- Transitioned Gemini provider to use the dedicated `systemInstruction` field for system prompts, resolving API role validation errors that occurred when using a 'system' role within the `contents` array.
- Enhanced response parsing to skip empty text parts, allowing for reliable extraction of content even when models return responses with intermediate empty parts.
- Added comprehensive tests to validate the new payload structure, ensure blank system instructions are ignored, and confirm robust response parsing behavior.

- Introduced type hints to the `content_text` variable within the Gemini API response handling.
- Applied linting rules to ensure consistent code style and readability.
- These changes improve code quality and maintainability by making the code more explicit and adhering to established standards.

fix: Adopt systemInstruction for Gemini API and improve response parsing
fix: Add type hints and linting to Gemini text extraction

thanks @vinsidious!

- Update all test references from gemini-1.5-pro to gemini-2.5-flash-lite model
- Add new integration test class for real API calls with valid credentials
- Improve test coverage with actual API endpoint verification when environment is configured

- Introduce custom-anthropic provider for Anthropic-compatible endpoints with support for custom base URLs and API version headers
- Add custom-openai provider for OpenAI-compatible endpoints including Azure OpenAI Service and self-hosted solutions
- Update CLI initialization to guide users through custom provider setup with required base URL and API key configuration
- Extend environment configuration examples with new custom provider variables and documentation
- Implement comprehensive test suites covering authentication, error handling, and edge cases for both custom providers
- Add integration tests for real API calls with valid credentials
- Handle system message extraction for Anthropic-compatible APIs and extended response formats with thinking traces

- Updated gac-simple-usage.png to dark mode variant for better visibility
- Reduced file size from ~100KB to ~52KB for faster loading

- Change "Custom Endpoint" to "Custom Endpoints" for proper pluralization
- Improves consistency with other provider listings in the supported providers section

- Remove <think> and </think> reasoning blocks from AI provider outputs
- Handle multi-line reasoning blocks while preserving inline tag mentions
- Clean orphaned opening/closing tags that appear before/after commit messages
- Support case-insensitive tag removal (<think>, <THINK>, etc.)
- Add comprehensive test coverage for various reasoning tag scenarios