digitarald

This commit introduces **new OpenTelemetry (OTel) metrics** to enhance **telemetry for AI-assisted code editing** within the **Copilot extension**. It adds capabilities to track **edit acceptance** and **code survival** for both inline edits and chat-based code modifications, providing granular insights into user interaction and the longevity of AI-generated code. Specifically, new OTel attributes are defined in `genAiAttributes.ts`, recording functions like `recordEditAcceptance` and `recordEditSurvivalFourGram` are implemented in `genAiMetrics.ts`, and these are integrated into user action handlers in `userActions.ts`. This **feature** provides critical data for evaluating the effectiveness and impact of AI suggestions, with new tests ensuring the accuracy of these metrics.

This commit **refactors and simplifies the `chat.autoReply` feature** within the **Chat subsystem**. Instead of attempting to answer questions via a separate LLM call, `chat.autoReply` now **skips questions entirely**, providing a default "user not available" response, which significantly reduces complexity by removing ~450 lines of prompt engineering and parsing logic. This **simplification** primarily affects the `AskQuestionsTool` and the overall `chat.autoReply` mechanism. Furthermore, it **improves robustness** by implementing logic in `chatListRenderer` to skip any pending question carousels when `chat.autoReply` is enabled mid-session. Users will experience a more streamlined and less resource-intensive auto-reply experience.

This commit **fixes** a regression in the **Chat feature's `askQuestions` tool** by **restoring the default behavior** of `allowFreeformInput` to `true`. Previously, the tool implicitly defaulted to `false` when the model omitted this setting, preventing users from providing freeform text answers alongside options. This change ensures users can now provide freeform input by default, enhancing the flexibility and usability of the `askQuestions` tool. Additionally, the relevant schema and model descriptions are updated to accurately reflect this default setting.

This commit **refines the documentation guidelines** for the **Copilot extension**, specifically within its prompt assets. It updates the `init.prompt.md` to instruct the agent to inventory existing documentation and link to it, rather than duplicating content. Additionally, the `workspace-instructions.md` clarifies the "Link, don't embed" principle, guiding the agent to search for existing documentation and only inline agent-critical information. This **documentation update** aims to improve the Copilot agent's efficiency and accuracy by promoting the reuse of existing knowledge and reducing redundancy.

This commit **fixes a bug** related to the **Chat Handoff Widget's visibility**, ensuring it is correctly derived from the chat response's `modeInfo.response.mode` and persists across updates. The **Chat Model** is enhanced with a new `updateResponse` method, allowing dynamic modification and persistence of a response's `modeInfo` and triggering UI updates. This change improves the accuracy of the chat interface by correctly displaying the handoff widget based on the current response state, and prepares the **Chat session operation log** for similar updates.

This commit introduces a **new capability** to the **Copilot custom instructions panel** by adding a visual hint that informs users when they are operating within a **multi-root workspace**. This **enhancement** improves the user experience by providing crucial context directly within the UI, specifically within the `customInstructions.tsx` component. Furthermore, it ensures that **workspace folder information** is now included in attachment tags, which helps in scoping AI interactions more accurately. This change primarily affects the **Copilot extension's prompt system**, enhancing how it understands and responds to user prompts in complex project setups.

This commit delivers **follow-up fixes and enhancements** for the **v2 taxonomy** within the **Copilot extension's prompt categorization system**. It **introduces a `taxonomyVersion` field** to telemetry events, specifically for prompt categorization, to accurately distinguish between v1 and v2 data. The changes also **resolve prompt formatting issues** by adding trailing newlines to ensure proper boundary handling with `SafetyRules` and standardize bullet formatting for signals in the `promptCategorizationTaxonomy.ts` file. This work is primarily **maintenance and enhancement**, improving data collection and consistency for the new taxonomy.

This commit **significantly improves the presentation of chat questions** within the **Chat UI's question carousel**. It **refactors the rendering logic** in `chatQuestionCarouselPart.ts` to display a distinct header as the question title, with the full question message, now supporting **multi-line content and markdown**, rendered separately below. This **enhancement** provides a clearer and more structured user experience for complex questions, with corresponding styling updates in `chatQuestionCarousel.css` and test coverage in `chatQuestionCarouselPart.test.ts`.

This commit **fixes Out Of Memory (OOM) errors** that occurred during **large workspace chunking and embedding operations** within the Copilot extension. It addresses the issue by **bounding concurrency** in the workspace embedding indexing process, specifically within `workspaceChunkEmbeddingsIndex.ts`, and by **reducing the SQLite cache size** in `workspaceChunkAndEmbeddingCache.ts`. This **bug fix** significantly improves the stability and reliability of the Copilot extension when processing extensive codebases, preventing crashes during resource-intensive tasks like `indexAllWorkspaceFiles`.

This commit **enables custom agents for subagent tools by default** within the **Chat feature**. It updates the `ChatConfiguration.SubagentToolCustomAgents` setting in `src/vs/workbench/contrib/chat/browser/chat.contribution.ts` from `false` to `true`, and **removes its experimental tag**. This **enhancement** makes the custom agent capability for subagent tools readily available to all users without requiring manual activation of an experimental setting, thereby expanding the default functionality of the chat experience.

This commit **refactors** the **Copilot extension's configuration service** to prepare for experiment-based feature rollouts. Specifically, the `SwitchAgentEnabled` configuration setting is updated to support **experimentation (ExP)**, moving it from a simple boolean toggle to an experiment-driven control. This change enables the `switchAgent` functionality to be managed and deployed through A/B testing or staged rollouts, ensuring a more controlled and data-driven release process.

This commit **simplifies the Copilot agent architecture** by **removing the `PlanAgentExploreSubagentEnabled` feature flag** and making the **'Explore' subagent always active**. The `PlanAgentProvider` in `extensions/copilot/src/extension/agents/vscode-node/planAgentProvider.ts` is updated to unconditionally include and enable the 'Explore' subagent, eliminating previous conditional configuration checks. Consequently, the `switchAgentTool` in `extensions/copilot/src/extension/tools/vscode-node/switchAgentTool.ts` is refactored to remove its dependency on the configuration service for this subagent, streamlining its `constructor` and `invoke` methods. This **feature enhancement** ensures the **'Explore' subagent is consistently available** to all users, simplifying its management and usage within the Copilot extension.

This commit **introduces a new `memory-leak-audit` skill** for the **AI coding agents**, enhancing their analytical capabilities. It **adds a new feature** by defining this skill within the `.github/skills/memory-leak-audit/SKILL.md` file, which outlines explicit rules and examples for agents to identify and prevent memory leaks. This enhancement significantly improves the **code quality and reliability** capabilities of the **AI coding agents** by equipping them with specialized knowledge for memory management. The new skill will enable agents to perform more thorough audits, leading to more robust and efficient software.

This commit **fixes telemetry reporting for agent properties during chat handoff events** within the **Chat feature**. It introduces a new shared utility, `getModeNameForTelemetry` in `src/vs/workbench/contrib/chat/common/chatModes.ts`, which hashes user-defined or local agent mode names to ensure **privacy-compliant telemetry**. This **refactoring** standardizes how agent mode names are reported, specifically updating the `chatWidget.ts` to use this new function for handoff events and removing a local telemetry function from `chatExecuteActions.ts`. The change ensures sensitive agent configuration details are not directly exposed in telemetry.

Remove preview label from review agent settings (#3949)

Merge pull request #297031 from microsoft/digitarald/rural-rooster

This commit **integrates the plan prompt directly into the application's codebase**, making it an intrinsic part of the system rather than an external configuration. This **refactoring** simplifies the management and deployment of the **plan prompt**, ensuring consistent behavior across all planning operations. It primarily affects the **prompt generation and planning subsystems**, potentially improving reliability and reducing external dependencies for prompt retrieval.

This commit **updates the proposed instructions for API versioning**, refining the guidelines developers should follow when designing and implementing new API versions. It specifically impacts the **API development and documentation** processes, ensuring clearer and more consistent practices. This is a **documentation update** aimed at improving the overall clarity and maintainability of the project's API surface. The changes help standardize how new API versions are introduced, reducing potential confusion for both internal teams and external consumers.

This commit introduces a **fix** to improve **telemetry data consistency** and **joinability** across various Copilot features. It ensures that telemetry events originating from prompt categorization and panel requests can be accurately linked by introducing a `telemetryMessageId`. This unique ID is generated for the first turn of a user interaction within the **conversation participants** (`chatParticipants.ts`) and then propagated through the `ChatParticipantRequestHandler` and `ChatTelemetryBuilder`. The `promptCategorizer.ts` now accepts and uses this ID, aligning its telemetry events with the main request ID, which significantly enhances the ability to analyze user interaction flows by providing a unified identifier for related telemetry.

This commit performs a **refactoring** to **standardize the `userInvokable` property name** across the **Copilot extension's agent configuration**. It corrects a typo by renaming the property from `userInvokable` to `userInvocable` within the `AgentConfig` interface defined in `agentTypes.ts`. This change is consistently applied to the `buildAgentMarkdown` function and updated in the `BASE_EDIT_MODE_AGENT_CONFIG` and `BASE_EXPLORE_AGENT_CONFIG` definitions in `editModeAgentProvider.ts` and `exploreAgentProvider.ts`, respectively. The update ensures a consistent and accurate **agent configuration API**, improving the overall maintainability of the **Copilot agent system**.