* The reckoning: Update all the vscode.proposed.*.d.ts files
and run the update on post install
* Improve proposed update and check
- Pin it to a commit
- Require that update be run manually, not as post install
- Add PR check
---------
Co-authored-by: João Moreno <joaomoreno@users.noreply.github.com>
* Background - fix changes paths in the working set
* Pull request feedback
---------
Co-authored-by: João Moreno <joaomoreno@users.noreply.github.com>
* Plan mode for background agents
* Updates
* Fix cache directory name and correct markdown syntax in PlanAgentProvider
* feat: add Plan Mode support for Copilot CLI agents
* fix: update configuration check for CLI Plan Mode in ChatSessionsContrib
* Ensure we log the error
* fix tests
* Remove Ask User Question tool in favor of Core tool
Fixes https://github.com/microsoft/vscode/issues/296134
* feedback
* remove duplicate & restore toolset
* feedback
* revert change in anthropic.ts because ask_questions is what is used in Claude
* Clean up MCP Gateway in Claude after 10min of idle
* Create a gateway per-session
* clean up after 10min in Claude by bringing it in to the session
* Have MCP Service use resource and address feedback
* Implement MCP gateway for Claude agent
This will hook up any MCP servers installed via VS Code and available via VS Code to Claude.
It think there should be some UX work in Core to enable/disable tools, but from an extension perspective this should be good.
ref https://github.com/microsoft/vscode/issues/287660
* feedback
* Add support for local repository memory and update telemetry events
* Update memory command labels for clarity in the UI
* update test
* Add repository memory section to snapshot tests for clarity
* Fix slash command handler not taking the command into consideration
This broke at some point, not sure when... but now it's plumbed through.
* feedback
* Initial plan
* Add tool-calling reinforcement to Gemini prompts to prevent hallucinated tool calls
The Gemini Pro model was hallucinating tool calls (describing them in text
instead of invoking them) after 2-3 turns. This adds explicit instructions
to both Gemini system prompts and the reminder instructions to reinforce
that tools must be invoked through the tool-calling mechanism, not described
in prose.
Co-authored-by: vijayupadya <41652029+vijayupadya@users.noreply.github.com>
* update prompts
* add br ending
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: vijayupadya <41652029+vijayupadya@users.noreply.github.com>
In general you should not call the refresh handler like this yourself because it won't show anything in the UI (it's fine to call the actual implementation that does the refresh). I'm going to update the api so that there is no confusing `refreshHandler` property toi avoid this confusion
* Initial plan
* Add usage handler mechanism for Claude context window widget
- Add UsageHandler type and methods to IClaudeSessionStateService
- Set usage handler in claudeChatSessionContentProvider request handler
- Wire up usage reporting in ClaudeStreamingPassThroughEndpoint
- Extract usage from AnthropicMessagesProcessor completion and call handler
Co-authored-by: TylerLeonhardt <2644648+TylerLeonhardt@users.noreply.github.com>
* Add unit tests for usage handler in ClaudeSessionStateService
- Add tests for getUsageHandlerForSession
- Add tests for setUsageHandlerForSession
- Verify handler preservation across state changes
- Verify handler can be called correctly
Co-authored-by: TylerLeonhardt <2644648+TylerLeonhardt@users.noreply.github.com>
* don't do the separate cache section
* a test
* include compaction
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: TylerLeonhardt <2644648+TylerLeonhardt@users.noreply.github.com>
Co-authored-by: Tyler Leonhardt <tyleonha@microsoft.com>
