Carry Code v0.7.5

Release Date: March 12, 2026
Version: 0.7.5
Commits: 121 commits since v0.7.4

✨ Major New Features

1. 🌐 Browser Automation Tool (`core_browser`)

Powerful built-in tool for web automation, enabling AI to interact with web pages:

Action	Description
`navigate`	Navigate to URLs with policy validation
`click`	Interact with page elements
`type`	Input text into form fields
`extract`	Extract structured content from pages
`screenshot`	Capture page screenshots
`evaluate`	Execute JavaScript in page context
`trace_start/trace_stop`	Record interaction traces
`close`	Clean up browser sessions

Key Features:

Session Pool Management: Multi-session support with automatic cleanup
URL Policy Validation: Configurable domain whitelist and scheme blocking
No Confirmation Required: Streamlined workflow for browser operations
Full Test Coverage: Comprehensive test suite for reliability

2. 🔍 SkillHub Integration (`/skill find-skills`)

Discover and install skills from Tencent SkillHub:

# Search for skills
/skill find-skills weather

Keyword Search: Find skills by name or description
Recent Search History: Quick access to past searches
Structured Results: Parsed and formatted skill information
One-Click Install: Direct installation from search results
Enhanced Prompts: Improved search guidance and suggestions

3. 🚀 `/init` Command - Project Initialization

New command to quickly initialize projects with AGENTS.md configuration:

/init

Features:

Auto-detect project types - Automatically identifies the project framework and language
Generate project-specific rules - Creates tailored AGENTS.md based on project type
Quick setup - One-command initialization for consistent project standards

Supported Project Types:

TypeScript/JavaScript projects
Python projects
Rust projects
Go projects
And more…

4. ❓ `/help` Command - Interactive Command Guide

Comprehensive built-in help system for all slash commands:

/help

Features:

Categorized by functionality - Commands organized by Session, Tools, Settings, etc.
Context-aware suggestions - Smart recommendations based on current context
Full i18n support - Available in both English and Chinese
Detailed examples - Practical usage examples for each command

5. 🔄 `/agent` Command - Quick Mode Switching

Instantly switch between Agent working modes:

/agent

Modes:

Plan Mode - Architecture and planning tasks (read-only analysis)
Build Mode - Implementation and coding tasks (read/write/execute)

Use Cases:

Quickly switch between analysis and implementation
Safer exploration in Plan mode before making changes
Streamlined workflow without manual mode selection

6. 📊 Token Budget Management & Statistics

Comprehensive token tracking and budget management system:

Input Token Estimation: Accurate prediction of request token usage before sending
Context Budget Tracking: Real-time monitoring of context window utilization
Cached Content Support: Proper handling of cached tokens for Gemini and Ollama
Call-Scoped Statistics: Detailed per-request token metrics with derive_call_scoped_stats
Smart Compaction Triggers: Auto-trigger context compaction based on token thresholds
Budget Snapshots: Logging and tracking of compaction budget decisions

Display Improvements:

Estimated token prefix in ProcessArea (e.g., “~1.2K tokens”)
Input/Output token breakdown in real-time
Idle-optimized stats fetching for better performance

7. 🛠️ Enhanced Tool Call Display

Completely redesigned tool result visualization:

Structured Tool Logs: Clear, consistent formatting across all tools
Duration Summary: Execution time displayed for each operation
Input/Output Summaries:
- Truncated to first 5 lines for readability
- Middle-omitted path truncation for long file paths
- 300-character summary limits
Markdown Rendering: Rich formatting with syntax highlighting
Key Path Sanitization: Safe display of paths in markdown autolinks

Tool-Specific Enhancements:

CoreFetch: Duration and output summaries with HTTP error handling
CoreBash: Input summary truncation, stdout/stderr filtering
MCP Tools: Consistent display formatting with input details

8. 📈 Mermaid Diagram Terminal Rendering

View Mermaid diagrams directly in your terminal:

Terminal-Native Rendering: ASCII-based diagrams without browser
Chinese Character Support: Optimized alignment for CJK characters
Smart Layout Handling: Dynamic width management and label preservation
Graceful Fallback: Plain text fallback when rendering fails
Beautiful Themes: Terminal color scheme integration

mermaid

9. 🔄 Session Checkpoints & Rollback

Advanced session management capabilities:

User Turn Checkpoints: Automatic checkpoint creation at user interactions
Rollback Mechanism: Revert to any previous checkpoint state
Session Snapshots:
- Compact, space-efficient storage
- Provider and model metadata preservation
- Source/target revision tracking
Soft-Delete Recovery: Reactivate previously cleared sessions
Async Approval: Non-blocking approval mode handling

10. 🎨 Terminal Background Detection

Intelligent terminal theming:

Background Color Detection: Automatic light/dark theme detection
OSC Sequence Support: Uses terminal escape sequences for detection
Theme Adaptation: UI components adapt to terminal background
State Preservation: Maintains paused state during detection

🔧 Additional Enhancements

Code Highlighting & Rendering

New summary Syntax: Specialized highlighting for operation summaries
Word-Boundary Wrapping: Improved readability for long lines
Comment Color Differentiation: Distinct styling for comments vs base text
Plaintext Handling: Better support for plain text output
Regex Highlighter: Enhanced syntax highlighting for various languages

CLI Improvements

Restructured Argument Parsing: More robust CLI command handling
New Commands: help, version, once, update with full functionality
Quick Start Guide: Built-in help with structured shortcuts
i18n Support: Full localization for CLI messages

Session Event Handling

Enhanced Subscription Management: Better event listener handling
Error Tracking: Improved error propagation and logging
Metadata Updates: Session metadata updates without full reload
Message Appending: Efficient message addition without re-rendering

Model Configuration

Simplified Configuration: Removed ui_model_name and api_model_name fields
Provider Information: Enhanced provider/agentl tracking in sessions
Usage Extraction: Improved token usage parsing for Ollama

🐛 Fixes & Improvements

Bash Tool: Simplified command banning logic, enhanced syntax validation with custom error handling
Browser Tool: Removed command confirmation requirement for smoother workflows
Latency Display: Cleaner output without providerId clutter
Tool Detection: Simplified logic in RichArea component
Response Summaries: File counts and row ranges in summaries
Session Stability: Improved test stability for snapshot handling

📦 Dependency Updates

Added beautiful-mermaid for diagram rendering
Updated Playwright and Puppeteer to latest versions
Enhanced test infrastructure with 20+ new test files

📝 Developer Updates

CI/CD: Scheduled cleanup workflow for old artifacts and caches
Test Coverage: Comprehensive tests for:
- Browser automation
- Token estimation
- Tool result formatting
- Mermaid rendering
- Session management
Code Quality: Refactored core modules for better maintainability
Documentation: Updated AGENTS.md and SKILL.md with clearer instructions

🎯 Upgrade Guide

Automatic Upgrade

# Using the install script
curl -fsSL https://carrycode.ai/install.sh | bash

# Or for Windows
irm https://carrycode.ai/install.ps1 | iex

Verify Installation

carry --version

🙏 Acknowledgments

Thanks to all contributors who made this release possible! Special thanks for the extensive work on browser automation, SkillHub integration, slash commands, and token management.

Happy coding with Carry Code v0.7.5! 🎉