All notable changes to ClaudeBox will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
0.3.1 - 2025-01-01
- Python 3.13 and 3.14 support in package classifiers
- Updated documentation to correctly show Python 3.10+ requirement (was incorrectly showing 3.9+)
0.3.0 - 2025-01-01
- Comprehensive Documentation: Complete documentation for painless developer onboarding and AI understanding
- Foundation: CHANGELOG.md, CONTRIBUTING.md, examples/README.md index, enhanced README.md
- Getting Started Guides (4 files): Installation, Quick Start, Authentication, First Session tutorials
- User Guides (6 files): Sessions, Skills, Templates, Security, RL Training, Workspace management
- Total: 14 documentation files with ~6,000+ lines
- Consistent structure: Overview → Reference → Patterns → Best Practices → Troubleshooting
- Complete coverage from installation to advanced features
- All 71 examples now indexed and discoverable in examples/README.md
- Enhanced README.md with comprehensive feature showcase and navigation
0.1.2 - 2025-01-01
- Session Persistence: Persistent sessions with workspace management at
~/.claudebox/sessions/session_idparameter for persistent sessionsClaudeBox.reconnect()classmethod to reconnect to existing sessionsClaudeBox.list_sessions()classmethod to enumerate all sessionsClaudeBox.cleanup_session()classmethod for manual cleanup
- Structured Logging: JSON Lines logging to
history.jsonlfor debugging and RL trainingActionLoggerclass for structured action logging- Session metadata tracking in
session.json - Resource metrics collection
- Modular Skills System: 9 built-in skills for extending capabilities
EMAIL_SKILL- Send emails via SendGridPOSTGRES_SKILL- PostgreSQL database connectionsMYSQL_SKILL- MySQL database connectionsREDIS_SKILL- Redis data storeAPI_SKILL- HTTP requests with requests/httpxAWS_SKILL- AWS SDK (boto3)DOCKER_SKILL- Docker CLISCRAPING_SKILL- Web scraping (BeautifulSoup, Playwright)DATA_SCIENCE_SKILL- Data science stack (pandas, numpy, matplotlib)- Custom skill creation with
Skilldataclass - Skill registry for managing skills
- Sandbox Templates: 6 pre-configured environments
SandboxTemplate.DEFAULT- Base runtimeSandboxTemplate.WEB_DEV- Web development (Node.js, TypeScript, Docker)SandboxTemplate.DATA_SCIENCE- Data science (Jupyter, pandas, scikit-learn)SandboxTemplate.SECURITY- Security research (nmap, wireshark) for authorized use onlySandboxTemplate.DEVOPS- DevOps tools (Docker, Kubernetes CLI)SandboxTemplate.MOBILE- Mobile development
- RL Training Support: Reinforcement learning scaffolding
- 5 built-in reward functions:
SuccessOnlyReward,CodeQualityReward,SafetyReward,EfficiencyReward,CustomReward combine_rewards()to merge multiple reward functionsTrajectoryExporterfor exporting training data- State-action pair extraction
- Trajectory merging for multi-session training
- 5 built-in reward functions:
- Security Policies: Fine-grained security control
- 5 pre-defined policies:
UNRESTRICTED,STANDARD,RESTRICTED,READONLY,RESEARCH - Custom security policy creation
- Network access restrictions
- Filesystem isolation
- Command blocking
- Resource limits
SecurityPolicyEnforcerfor policy enforcement
- 5 pre-defined policies:
- Observability: Enhanced monitoring capabilities
get_metrics()method for current resource usageget_history_metrics()method for historical metrics- Structured action logging
- Enhanced
ClaudeBoxconstructor with new parameters:session_id- For persistent sessionsworkspace_dir- Custom workspace locationenable_logging- Toggle structured loggingskills- List of skills to pre-loadtemplate- Sandbox template selectionreward_fn- Reward function for RL trainingsecurity_policy- Security policy to enforce
CodeResultdataclass now includes:rewardfield - Calculated reward (if reward function provided)action_logfield - List of actions taken during execution
- Improved error handling and type hints throughout codebase
- Import ordering issues (ruff compliance)
- Type checking errors (mypy compliance)
- File opening modes in trajectory export
- Unused variable warnings
0.1.1 - 2024-12-XX
- Initial PyPI publishing workflow
- Docker image publishing to GHCR
- Fixed Docker image name in publish workflow
- Improved import ordering in box.py
- Docker image naming consistency
0.1.0 - 2024-12-XX
- Initial release of ClaudeBox
- Basic
ClaudeBoxclass for running Claude Code in micro-VMs - Integration with BoxLite for hardware isolation
code()method for executing Claude Code commands- OAuth token authentication support
- Basic error handling and result types
- MIT License