Skip to main content

Roo Code 3.35 Release Notes (Combined)

This release introduces metadata-driven subtasks that never lose their parent connection, plus native tool calling support for 15+ providers.

Roo Code v3.35 Release

Metadata-Driven Subtasks​

The connection between subtasks and parent tasks no longer breaks when you exit a task, crash, reboot, or reload VS Code (#9090). Subtask relationships are now controlled by metadata, so the parent-child link persists through any interruption.

Native Tool Calling Expansion​

Native tool calling support has been expanded to 15+ providers:

QOL Improvements​

  • Improved Onboarding: Simplified provider settings during initial setup—advanced options remain in Settings (#9709)
  • Cleaner Toolbar: Modes and MCP settings consolidated into the main settings panel for better discoverability (#9710)
  • Tool Format in Environment Details: Models now receive tool format information, improving behavior when switching between XML and native tools (#9661)
  • Debug Buttons: View API and UI history with new debug buttons (requires roo-cline.debug: true) (#9684)
  • Grok Code Fast Default: Native tools now default for xai/grok-code-fast-1 (#9717)
  • New Welcome View: Simplified welcome view with consolidated components for a cleaner, more consistent onboarding experience (#9741)
  • Simplified write_to_file Tool: The line_count parameter has been removed from the write_to_file tool, making tool calls cleaner and reducing potential errors from incorrect line counts (#9667)
  • OpenRouter Embeddings Provider Routing: Select specific routing providers for OpenRouter embeddings in code indexing settings, enabling cost optimization since providers can vary by 4-5x in price for the same embedding model (thanks SannidhyaSah!) (#9693)
  • CloudView Interface Updates: Cleaner UI with refreshed marketing copy, updated button styling with rounded corners for a more modern look (#9776)

Bug Fixes​

  • Parallel Tool Calls Fix: Preserve tool_use blocks in summary during context condensation, fixing 400 errors with Anthropic's parallel tool calls feature (thanks SilentFlower!) (#9714)
  • Navigation Button Wrapping: Prevent navigation buttons from wrapping on smaller screens (#9721)
  • Task Delegation Tool Flush: Fixes 400 errors that occurred when using native tool protocol with parallel tool calls (e.g., update_todo_list + new_task). Pending tool results are now properly flushed before task delegation (#9726)
  • Malformed Tool Call Fix: Fixed a regression where malformed native tool calls would cause Roo Code to hang indefinitely. Tool calls now proceed to validation which catches and reports the missing parameters properly (#9758)

Misc Improvements​

  • Model-specific Tool Customization: Configure excludedTools and includedTools per model for fine-grained tool availability control (#9641)
  • apply_patch Tool: New native tool for file editing using simplified diff format with fuzzy matching and file rename support (#9663)
  • search_and_replace Tool: Batch text replacements with partial matching and error recovery (#9549)
  • Better IPC Error Logging: Error logs now display detailed structured data instead of unhelpful [object Object] messages, making debugging extension issues easier (#9727)
  • Stealth Model Privacy: Models tagged with "stealth" in the Roo API now receive vendor confidentiality instructions in their system prompt, enabling white-label or anonymous model experiences (#9742)

Provider Updates​

  • Model Default Temperatures: Models can now specify their own default temperature settings. Temperature precedence is: user's custom setting → model's default → system default (#9734)
  • Roo Provider Native Tools: Models with the default-native-tools tag automatically use native tool calling by default (#9735)
  • LiteLLM Native Tool Support: All LiteLLM models now assume native tool support by default, improving tool compatibility (#9736)
  • App Version Tracking: The Roo provider now sends app version information with API requests for improved request tracking (#9730)
  • z.ai GLM Model Fix: Removed misleading reasoning toggle UI for GLM-4.5 and GLM-4.6 models on z.ai provider, as these models don't support think/reasoning data for coding agents (#9752)
  • Roo Provider Tool Compatibility: Improved tool conversion for OpenAI-compatible API endpoints, ensuring tools work correctly with OpenAI-style request formats (#9769)
  • MiniMax M2 Free Tier Default: MiniMax M2 model now defaults to the free tier when using OpenRouter (#9778)