Copilot Chat v0.42: Gemini Fixes & Token Budget Boost
Copilot Chat v0.42 fixes Gemini 400 errors, boosts token budgets for better context, and adds prompt cache experiments. Solid stability release with CLI improvements.
TL;DR
- Fixed Gemini 400 errors from orphaned tool calls during conversation summarization
- Increased cursor jump token budget from 2K to 3K for better context handling
- Prompt cache experiment and session management improvements across the board
Fixed
- Gemini 400 errors — Orphaned tool calls during conversation summarization no longer crash with 400 errors
- PR detection — Sessions now properly detect pull requests
- Apply button — Background feature's Apply button now works correctly
- Git state issues — Fixed session-related git state tracking problems
- NES telemetry — Corrected stateless telemetry reporting for first edits from Neural Edit Suggestions
New
- Token budget increase — Cursor jump now allocates 3K tokens for current file context (up from 2K)
- Rate limit tracking — Added rate limit time monitoring
- Prompt cache experiment — Set up experiment for
prompt_cache_keyin Responses API - CLI session management — Enhanced CopilotCLISessionService with session ID management and fork session support
- Scheme exclusions — Expanded default scheme exclusions for better file filtering
- Temp file handling — Background indexing now creates temp files in global storage directory
Under the Hood
- Separated Copilot CLI implementation (controller vs old API)
- Enhanced CLI chat request handling with mode instructions
- Moved IGitService from vscode to vscode-node
- Improved output channel logging resilience during shutdown
- Refactored change computation for sessions with auto-commit disabled
- Claude Code now respects model picker settings
Update via VS Code extensions or run code --install-extension GitHub.Copilot-Chat.
Source: Copilot Chat
