generate_early
Added
- clear_thinking setting for GLM-4.7 model thinking control
- enable_pack_agents config option (defaults to false)
Changed
- Replace google-genai SDK with direct httpx implementation for Gemini
- Add proactive token refresh and malformed JSON repair
- Flatten union types for Gemini/Claude function schema compatibility
- Streamline dependencies with pydantic-ai-slim and remove dev tools
Removed
- google-genai dependency
- logfire dependency
- docs/PROACTIVE_TOKEN_REFRESH.md
generate_early
Added
- Add MCP server support and streaming to sub-agents with conditional handling based on configuration
Changed
- Refactor event stream handler into dedicated module for improved separation of concerns
- Consolidate MCP server management by removing /mcp add command and improving configuration handling with environment variable expansion
Removed
- Remove DBOS durable execution support including all integration code, configuration, and dependencies
Fixed
- Resolve Windows console input issues after prompt_toolkit exits with safe_input utility
- Prevent event loop blocking in MCP server start/stop operations by replacing blocking calls with async task scheduling
Jan 2, 2026
Added
- Implement real-time markdown streaming with termflow library for immediate visual feedback
- Add automatic OAuth token refresh for Claude Code integration with expiry detection and retry logic
Changed
- Update Cerebras GLM model from version 4.6 to 4.7 in models configuration
- Add truecolor terminal detection and warning system with comprehensive environment variable checks
Fixed
- Fix broken Kittylog changelog links in README documentation
Jan 1, 2026
Added
- Add environment variable CODE_PUPPY_SKIP_TUTORIAL to skip interactive onboarding wizard
- Add interleaved thinking support for Claude models via anthropic-beta header
Changed
- Refactor proxy configuration into centralized ProxyConfig dataclass and add background process support for shell commands
- Replace verbose file-by-file listings with compact directory summaries showing file count, subdirectory count, and total size
Fixed
- Fix JSON serialization issues in Antigravity model by converting binary data to base64 strings
- Fix parallel function call handling for Gemini API by merging consecutive same-role messages and ensuring required thoughtSignature field
Dec 31, 2025
Added
- Introduce AntigravityModel for enhanced thinking signature support with custom HTTP client and streaming response handling
- Implement modular Antigravity OAuth plugin with multi-account management, load balancing, and automatic failover
Changed
- Refactor HTTP utilities to remove broken TenacityTransport and implement robust async client with retry logic
- Update MOTD to New Year 2026 version with OAuth subscription reminders
- Enhance share_your_reasoning function to accept both string and list formats for next_steps with proper enumeration
Removed
- Remove obsolete upgrade_to_standard_tier functionality and associated CLI commands from Antigravity plugin
- Remove older Gemini models (2.5, 2.0, 1.5 series) and GPT-OSS model from Antigravity plugin configuration
Fixed
- Resolve HTTP transport issues by removing broken TenacityTransport and deleting failing tests