dileepyavan

This commit **enhances the chat feature's response stringification** by prioritizing display commands and presentation overrides for terminal commands. Specifically, the `AbstractResponse.toString` method within the **chat model** (`chatModel.ts`) now intelligently selects the most user-friendly representation when converting terminal commands to strings. This **feature improvement** ensures that chat responses containing commands are more readable and contextually appropriate, directly impacting how users perceive command output in the chat interface. New test cases have been added to `chatModel.test.ts` to validate this updated logic.

This commit **reverts** a previous change that introduced `responseOutputMessageId` into various requests within the **Copilot extension**, especially when message compaction was enabled. It **removes** the `responseOutputMessageId` property from the `IToolCallRound` interface and `ToolCallRound` class, and eliminates its handling and rendering from assistant messages in `languageModelAccessPrompt.tsx`. The commit also refactors phase data handling by removing this ID from data structures and simplifying API request processing in `responsesApi.ts`, along with associated test cases. This **bug fix** simplifies the internal message and phase data model across the **Copilot conversation and endpoint platforms**, ensuring correct behavior by removing an identifier that was causing issues.

This commit significantly enhances the **security** of **chat agent tools** by introducing **network domain sandboxing** for commands executed via `runInTerminalTool`. Users will now receive an **explicit notification and be prompted for confirmation** if a command attempts to access a network domain not included in the allowed list, preventing unauthorized network activity. This **new capability** also involves a **refactoring** of the sandbox setting, moving from `chat.tools.terminal.sandbox.enabled` to a more general `chat.agent.sandbox` to reflect its broader application across agents.

This commit introduces a **new capability** to attach a `responseOutputMessageId` to uniquely identify outgoing messages, specifically targeting **phased assistant messages** and those generated during **tool calling**. It integrates this identifier across the **Copilot extension's language model access**, **tool calling loop**, and **UI rendering components** to ensure proper tracking and display. The `responseOutputMessageId` is now captured, propagated, and rendered throughout the message lifecycle, from generation to display, enhancing the system's ability to manage complex conversational flows. This involves significant updates to the **platform's endpoint and networking layers** for proper encoding, decoding, and API communication of these new message IDs.

This commit performs a significant **refactoring** of the **chat agent sandboxing configuration**, renaming the setting from `chat.tools.terminal.sandbox.enabled` to the more general `chat.agent.sandbox`. This change reflects that sandboxing is a core concept for **AI agents** rather than being specific to terminal interactions. The work involves **deprecating the old setting**, introducing the new one, updating all internal references across **terminal chat agent tools**, and ensuring **configuration migration** for existing users. Additionally, **telemetry events** are updated to align with the new, broader scope, improving the consistency and future extensibility of agent-related features.

This commit introduces a **new capability** to proactively detect missing dependencies required for **agent sandboxing** on **Linux** systems, offering users the option to install them before execution. It establishes a new `ISandboxHelperService` across main, browser, and node environments to perform these prerequisite checks and manage their status. The **chat agent tools** and **terminal sandbox service** are updated to leverage this new service, ensuring sandboxed operations are only attempted when dependencies are met. Furthermore, a new `ChatMissingSandboxDepsConfirmationSubPart` is integrated into the **chat UI** to inform users about missing dependencies and facilitate their installation, significantly improving the robustness and user experience of agent execution.

This commit **fixes a bug** in the **terminal chat agent tools** by ensuring that the `$TMPDIR` environment variable is correctly preserved when retrying terminal commands outside the sandbox. The `commandLineSandboxRewriter` and `terminalSandboxService` modules were updated to properly handle the `requestUnsandboxedExecution` flag and conditionally preserve `$TMPDIR`. This **enhances the reliability** of unsandboxed command execution, particularly for commands initiated by AI chat agents, by preventing issues related to incorrect temporary directory paths. The change improves the overall **stability and correctness** of terminal command execution in scenarios requiring unsandboxed retries.

This commit introduces **two new experimental prompt variations** for the **Copilot agent's GPT-5.4 model**: a concise prompt and a large prompt. This **new capability** allows for testing different prompting strategies by conditionally enabling these prompts via new configuration flags, which are defined in `configurationService.ts` and checked by `chatModelCapabilities.ts`. The **Copilot extension** is primarily affected, with updates to `allAgentPrompts.ts` and `gpt54Prompt.tsx` to register and resolve these new prompts, alongside new localization strings in `package.nls.json`. This work enables the evaluation of alternative prompt designs to potentially enhance GPT-5.4's performance and behavior within the agent.

This commit **fixes** an issue with **terminal sandbox temporary directory scoping** within the **chat agent tools** feature. It modifies the `terminalSandboxService.ts` to ensure temporary directories are uniquely identified per window, preventing potential conflicts in multi-window or remote environments. The change involves updating the command wrapping logic and introducing a new helper `_getSandboxWindowTempDirName` to generate window-specific temporary directory paths. This **maintenance** work improves the isolation and reliability of the **terminal sandbox**, with corresponding updates to test cases in `terminalSandboxService.test.ts`.

This commit **fixes a rendering bug** affecting **tool invocation messages** within the **Chat UI's sandbox mode**. Previously, markdown syntax tokens were not correctly escaped, causing malformed output in the `ChatTerminalToolProgressPart`. The fix modifies `src/vs/workbench/contrib/chat/browser/widget/chatContentParts/toolInvocationParts/chatTerminalToolProgressPart.ts` to properly escape these tokens and now uses `MarkdownString` for the collapse button label. This ensures that users receive correctly formatted and readable tool invocation feedback, enhancing the overall user experience in the chat interface.

This commit **fixes a critical security vulnerability** by preventing sandboxed commands within the **terminal sandbox service** from improperly utilizing the `/tmp` directory. It **modifies the core logic** in `terminalSandboxService.ts` and `runInTerminalTool.ts` to ensure proper isolation for commands executed by features like **chat agents**. The work also includes an **upgrade to the sandbox runtime** for enhanced security and stability, with comprehensive **test updates** to validate the corrected temporary directory handling. This significantly improves the reliability and security posture of sandboxed terminal command execution.

This commit **enhances the terminal output** by **integrating analyzer messages** directly into the background terminal. It introduces a new private method, `_getOutputAnalyzerMessage`, within the `runInTerminalTool.ts` file of the **terminal chat agent tools contribution**. This **new capability** ensures that users receive richer feedback, as analyzer messages are now displayed in the terminal for both idle and running commands. The change improves the contextual information available during command execution within the **workbench terminal**.

This commit **fixes an issue** related to the display and handling of **unsandboxed terminal confirmations** within the **Terminal Contrib** module. Specifically, it modifies the condition in `src/vs/workbench/contrib/terminalContrib/chatAgentTools/browser/tools/runInTerminalTool.ts` by removing `requiresUnsandboxConfirmation` from a primary check, which was causing incorrect confirmation window behavior. This ensures that **allow-list actions** are properly presented for user approval when commands are executed in an **unsandboxed terminal environment** via **chat agent tools**. The change improves the reliability and security of user interactions by guaranteeing correct authorization prompts for potentially sensitive operations.

This commit introduces a **new capability** to manage user confirmation for commands suggested by the **Terminal Chat Agent Tools** that are intended to run outside of a sandbox. It adds a `requiresUnsandboxConfirmation` option to the `ICommandLineAnalyzerOptions` interface, allowing specific commands to be **allow-listed** for automatic approval. The `run` method in `runInTerminalTool.ts` now passes this option to the `commandLineSandboxAnalyzer.ts`, which then conditionally sets `forceAutoApproval` to bypass the unsandboxed terminal confirmation. This enhancement provides more granular control over security prompts, improving the user experience for trusted commands while maintaining necessary safeguards.

This commit **restores** the display of the **sandbox lock message** for **background commands** executed through the **terminal tool**. It re-enables the necessary display logic within the `execute` function of `src/vs/workbench/contrib/terminalContrib/chatAgentTools/browser/tools/runInTerminalTool.ts`. This **bug fix** addresses a recent regression, ensuring users are properly informed when background operations encounter sandbox restrictions. The change improves user feedback and security awareness within the **terminal contribution's chat agent tools**.

This commit **enhances the user experience** by providing clearer feedback when chat agent commands are executed in a sandbox. It introduces a **new capability** to display a specific label within the **Chat terminal tool progress part** (`ChatTerminalToolProgressPart`) to indicate sandbox execution. The `runInTerminalTool` is updated to generate this sandbox-specific invocation message, improving transparency for **chat agent tool invocations**. New tests have been added to ensure the correctness of this sandbox messaging.

This commit **resolves a display bug** in the **chat agent tools' terminal integration** where commands executed within a **sandbox environment** were not shown in the tool's output. It introduces an `isSandboxWrapped` flag, propagated through `IChatTerminalToolInvocationData` and `ICommandLineRewriterResult`, which is set by the `commandLineSandboxRewriter` when a command is wrapped. The `runInTerminalTool` then utilizes this flag to **conditionally adjust the display**, ensuring that users can now see the actual command being executed in sandbox mode, thereby **improving transparency and user experience**.

This commit **enhances the terminal sandboxing mechanism** by automatically including **workspace folders** in the list of allowed write paths. It introduces a new method `_updateAllowWritePathsWithWorkspaceFolders` in `src/vs/workbench/contrib/terminalContrib/chatAgentTools/common/terminalSandboxService.ts` to integrate workspace folder changes into the sandbox configuration. This **new capability** ensures that the **terminal sandbox**, particularly for **AI chat agent tools**, can correctly interact with files within the user's workspace. **New tests** have been added in `src/vs/workbench/contrib/terminalContrib/chatAgentTools/test/browser/terminalSandboxService.test.ts` to validate the refreshing of `allowWrite` paths when workspace folders change.

This commit **reverts** previously introduced **platform-level sandbox manager changes**, effectively **removing** the dedicated `sandboxHelperService` and its associated common interfaces, IPC channels, and implementations from the platform. This **cleanup** impacts the **main application**, **server services**, and **terminal contributions**, which no longer import or register the external sandbox helper. Consequently, the **Terminal Chat Agent Tools**' `ITerminalSandboxService` has been **refactored** to manage sandbox configuration internally, eliminating its dependency on the removed external service. This **maintenance** effort streamlines the codebase by consolidating sandbox configuration logic within the specific feature that requires it, simplifying the overall platform architecture.

This commit implements a **fix** for the **terminal sandboxing** feature, specifically addressing issues with the `rgpath` configuration. It modifies the `SandboxHelperService` in `src/vs/platform/sandbox/node/sandboxHelperService.ts` to correctly configure ripgrep, including adjusting its settings and introducing a new method, `_getPathWithRipgrepDir`, to manage the `PATH` environment variable. This ensures that ripgrep can be properly located and executed within sandboxed terminal environments, resolving functionality problems and improving the reliability of sandboxed operations.