Coder / Agent improvements

• Support instant request for chat initialized with task context and agent #15778
• Refine Coder prompt #15774
• feat: Provide a Task Context Management / Agent #15732
• Enact code actions when change set file created #15724

Chat View improvements

• Add chat retry button #15779
• Fix paste handling in chat input #15766
• Fix chat input widget not tracking branches and pending state #15727
• Show active chat in chat dropdown #15723
• fix: temporary scroll lock when user scrolls up in AI Chat #15683
• Use chatInputWidget in AskAI Input and fix keyboard scrolling #15781
• fix: pin the last mentioned agent, rather than the first one #15777

General features / improvements

• Chat Variable for All Open Editors #15775
• Improve image support for tool calls #15765
• Sanitize Task Context Labels #15762
• Add async initialization to ChangeSetFileElement #15761
• Replace YOLO terminology with "Always Allow"  #15756
• Add support for the new vscode.lm.registerMcpServerDefinitionProvider API #15755
• Make Workspace Search function more efficient / use less tokens #15743
• chore: update the Google AI SDK #15737
• enable Agent-to-Agent Delegation via Tool Calls #15736
• Consolidate Coder Prompts #15733
• feat: implement prompt caching for Anthropic #15731
• Allow to increase the number of retries before failing #15730
• Move AI Configuration View to main #15726
• Improve Ask AI feature #15725
• Update project info with styling information #15722
• Add ai tools for write through file changes #15717
• Work on improving code completion #15715
• Tool Call Confirmation UI #15714
• AppTester Agent via Browser Automation #15713
• Start Chat from Editor Context #15712
• Make sure to add openai stream options only in streaming models #15706
• Use relative path in search results in workspace function #15704
• Make workspace functions token efficient #15703
• Start Chat from Task Context #15700
• Workspace search function allows specifying file extensions #15699
• Add Widgets, Commands and Toolbar to project info. #15697
• Support instant request for chat initialized with task context and agent #15778
• Streamline Ask AI commands #15849
• Prompt Fragment view reset should only reset built-ins #15773
• Mentioning and agent needs to be at the beginning of prompt #15272
• Adapt all prompt ids to common schema #15880
• Reorganize architect prompts #15859
• Add fall back to default prompts if selected does not exist #15879
• Refine inline Ask AI chat input appearance #15885
• feat: theia-ai agent completion notification #15816
• feat: enhance mcp server management #15842

• Claude Opus 4 request fails because max token too high (64000 but should be 32000) #15785
• AI features accessible when AI is not enabled #15780
• Fix change set deletion #15759
• Fix Toolcall Spinner #15757
• Include the ai-ide stylesheet in the secondary window #15752
• fix: difficult to scroll up when auto scroll is enabled #15748
• Fix AI Configuration Widget Tab Bar Overflow #15739
• fix: filter out error messages from the messages send to the llm #15720
• fix: remove tool results from subsequential requests #15702
• fix: support more than 10 tool calls via OpenAI SDK #15696
• Fix bug that customizations are treated as variants #15846
• Send to AI Chat throws an error #15827
• Caching break code completion #15831
• feat(ai-core): handle toolCall errors gracefully instead of throwing #15800