- Replace --dangerously-skip-permissions with official headless mode flags
- Use -p (--print) for non-interactive mode
- Add --output-format json for structured responses
- Use --permission-mode acceptEdits for automation
- Parse JSON response and extract metadata (cost, duration, session_id)
- Handle both successful and error responses properly
- Reference: https://docs.anthropic.com/en/docs/claude-code/sdk/sdk-headless
- Created extensible agent interface with install() and execute() methods
- Moved Claude Code logic to agents/claude.ts implementing Agent interface
- Added utilities directory for reusable functions (exec, files)
- Refactored index.ts to be minimal (35 lines) using agent abstraction
- Made anthropic_api_key optional in action.yml
- Updated Node.js imports to use node: prefix convention
- Bumped version to 0.0.5
- Architecture now supports multiple agents (OpenAI, Gemini, etc.)
Colin McDonnell