Roo Code 3.23.17 Release Notes (2025-07-22)

This release introduces new AI providers, enhanced stability, and numerous quality-of-life improvements.

QOL Improvements

Jump to New Files: Added jump icon for newly created files, matching the experience of edited files (thanks mkdir700!) (#5738)
Apply Diff Error Messages: Added case sensitivity reminder when apply_diff fails, helping users understand matching requirements (thanks maskelihileci!) (#6076)
Context Condensing Prompt Location: Moved to Prompts section for better discoverability and persistent visibility (thanks SannidhyaSah, notadamking!) (#5279)
Todo List Tool Control: Added checkbox in provider settings to enable/disable the todo list tool (#6032)
MCP Content Optimization: Automatically omits MCP-related prompts when no servers are configured (#5889)
Git Installation Check: Shows clear warning with download link when Git is not installed for checkpoints feature (thanks MuriloFP!) (#5920)
Configurable Eval Timeouts: Added slider to set evaluation timeouts between 5-10 minutes (#5865)

Terminal Output Limits: Fixed critical crashes caused by terminal commands outputting extremely long lines. Now properly limits output to prevent context explosion (thanks robottwo!) (#5777)
API Error Messages: Fixed issue where all API failures showed as "API Request Cancelled". Now displays correct error messages for better troubleshooting (thanks MDean-Slalom, daniel-lxs!) (#6025)
AWS Bedrock Token Counting: Fixed inaccurate token counts and false context window errors for Bedrock users (thanks daniel-lxs, fedorbass!) (#6019)
Export During API Operations: Restored ability to export, share, and copy content while API is running (thanks MuriloFP, KJ7LNW!) (#5849)
Symlinked Rules Files: Fixed sorting of symlinked rules files to use symlink names, allowing proper ordering with numbered prefixes (thanks dsent!) (#5903)
Global Mode Export: Fixed issue where exporting global modes didn't include associated rules files (#5837)
Vitest Command: Added missing run parameter to prevent tests hanging in watch mode (thanks KJ7LNW!) (#5991)
Evaluation System: Fixed evaluation timeout handling that was broken in a previous update (#6065)
Cloud Session Handling: Improved session error handling to prevent unnecessary logouts during service issues (#6051)

Mistral Embedding Provider: Codebase indexing gets a major upgrade with Mistral as a new embedding provider, offering superior performance at no cost. Simply select Mistral's codestral-embed model in your embedding settings for better code understanding and more accurate AI responses (thanks SannidhyaSah, shariqriazz!) (#5946)
Qwen3-235B Model: Added support for Qwen3-235B-A22B-Instruct-2507 with massive 262K token context window on Chutes AI (thanks apple-techie!) (#6052)
Moonshot AI: Added Moonshot as a new AI provider option (thanks CellenLee!) (#6046)
Llama 4 Maverick: Added Llama 4 Maverick model (128K context) to Google Vertex AI in US-EAST5 region (thanks MuriloFP, konstantinosbotonakis!) (#6023)

Merge Resolver Mode: Added intelligent merge conflict resolution mode that analyzes commit history to make smart decisions about which changes to keep (thanks daniel-lxs!) (#6090)
Token Fallback Logic: Improved sliding window with consistent 8192 token fallback for more predictable context handling (#5993)
Todo List Clarification: Updated documentation to clarify that todo list tool is for complicated tasks, not just multi-step ones (#5926)