Zhichao Li

This commit introduces **telemetry for OpenTelemetry (OTel) feature adoption tracking** within the Copilot extension. It adds two new events: `otel.activated` to capture OTel configuration at startup, including a new `enabledVia` field indicating how OTel was enabled, and `otel.exportAgentTracesDB` to track agent trace database exports. This **new capability** provides crucial insights into OTel usage patterns and configuration methods, primarily affecting the **OTel configuration resolution** and **telemetry reporting** subsystems. Additionally, it includes a **refactoring** to extract the `DEFAULT_OTLP_ENDPOINT` constant, improving maintainability across the `otelConfig.ts` and `otelContrib.ts` files.

This **feature** enhances **OpenTelemetry (OTel)** and **GitHub telemetry** by adding **workspace metadata**, including Git branch, commit hash, remote URL, and relative file path. It introduces new OTel attributes and a `resolveWorkspaceOTelMetadata` helper to gather this context from the active repository. This crucial information is then injected into various **Copilot editing features**, enriching `invoke_agent` OTel spans and numerous `edit` and `survival` related telemetry events such as `emitEditFeedbackEvent` and `trackEditSurvival`. The change provides deeper contextual data for analyzing user interactions and the effectiveness of AI-assisted code modifications.

This commit primarily **fixes and refines OpenTelemetry (OTel) agent activity metrics** and related telemetry within the **Copilot extension**, addressing review feedback. It **corrects the `recordEditAcceptance` metric** to guard against invalid states and adjusts `incrementLinesOfCode` to properly handle optional language IDs, while also **removing the `incrementCommitCount` metric** for better data hygiene. Furthermore, the commit **introduces new telemetry events** such as `invokeCloudSession` for cloud session initiation, and `onChatEditingSessionEnd`, `reportInlineDoneEvent`, and `reportInlineEditSurvivalEvent` to capture more granular user interactions with chat and inline edits. The **monitoring documentation** (`agent_monitoring.md`) is updated to reflect these changes in metric names, descriptions, and attributes, ensuring accurate reporting and understanding of **Copilot's agent activity**.

This commit introduces comprehensive **OpenTelemetry (OTel) instrumentation** to monitor **agent activity and outcome signals** across various agentic features. It adds new OTel events, counters, and histograms to track user actions, edit acceptance/rejection, inline edit survival, agent summarization, PR creation via CLI, and cloud session invocations within modules like `userActions.ts`, `copilotCloudSessionsProvider.ts`, and several agent intents and tools. This **new capability** provides critical observability into the performance and user interaction with agentic features, enabling better analysis and future improvements to the **Copilot agent's quality and effectiveness**. The work also includes a **refactoring** to adopt a type-safe metrics API and a **bug fix** for TypeScript errors related to threading the OTel service.

This commit introduces a **new OpenTelemetry (OTel) SQLite store** (`OTelSqliteStore`) to persistently capture agent trace data, effectively **replacing the memory-leaking legacy trajectory logger**. It significantly **enhances observability and stability** by storing OTel spans in `agent-traces.db`, complete with automatic data retention policies (7-day TTL, 100-session cap) and robust session aggregation views. Key **bug fixes** ensure correct parent-child span hierarchy for agent tool invocations and CLI sessions, while a new `SqliteSpanExporter` properly integrates with the OTel pipeline, controlled by the `dbSpanExporter.enabled` setting. The **refactored data export mechanism** now provides a dedicated command (`exportAgentTracesDB`) to copy the raw SQLite database, primarily supporting external evaluation harnesses for ATIF conversion.

This commit **fixes a critical bug** in the **Copilot CLI session service** that caused unintended file creation on Windows. Previously, the OpenTelemetry (OTel) null sink would create a literal `NUL` file in the application's install directory, leading to large telemetry files and **update errors**. The change replaces a problematic platform-conditional string with `os.devNull`, ensuring the correct null device is used across all operating systems and **resolving the update failures** for Windows users.

This commit introduces a **new capability** and **bug fix** to significantly enhance the **debuggability of the background CopilotCLI agent**. Previously, SDK hook spans from the background agent lacked crucial input, output, and result details in the **Agent Debug Log panel**. This change implements a mechanism where the `copilotcliSession` now stashes hook event data, which the `copilotCliBridgeSpanProcessor` then uses to **enrich SDK hook spans** with comprehensive `copilot_chat.*` attributes. This ensures that the debug panel provides full visibility into the execution of background agent hooks, including a fix for displaying hook duration and formatting, thereby improving the **transparency and diagnostic capabilities** for developers.

This commit **fixes** an issue where the **Copilot CLI debug panel** would display empty content for prompts, responses, and tool arguments. It ensures **full content capture** by programmatically setting `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=true` within the `startCopilotCliSession` method before SDK initialization, overriding user settings only when OpenTelemetry is explicitly enabled. This significantly improves **observability and debugging** for GenAI interactions within the CLI, as users will now consistently see complete interaction details. Additionally, the **monitoring documentation** has been updated to clearly explain this content capture behavior and its implications for the agent debug log panel.

This extensive commit introduces **native OpenTelemetry (OTel) instrumentation** for the **Copilot CLI**'s background and terminal agents, significantly enhancing observability. It **refactors** the system to bridge SDK-native OTel spans directly to the **Agent Debug panel** via a new `CopilotCliBridgeSpanProcessor`, replacing previous synthetic span generation with a full, accurate trace hierarchy. This **new capability** ensures the debug panel displays detailed subagent, permissions, and tool call information, while also filtering external OTLP exports to include only standard GenAI operations. Numerous **fixes** address trace context propagation, debug panel routing via `chatSessionId`, and correct OTel environment variable handling for terminal sessions, alongside comprehensive **documentation updates** and **new tests**.

This commit introduces a **new feature** to the **Copilot extension's trajectory export functionality**, enabling **programmatic ATIF trajectory export** with enhanced filtering. It adds an `agentOnly` option to the `exportTrajectories` command, which filters out internal sessions and subagent interactions, ensuring only the main agent's trajectory tree is exported. Additionally, the commit allows skipping UI dialogs when a `savePath` is provided, facilitating seamless use in **headless environments**. This provides greater control and precision when exporting relevant trajectory data.

This commit significantly **improves the OpenTelemetry monitoring documentation** within the `copilot` extension by providing a more practical and user-friendly setup experience. It **replaces the environment variable-based Quick Start with a detailed, step-by-step guide for setting up the Aspire Dashboard**, and **converts all example configurations to VS Code settings JSON** for easier integration. The update also introduces new backend setup guides for **Jaeger** and **Langfuse**, expanding the scope of monitoring options. This **documentation enhancement** makes it easier for developers to configure and utilize OTel monitoring, particularly for those using VS Code, while retaining an environment variable reference table.

This commit introduces a **new feature** by adding the `github.copilot.chat.otel.flush` **VS Code command** to the **Copilot Chat extension's OpenTelemetry integration**. This command allows for the explicit and immediate flushing of all buffered OpenTelemetry spans, metrics, and events by calling `IOTelService.flush()`. The primary purpose is to ensure **data integrity** by preventing telemetry data loss, particularly critical `invoke_agent` root spans, when the VS Code process is terminated, especially within evaluation runtimes. This provides deterministic control over telemetry export, enhancing the reliability of collected data.

This commit **enhances user privacy** by **disabling the transmission of external telemetry** related to repository information for **non-internal users**. It specifically modifies the `copilot` extension's telemetry module, preventing the `sendBeginTelemetryIfNeeded` function from sending metadata such as the head commit hash and repository type. This **privacy enhancement** and **policy adjustment** ensures that sensitive repository details are no longer collected from general users. An accompanying test update verifies that no telemetry is sent for non-internal users, reflecting the updated data collection policy.

This commit implements a **security and privacy fix** and **performance optimization** within the **Copilot extension's telemetry and repository information handling**. It **removes the sensitive `repoId` from telemetry events sent to Microsoft for external users**, enhancing user privacy. Concurrently, it **optimizes the amount of repository information** provided to external users by gating detailed data behind an internal user check, ensuring they only receive lightweight metadata. The change also includes **refactoring** to deduplicate repository context resolution logic in `_resolveRepoContext` and updates tests to validate these new privacy and optimization behaviors.

This commit **enhances telemetry collection** within the **Copilot extension** by including `repoId` and `headCommitHash` in repository telemetry for all users. This **feature enhancement** provides more granular data for understanding repository usage patterns, while also **fixing a potential bug** by gracefully handling cases where the active repository might be null. The changes primarily affect the `repoInfoTelemetry.ts` module, ensuring more comprehensive and robust telemetry data is gathered for internal and external analysis.

This commit introduces a significant **performance improvement** to the **Chat** feature by deferring expensive repository diff capture. It refactors the `ChatRepoInfoContribution` to separate lightweight metadata collection, performed on the first chat message, from full file diff generation, which now only occurs when a user explicitly triggers a chat export. This change **reduces per-message overhead** by eliminating synchronous file I/O and diff computation during active chat sessions, addressing previous performance concerns and fixing a root cause of slow interactions. As a result, the **Chat** experience is smoother, with full repository context still available on demand for export.

Merge pull request #295253 from microsoft/frightened-cricket

This commit introduces **streaming support** for **Bring-Your-Own-Key (BYOK) model configurations** within the `copilot` extension. This **new feature** enhances the `BYOK` integration by allowing models to send responses incrementally, improving user experience. The `streaming` capability is added to the common `BYOK` model information resolution via `resolveModelInfo` and integrated into both the **Azure** and **Custom OpenAI** `BYOK` providers through their `toBYOKModelInfo` functions. This enables a more responsive and interactive experience for users leveraging their own LLM keys with these providers.

This commit **fixes and enhances request logging** for **Bring Your Own Key (BYOK) language models** by introducing a **correlation ID mechanism** to preserve and restore `CapturingToken` context across IPC boundaries. This ensures that requests from **Anthropic BYOK**, **Gemini Native BYOK**, and `CopilotLanguageModelWrapper` are correctly logged as children of their prompt items with accurate token usage, while also eliminating **duplicate logging** from `ExtensionContributedChatEndpoint`. Additionally, it includes a **critical bug fix** to prevent memory leaks in the `CapturingToken` correlation map by ensuring proper cleanup, and **refactors** the `requestLogTree.ts` export logic for simplification. The overall impact is more **accurate, hierarchical, and leak-free request logging** for BYOK model interactions.

This change refines the chat feature's subagent request mechanism by removing a hardcoded fallback. This improves the robustness and flexibility of subagent interactions within the chat experience, likely preventing incorrect subagent identification in certain scenarios.