Martin Aeschlimann

This commit introduces the `IHookDiscoveryInfo` interface to summarize information about discovered hooks within the **Chat prompt service**. It then **refactors** the `promptsServiceImpl.ts` to leverage this new interface for **caching and discovery logic** of hooks, improving internal efficiency. This **maintenance** effort also enhances error reporting for files skipped during hook discovery, contributing to a more robust and reliable prompt processing system.

This commit introduces **caching for agent and slash command discovery information** within the **chat prompt service**, specifically for `ISlashCommandDiscoveryInfo` and `IAgentDiscoveryInfo`. It **refactors** the `promptsServiceImpl.ts` to store this data, improving performance by avoiding redundant computations when resolving chat prompts. New interfaces and a `sanitizePromptDiscoveryInfo` function are added in `promptsService.ts` to support this optimization. This **maintenance** work enhances the efficiency of the chat feature, potentially leading to a more responsive user experience when interacting with agents and slash commands.

This commit provides a **bug fix** addressing an issue where **variable references in chat prompts and slash commands** were only partially replaced, leaving trailing characters. The underlying problem was an incorrect calculation of the reference's length, which is now resolved by introducing and utilizing a `fullLength` property for `IBodyVariableReference` during **prompt file parsing**. This correction impacts the **chat widget's display**, the **prompts service's processing**, and the **chat sessions' slash command resolution**, ensuring accurate range calculations for replacement. Consequently, all **variable references** will now be **fully and correctly replaced**, improving the reliability and user experience of the **chat feature**.

This commit **renames** the default "VS Code Dark" theme to "**Dark 2026**" (and implicitly "VS Code Light" to "Light 2026") as part of a **maintenance refactoring**. It updates theme constants like `COLOR_THEME_DARK` within the **workbench theme service** (`src/vs/workbench/services/themes/common/workbenchThemeService.ts`) and adjusts the default `workbench.colorTheme` setting in the **configuration contribution**. The `migrateThemeSettingsId` function was enhanced to ensure **seamless migration** for users with the old theme IDs, preventing theme resets upon update. Localized theme labels in `extensions/theme-defaults/package.nls.json` are updated, and relevant tests for theme ID migration were adjusted to reflect these changes.

This commit implements **robust error handling** within the `_autoAttachInstructions` method of the **Chat widget** in `chatWidget.ts`. This **bug fix** prevents the chat interface from becoming unresponsive or blocked if the background process of collecting contextual instructions encounters a failure. By ensuring that instruction collection errors are gracefully handled, the change significantly enhances the **stability and user experience** of the chat feature, allowing users to continue interacting without interruption.

This commit **fixes a critical bug** in the **Chat feature** where the send button would become non-responsive in existing workspaces after enabling parent repository customizations. The issue was traced to an incorrect loop within the `promptFilesLocator.ts` utility, specifically in the `findParentRepoFolders` function. The fix **refactors the path traversal logic** to correctly handle filesystem roots, user home directories, and previously visited folders, thereby preventing infinite loops or incorrect path resolution. This ensures the **Chat interface remains responsive and stable** when interacting with complex repository structures.

This commit performs a **maintenance refactoring** within the **Chat feature's prompt syntax and service components** to **simplify type definitions**. It **removes the `ExtensionAgentSourceType` enum and the `IResolvedPromptFile` interface**, consolidating prompt file source tracking into a unified `PromptFileSource` enum. This change impacts how **prompt files and agent skills are discovered, registered, and managed**, requiring updates to functions like `findAgentSkills` and various registration logic within the `promptsService` and `chatModes` modules. The refactoring aims to create a more consistent and streamlined type system for prompt file handling.

This commit introduces a **refactoring** within the **Copilot extension's prompt file handling** to prevent unintended validation triggers. Specifically, the `PromptsServiceImpl` now utilizes `IFileSystemService` instead of `openTextDocument` for reading prompt file content. This **internal implementation change** ensures that accessing prompt files does not inadvertently activate document validations, improving the stability and efficiency of **prompt file loading**. Tests for `copilotCLIChatSessionParticipant` were also updated to accommodate this dependency change, ensuring the **Copilot extension** functions correctly without unnecessary side effects.

This commit **refactors** the **PromptsService** to **cache prompt discovery information** and **simplify the `IPromptDiscoveryInfo` interface** within the **Chat feature**. It introduces a `promptPath` property for more consistent prompt identification, streamlining the internal prompt discovery and caching logic in `promptsServiceImpl.ts`. This architectural improvement significantly **enhances the performance** of prompt lookups by avoiding redundant discovery operations. Consequently, the prompt debugging contribution and related test cases are updated to reflect these new data structures and access patterns.

This commit performs a significant **refactoring** of the **Chat prompt management system** by completely removing all integration and references to internal prompt files and `ChatInternalCustomizations`. The `PromptsService` implementation, chat mode source data types, and the `ChatContentMarkdownRenderer` are updated to no longer recognize or process these internal prompt mechanisms. This change streamlines the **Chat feature's** prompt handling, eliminating a deprecated or internal storage mechanism and simplifying the codebase related to prompt discovery, rendering, and mode definition.

This commit performs a **refactoring** of the **Chat Prompts Service** to **remove unnecessary change event emissions**, enhancing its efficiency and correctness. It simplifies event handling within `CachedPromise` in `src/vs/workbench/contrib/chat/common/promptSyntax/service/promptsServiceImpl.ts`, preventing redundant `onDidChange` events for various prompt types like instructions, skills, custom agents, and slash commands. Additionally, the `promptFilesLocator` is updated to explicitly fire events only when configuration or workspace folder changes occur, removing an unnecessary parameter. This **maintenance** work ensures more predictable behavior and potentially minor performance improvements for the **prompt syntax subsystem**, with updated tests reflecting these changes.

This commit introduces a **performance improvement** and **refactoring** to the **Chat feature's prompt file validation** by ensuring that problems are only shown for files currently open in the editor. It refactors the generic `PromptValidatorContribution` and `ModelTracker` from `src/vs/workbench/contrib/chat/common/promptSyntax/languageProviders/promptValidator.ts` into a new browser-specific implementation within `src/vs/workbench/contrib/chat/browser/promptSyntax/promptFileContributions.ts`. This change optimizes resource usage by preventing unnecessary validation of closed prompt files, leading to a more efficient and responsive user experience within the **Chat prompt syntax** subsystem.

This commit introduces a **new warning** within the **chat prompt validator** to inform users when they attempt to utilize **subagents** (indicated by `disable-model-invocation: false`) without having the `chat.customAgentInSubagent.enabled` setting enabled. This **feature enhancement** improves the **Chat feature's user experience** by providing immediate feedback on unsupported configurations for advanced prompt syntax. It guides users to enable the required setting, preventing confusion and ensuring proper usage of **subagent functionality**. The change includes a new helper function `isCustomAgentInSubagentEnabled` and corresponding test updates.

This commit introduces the **new feature** of **parsing and resolving slash commands** within **chat sessions**. It implements the core logic in `src/vs/workbench/contrib/chat/browser/chatSessions/chatSessions.contribution.ts` to interpret user input containing slash commands, integrating with existing prompt and tool services to handle attached context. This enhancement significantly improves the interactivity of the **chat experience** by allowing users to trigger specific actions or provide structured input directly through commands. The `src/vs/workbench/contrib/chat/common/requestParser/chatRequestParser.ts` module is updated to export the necessary regular expression for command parsing.

This commit **refactors** the **chat subagent invocation mechanism** by replacing the `chat.subagents.maxDepth` configuration setting with a new boolean `chat.subagents.allowInvocationsFromSubagents`. This **feature change** simplifies the user control over subagent nesting from a numerical depth to a simple on/off switch within the **chat configuration**. The `RunSubagentTool` is updated to respect this new setting, introducing a **hardcoded maximum nesting depth of 5** when subagent invocations are enabled, ensuring a controlled environment. Test cases for `RunSubagentTool` have also been updated to validate this new behavior, impacting the overall stability and predictability of subagent interactions.

This commit addresses **log spam** within the **Chat** feature's **agent skill discovery and validation** process. It **reduces log verbosity** by downgrading several `warn` level messages to `debug` level in `src/vs/workbench/contrib/chat/common/promptSyntax/service/promptsServiceImpl.ts`. Specifically, messages related to `truncateAgentSkillName`, `truncateAgentSkillDescription`, and `computeSkillDiscoveryInfo` will now only appear in debug logs. This **maintenance fix** improves the clarity of application logs, making it easier for developers to identify critical issues without being overwhelmed by non-critical warnings about duplicate agent skill names.

This commit **resolves a parsing bug** within the **Copilot extension's agent planning mechanism** by correcting the format of a tool identifier. Specifically, it adds a crucial space character after the `#tool:vscode/memory` string in `extensions/copilot/src/extension/agents/vscode-node/planAgentProvider.ts`, which was causing the system to report an "Unknown tool or toolset" error. This **bug fix** ensures that the **`vscode/memory` tool is correctly recognized and utilized** by the Copilot agent, preventing misinterpretations and allowing for proper execution of plans involving this tool. The change directly improves the reliability of agent operations within the extension.

This commit implements a **bug fix** within the **Copilot extension** to prevent the attachment of empty custom instruction files. It modifies the `customInstructions.tsx` module to **trim the content of custom instructions** and, if the content is empty after trimming, it explicitly returns `undefined` to signify no instruction. This **improves the robustness** of the custom instructions feature by ensuring that only meaningful content is processed. Consequently, users will experience a more reliable and cleaner interaction with Copilot's custom instruction capabilities, avoiding potential issues from empty inputs.

This commit **fixes a critical issue** where **Copilot's prompt instruction files and the customizations index were not correctly utilized for read access**. The **bug fix** involves a significant **refactoring of the internal logic** across the **Copilot extension's prompt and instruction handling system**. Specifically, it replaces imprecise checks like `isPromptInstruction` and `isPromptInstructionText` with dedicated and accurate identifiers such as `isInstructionFile` and `isCustomizationsIndex`. This ensures that user-defined prompt instructions and the customizations index are properly recognized and applied in features like CLI prompt resolution, code editing intents, chat variable rendering, and custom instruction display, thereby improving the reliability of Copilot's contextual understanding.

This commit introduces **support for nested subagents** within the **chat system**, allowing subagents to invoke other subagents up to a configurable maximum depth. It adds a new configuration setting, `chat.subagents.maxDepth`, to control this behavior and prevent infinite recursion. The `RunSubagentTool` is updated to track and enforce the nesting depth, disabling further subagent invocation if the maximum is reached. This **new capability** enhances the flexibility and power of agent interactions by enabling more complex, hierarchical agent workflows.