Claude Code v2.0.41

What: Comprehensive git status detection for branch management and workflow automation

How it works:

• Automatically detects commits ahead of default branch (main/master)
• Checks for uncommitted changes in working directory
• Detects unpushed commits to remote
• Warns before switching branches with uncommitted/unpushed work

Details:

• New functions: lJ1() at line 87137 (comprehensive status aggregator), Hn9() at line 87127 (commit counting), dJ1() at line 87177 (default branch detection)
• Integrated into branch switching workflows
• Prevents accidental loss of work during teleport operations
• Evidence: lJ1() at line 87137, Hn9() at line 87127, dJ1() at line 87177 in pretty-v2.0.41.js

What: Interactive dialog when switching branches with uncommitted or unpushed changes

How to trigger:

# Automatically appears when switching branches with uncommitted work
claude  # then use teleport feature

Details:

• Shows different warnings for uncommitted changes vs unpushed commits
• Provides three options: Continue without committing, Commit and push changes, Cancel operation
• Prevents data loss during branch switches
• Evidence: uT2() at line 428379 in pretty-v2.0.41.js (replaces unrelated function at line 428358 in v2.0.37)

What: Automatically configures upstream tracking after git checkout

How it works: After checking out a branch, automatically runs:

git branch --set-upstream-to=origin/<branch> <branch>

Details:

• Checks if upstream is already configured
• Verifies remote branch exists before setting upstream
• Logs all actions for debugging
• Called automatically after branch checkout operations
• Evidence: tl5() at line 427581 in pretty-v2.0.41.js (new function, doesn't exist in v2.0.37)

What: Glob-style brace expansion in CLAUDE.md frontmatter configuration

How to use:

paths: src/{components,utils}/{*.ts,*.tsx}

Expands to:

src/components/*.ts
src/components/*.tsx
src/utils/*.ts
src/utils/*.tsx

Details:

• Custom implementation supporting nested braces
• Recursive expansion for complex patterns
• Used in CLAUDE.md frontmatter paths: field
• Evidence: nP0() at line 52984, aP0() at line 53002 in pretty-v2.0.41.js (not present in v2.0.37)

What: OAuth-based GitHub account linking with Claude account

How it works:

• Automatically detects GitHub authentication via gh CLI
• Links GitHub account to Anthropic backend
• Enables enhanced GitHub integration features

Details:

• API endpoint: https://api.anthropic.com/api/claude_code/link_vcs_account
• Feature flag: vcs_account_linking_enabled
• Runs in background after authentication
• Tracks organization UUID and billing type
• Evidence: cl5() at line 426586, mt1() at line 426608 in pretty-v2.0.41.js (not in v2.0.37)

What: Detects when running inside Conductor desktop app

Details:

• Checks process.env.__CFBundleIdentifier === "com.conductor.app"
• Allows behavior customization for Conductor environment
• Similar to existing IDE detection (VSCode, Cursor, etc.)
• Evidence: Vb9() at line 47155 in pretty-v2.0.41.js (v2.0.37 has different code at line 47461)

What: Optional performance optimization using ripgrep for file pattern matching

How to use: Requires feature flag tengu_glob_with_rg to be enabled.

Details:

• Uses rg --files --glob <pattern> for matching
• Sorts results by modification time
• Includes hidden files
• Respects ignore patterns
• Performance benefit for large codebases
• Evidence: Feature flag check at line 483919 in pretty-v2.0.41.js (not in v2.0.37)

What: Structured 4-phase planning workflow with parallel agent coordination

How it works:

1. Phase 1: Initial Understanding - Read 3-4 relevant files, clarify ambiguities
2. Phase 2: Multi-Agent Planning - Launch 2 to N agents in parallel with different perspectives
3. Phase 3: Synthesis - Collect responses, ask trade-off questions
4. Phase 4: Final Plan - Present synthesized recommendation

Details:

• Quality over quantity: meaningful contrasts between agent perspectives
• Dynamic perspective generation (e.g., "simplicity vs performance vs maintainability")
• Parallel agent launches in single message (multiple tool calls)
• Context-aware perspective selection based on task type
• Evidence: Enhanced workflow at line 475195 in pretty-v2.0.41.js (basic version at line 473774 in v2.0.37)

What: Expanded detection of suspicious Windows network paths

Improvements over v2.0.37:

• Added DavWWWRoot detection (WebDAV mounted drives)
• Added IP address-based UNC path detection (IPv4 and IPv6)
• Added forward slash variants for all patterns
• Added SSL port notation patterns (@SSL@, @\d+@SSL)

Details:

• Protects against WebDAV attacks
• Validates paths before execution
• Shows security warnings for suspicious paths
• Evidence: ddA() at line 222047 in pretty-v2.0.41.js (enhanced from i2Q() at line 222404 in v2.0.37)

What: Improved parsing of terminal color codes and escape sequences

Details:

• Proper handling of 256-color codes (ESC[38;5;Nm)
• Proper handling of RGB color codes (ESC[38;2;R;G;Bm)
• Fixes rendering bugs with complex color sequences
• More accurate terminal width calculations
• Evidence: Km9() at line 76407, Em9() at line 76423 in pretty-v2.0.41.js (different code at same lines in v2.0.37)

What: Better handling of emoji and full-width characters in terminal output

Details:

• Uses Intl.Segmenter for proper grapheme cluster detection
• Detects full-width characters (CJK, emoji) for alignment
• Prevents terminal rendering issues with Unicode
• Evidence: Nm9() at line 76699, $m9 initialization at line 76716 in pretty-v2.0.41.js (enhanced from v2.0.37)

What: Comprehensive examples and guidance for todo list usage

Improvements:

• Added 4 detailed examples of when TO use (dark mode implementation, function renaming, e-commerce features, React optimization)
• Added 4 examples of when NOT to use (simple questions, informational requests, single edits, commands)
• Clarified activeForm requirement ("Fix bug" → "Fixing bug")
• Emphasized exactly ONE task must be in_progress at a time
• Added guidance not to mark tasks complete if tests fail

Details:

• Better user experience through clearer task status
• Prevents misuse for trivial tasks
• Ensures systematic progress tracking
• Evidence: Enhanced documentation in Kv0 tool description at lines 1287-1469 in diff

What: Clearer guidance on appropriate use of plan mode

Details:

• Only for tasks requiring code implementation planning
• NOT for research/understanding tasks
• Must clarify ambiguities before exiting
• Added 3 specific examples distinguishing planning vs research
• Evidence: Enhanced guidance in a12() at lines 1902-1917 in diff

What: Better handling of Language Server Protocol diagnostics

Details:

• Deduplication of diagnostics across multiple LSP servers
• More detailed logging for diagnostic delivery
• Better error messages for malformed LSP responses
• Reduces duplicate error reports in editor
• Evidence: Enhanced diagnostic handling at lines 1710-1734 in diff

What: More specific error types and clearer user feedback

New error types:

• repeated_529 - Rate limit errors with specific guidance
• prompt_too_long - Oversized prompts with size information
• pdf_too_large - PDF page limit violations
• oauth_org_not_allowed - OAuth restriction details

Details:

• Better categorization of error conditions
• More actionable error messages
• Context-specific troubleshooting hints
• Evidence: New error types at lines 1499-1554 in diff

What: Enhanced explanation of slash command behavior

Details:

• Clearer explanation that commands expand to prompts
• Shows <command-message> format example
• Warning not to invoke commands already running
• Better understanding of command execution model
• Evidence: Enhanced documentation at lines 96963-96982 in diff

Separate token usage tracking for MCP tools, distinguishing them from built-in tools in context window breakdown.

Consolidated plugin hook execution summaries showing hook count, errors, and prevention status with better formatting.

Added OSC 8 hyperlink support for creating clickable links in supported terminals using standard ANSI escape sequences.

Better warnings for truncated content, showing kilobytes removed for large tool results and warnings for git status truncation.

Enhanced thinking mode integration with better metadata passing and settings synchronization for alwaysThinkingEnabled.