# Claude Code v1.0.74

> Claude Code v1.0.74, released 12 Aug 2025 (2025-08-12). 11 entries read out of the shipped bundle. Unofficial, and not affiliated with Anthropic.

Web version: https://changelogs.core-directive.com/v/1.0.74

Claude Code 1.0.74 introduces Sonnet 4 with 1M token context for eligible organizations, a powerful new custom output modes system that lets you personalize how Claude communicates, and improved bash command handling with enhanced security controls.

## Changes

### Sonnet 4 with 1M Token Context

**What:** Access to Claude Sonnet 4 with an extended 1 million token context window for processing large codebases and documents **How to use:**

`# Select from the model menu claude /model # Choose "Sonnet (1M context)" option # Or set directly claude /model sonnet[1m]` **Details:**

- Available to eligible organizations only (checked automatically)
- Higher token costs: $6/$22.50 per million tokens (vs $3/$15 for regular Sonnet)
- Automatically selected as default for users with Opus mode and 1M access
- One-time welcome notification explains the feature and rate limit implications
- Access status cached for 1 hour to minimize API calls

### Custom Output Modes

**What:** Create personalized output modes that modify how Claude Code communicates with you **How to use:**

`# Create a new custom output mode claude /output-mode:new # Select from available modes claude /output-mode # Set a mode directly claude /output-mode Concise` **Details:**

- Output modes are markdown files with YAML frontmatter stored in `~/.claude/output-modes/` or `.claude/output-modes/`
- Built-in modes "Insights" and "Learn By Doing" can be overridden with custom versions
- New `output-mode-setup` agent guides you through creating modes with natural language
- Modes can customize response length, tone, format, and workflow preferences
- Project-level modes allow teams to share communication styles

### Opus Plan Mode

**What:** New hybrid model option that uses Opus 4.1 in plan mode and Sonnet 4 for execution **How to use:**

`claude /model # Select "Opus Plan Mode" option` **Details:**

- Description: "Use Opus 4.1 in plan mode, Sonnet 4 otherwise"
- Provides optimal model selection based on the current operation mode
- Automatically switches between models for planning vs implementation

### Enhanced Bash Command Generation

**What changed:** Improved handling of bash redirections and file descriptors **Impact:** More reliable command execution with complex shell redirections like `2>&1`, `1>/dev/null`, and file descriptor routing

### Granular Permission System

**What changed:** Separated trust dialogs into three distinct categories **Impact:** Users can now approve bash commands, hooks, and general operations independently, providing finer control over security permissions

### UI Polish with Shimmer Effects

**What changed:** Added shimmer colors to all themes for Claude-related UI elements **Impact:** Subtle visual feedback improvements with colors like `rgb(245,149,117)` for light theme and `rgb(235,159,127)` for dark theme

### Command Source Tracking

**What changed:** All built-in commands now include a `source: "builtin"` field **Impact:** Better organization and potential for future command source filtering or custom command sources

### Organization Feature Tracking

**What changed:** Added first token date tracking and per-organization feature flag caching **Impact:** Better analytics and feature rollout capabilities for organization-specific features

### Fixed: Process Import Issue

- **Issue:** Incorrect import statement for process.cwd
- **Cause:** Using default import instead of named import
- **Resolution:** Changed from `import ndB from "node:process"` to `import { cwd as RM2 } from "node:process"`

### Fixed: Output Mode Key Consistency

- **Issue:** Inconsistent naming between internal keys and display names
- **Cause:** Built-in modes used kebab-case internally but title case for display
- **Resolution:** Standardized on title case for both internal keys and display ("Insights", "Learn By Doing")

### Fixed: Statsig Gate Caching

- **Issue:** Feature gates were being fetched repeatedly without caching
- **Cause:** Missing cache implementation for Statsig gates
- **Resolution:** Added `cachedStatsigGates` to configuration with proper cache management