
  • Enables clearer error messages when tool is used incorrectly
  • Supports both inputSchema (Zod) and inputJSONSchema (JSON Schema)
  • Evidence: UM3() at line 497621 vs r$3() at line 494892 in v2.0.47

Code Organization#

  • Refactored text segmentation functions for better emoji handling (y44() at line 76651, k44() at line 76643)
  • Consolidated import statements for better module organization
  • Updated internal function names for consistency (147 functions added, 22 removed)

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

Infrastructure#

  • Added support for dynamic feature rollouts via remote configuration
  • Implemented stale-while-revalidate caching strategy for feature flags
  • Enhanced analytics tracking for feature evaluation and experiment participation
Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.0.49. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.

  • Added readline-style ctrl-y for pasting deleted text
  • Improved clarity of usage limit warning message
  • Fixed handling of subagent permissions
System prompt

The system prompt was not captured for this release, so this page cannot say whether it moved.