This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and publish to npm yourself or setup this action to publish automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

[email protected]

Major Changes

• #5400 5a49128 Thanks @Sureshkumars! - Add Local review mode

Minor Changes

• #5234 796e188 Thanks @kevinvandijk! - Include changes from Roo Code v3.39.0-v3.41.2

  • Add button to open markdown in VSCode preview for easier reading of formatted content (PR #10773 by @brunobergher)
  • Fix: Add openai-codex to providers that don't require an API key (PR #10786 by @roomote)
  • Fix: Detect Gemini models with space-separated names for proper thought signature injection in LiteLLM (PR #10787 by @daniel-lxs)
  • Feat: Aggregate subtask costs in parent task (MCP management design is unreasonable #5376 by @hannesrudolph, PR #10757 by @taltas)
  • Fix: Prevent duplicate tool_use IDs causing API 400 errors (PR #10760 by @daniel-lxs)
  • Fix: Handle missing tool identity in OpenAI Native streams (PR #10719 by @hannesrudolph)
  • Fix: Truncate call_id to 64 chars for OpenAI Responses API (PR #10763 by @daniel-lxs)
  • Fix: Gemini thought signature validation errors (PR #10694 by @daniel-lxs)
  • Fix: Filter out empty text blocks from user messages for Gemini compatibility (PR #10728 by @daniel-lxs)
  • Fix: Flatten top-level anyOf/oneOf/allOf in MCP tool schemas (PR #10726 by @daniel-lxs)
  • Fix: Filter Ollama models without native tool support (PR #10735 by @daniel-lxs)
  • Feat: Add settings tab titles to search index (PR #10761 by @roomote)
  • Fix: Clear terminal output buffers to prevent memory leaks that could cause gray screens and performance degradation (#10666, PR #7666 by @hannesrudolph)
  • Fix: Inject dummy thought signatures on ALL tool calls for Gemini models, resolving issues with Gemini tool call handling through LiteLLM (PR #10743 by @daniel-lxs)
  • Fix: Add allowedFunctionNames support for Gemini to prevent mode switch errors (#10711 by @hannesrudolph, PR #10708 by @hannesrudolph)
  • Add settings search functionality to quickly find and navigate to specific settings (PR #10619 by @mrubens)
  • Improve settings search UI with better styling and usability (PR #10633 by @brunobergher)
  • Display edit_file errors in UI after consecutive failures for better debugging feedback (PR #10581 by @daniel-lxs)
  • Improve error display styling and visibility in chat messages (PR #10692 by @brunobergher)
  • Improve stop button visibility and streamline error handling (PR #10696 by @brunobergher)
  • Fix: Omit parallel_tool_calls when not explicitly enabled to prevent API errors (#10553 by @Idlebrand, PR #10671 by @daniel-lxs)
  • Fix: Encode hyphens in MCP tool names before sanitization (#10642 by @pdecat, PR #10644 by @pdecat)
  • Fix: Correct Gemini 3 thought signature injection format via OpenRouter (PR #10640 by @daniel-lxs)
  • Fix: Sanitize tool_use IDs to match API validation pattern (PR #10649 by @daniel-lxs)
  • Fix: Use placeholder for empty tool result content to fix Gemini API validation (PR #10672 by @daniel-lxs)
  • Fix: Return empty string from getReadablePath when path is empty (PR #10638 by @daniel-lxs)
  • Optimize message block cloning in presentAssistantMessage for better performance (PR #10616 by @ArchimedesCrypto)
  • Improve ExtensionHost code organization and cleanup (PR #10600 by @cte)
  • Fix: Ensure all tools have consistent strict mode values for Cerebras compatibility (#10334 by @brianboysen51, PR #10589 by @app/roomote)
  • Fix: Remove convertToSimpleMessages to restore tool calling for OpenAI-compatible providers (PR #10575 by @daniel-lxs)
  • Fix: Make edit_file matching more resilient to prevent false negatives (PR #10585 by @hannesrudolph)
  • Fix: Order text parts before tool calls in assistant messages for vscode-lm (PR #10573 by @daniel-lxs)
  • Fix: Ensure assistant message content is never undefined for Gemini compatibility (PR #10559 by @daniel-lxs)
  • Fix: Merge approval feedback into tool result instead of pushing duplicate messages (PR #10519 by @daniel-lxs)
  • Fix: Round-trip Gemini thought signatures for tool calls (PR #10590 by @hannesrudolph)
  • Feature: Improve error messaging for stream termination errors from provider (PR #10548 by @daniel-lxs)
  • Feature: Add debug setting to settings page for easier troubleshooting (PR #10580 by @hannesrudolph)
  • Chore: Disable edit_file tool for Gemini/Vertex providers (PR #10594 by @hannesrudolph)
  • Chore: Stop overriding tool allow/deny lists for Gemini (PR #10592 by @hannesrudolph)
  • Fix: Stabilize file paths during native tool call streaming to prevent path corruption (PR #10555 by @daniel-lxs)
  • Fix: Disable Gemini thought signature persistence to prevent corrupted signature errors (PR #10554 by @daniel-lxs)
  • Fix: Change minItems from 2 to 1 for Anthropic API compatibility (PR #10551 by @daniel-lxs)
  • Implement sticky provider profile for task-level API config persistence (#8010 by @hannesrudolph, PR #10018 by @hannesrudolph)
  • Add support for image file @mentions (PR #10189 by @hannesrudolph)
  • Add debug-mode proxy routing for debugging API calls (#7042 by @SleeperSmith, PR #10467 by @hannesrudolph)
  • Add Kimi K2 thinking model to Fireworks AI provider (#9201 by @kavehsfv, PR #9202 by @roomote)
  • Add image support documentation to read_file native tool description (#10440 by @nabilfreeman, PR #10442 by @roomote)
  • Add zai-glm-4.7 to Cerebras models (PR #10500 by @sebastiand-cerebras)
  • Tweak the style of follow up suggestion modes (PR #9260 by @mrubens)
  • Fix: Handle PowerShell ENOENT error in os-name on Windows (#9859 by @Yang-strive, PR #9897 by @roomote)
  • Fix: Make command chaining examples shell-aware for Windows compatibility (#10352 by @AlexNek, PR #10434 by @roomote)
  • Fix: Preserve tool_use blocks for all tool_results in kept messages during condensation (PR #10471 by @daniel-lxs)
  • Fix: Add additionalProperties: false to MCP tool schemas for OpenAI Responses API (PR #10472 by @daniel-lxs)
  • Fix: Prevent duplicate tool_result blocks causing API errors (PR #10497 by @daniel-lxs)
  • Fix: Add explicit deduplication for duplicate tool_result blocks (#10465 by @nabilfreeman, PR #10466 by @roomote)
  • Fix: Use task stored API config as fallback for rate limit (PR #10266 by @roomote)
  • Fix: Remove legacy Claude 2 series models from Bedrock provider (#9220 by @KevinZhao, PR #10501 by @roomote)
  • Fix: Add missing description fields for debugProxy configuration (PR #10505 by @roomote) @objectiveSee)

Patch Changes

• #5354 7156a35 Thanks @marius-kilocode! - Fixed broken image display in Agent Manager message list. Images pasted or attached to messages now render correctly as thumbnails in both user feedback messages and queued messages.

• #5373 cb41705 Thanks @sebastiand-cerebras! - Set default temperature to 1.0 for Cerebras zai-glm-4.7 model

• #5402 930931e Thanks @PeterDaveHello! - Improve zh-TW translations

• #5407 77cfa54 Thanks @chrarnoldus! - Add native tool calling support to Nano-GPT provider

• #5396 fdae881 Thanks @markijbema! - Revert "Using Kilo for work?" button in low credit warning, restore free models link

• #5364 5e8ed35 Thanks @huangdaxianer! - Removed forced context compression for volces.com

@kilocode/[email protected]

Patch Changes

• #5359 e4076d9 Thanks @PeterDaveHello! - Add missing openai-codex provider type definition

• #5400 5a49128 Thanks @Sureshkumars! - Add Local review mode

• #5374 1f06759 Thanks @PeterDaveHello! - Add openai-responses provider support in CLI config validation.