Alexandru Dima

This commit **fixes an incorrect UI display** within the **chat thinking dropdown**, specifically for **terminal sandbox icons**. Previously, unsandboxed terminal commands would erroneously show a lock icon (`terminalSecure`) because the icon was set prematurely during tool registration. The fix ensures that the default icon for terminal tools is always `Codicon.terminal` in `runInTerminalTool.ts`, and then dynamically updates the icon based on the actual sandbox status (`isSandboxWrapped`) once `toolSpecificData` becomes available in `chatThinkingContentPart.ts`. This **improves user clarity** by accurately reflecting the sandbox status of terminal commands and **resolves a UI inconsistency** in the **chat agent tools** and **chat UI components**.

This commit **reverts** a previous change that introduced file paths for inline references within chat responses in the **Accessible View**. It specifically undoes modifications to the `provideContent` logic in `src/vs/workbench/contrib/chat/browser/accessibility/chatResponseAccessibleView.ts`, which was responsible for adding these details. This **revert** restores the prior behavior, meaning users of the accessible view will no longer see file paths associated with inline references in chat outputs. The change impacts the **accessibility experience** for chat interactions by removing this specific information from the accessible content.

This commit **enhances the terminal tool confirmation widget** within the **Chat** feature by displaying the model-provided reason for requesting unsandboxed execution. Previously, users lacked context on why a model required sandbox bypass for a terminal command. This **improvement** addresses a user experience gap, providing crucial information to enhance transparency and security awareness. The change specifically modifies the rendering logic in `chatTerminalToolConfirmationSubPart.ts` to present this additional context to the user, allowing for more informed decisions before approving potentially risky operations.

This commit performs a significant **refactoring** of the **Terminal Chat Agent Tools**' **sandbox network configuration**, flattening the previously nested `chat.tools.terminal.sandbox.network` object into three distinct settings: `allowedDomains`, `deniedDomains`, and `allowTrustedDomains`. This **maintenance** effort improves configuration clarity and manageability for network access within the sandbox. It includes an **automatic configuration migration** to seamlessly update existing user settings, ensuring backward compatibility while marking the old setting as deprecated. All internal logic, configuration listeners, and tests within the **terminal sandbox service** and related components have been updated to utilize these new, flattened settings.

This commit implements a **bug fix** to enhance the resilience of **Copilot extension's output channel logging** during extension host shutdown. It wraps logging calls, specifically within the `logIt` function in `outputChannelLogTarget.ts`, in a try-catch block. This prevents unhandled exceptions that previously occurred if the debug channel was already closed when logging attempts were made. Consequently, the **Copilot extension** now handles logging gracefully throughout its lifecycle, improving overall stability and resolving issue #303916.

This commit delivers several **bug fixes** to improve the reliability of **terminal sandboxing** and **command rewriting** for tool terminal commands. It resolves issues where sandboxed commands incorrectly appeared in shell history by reordering the `PreventHistoryRewriter` to run last, and fixes command display problems by ensuring `SandboxedCommandLinePresenter` uses the rewritten command's `forDisplay` property. Crucially, the **bash shell integration** script is updated to correctly handle `ignorespace` in `HISTCONTROL`, preventing `__vsc_current_command` from being empty and thus restoring proper exit code detection for tool terminal commands. These **maintenance** changes ensure accurate command history, correct display, and reliable exit code reporting for sandboxed operations, while also upgrading logging for **CI diagnostics**.

This commit introduces a **test fix** within the **`vscode-api-tests`** suite, specifically for the `chat.runInTerminal` functionality. It **modifies an assertion** in `extensions/vscode-api-tests/src/singlefolder-tests/chat.runInTerminal.test.ts` to **accept "Command produced no output"** when shell integration is disabled. This change **prevents false negatives** and ensures the reliability of API tests by accurately validating terminal command execution behavior under various configurations.

This commit **fixes a display bug** in the **ChatTerminalThinkingCollapsibleWrapper** within the **Chat** workbench contribution, where **terminal command labels** were incorrectly escaped, leading to visible backslashes or broken markdown rendering. The rendering logic has been **refactored** to use direct DOM manipulation with `textContent` and `<code>` elements, ensuring that all command text, including those with special characters like backticks, is displayed literally and safely. This **bug fix** significantly improves the visual fidelity and robustness of how terminal commands are presented in the chat interface. Furthermore, **visual regression tests** have been added for the terminal collapsible wrapper to prevent similar display issues in the future.

This commit delivers a **bug fix** to the **terminal command confirmation** system, specifically addressing an issue within the `chatAgentTools` contribution. Previously, the `CommandLineSandboxAnalyzer` would incorrectly force auto-approval and skip confirmation prompts even when sandbox command rewriting was disabled via `_enableCommandLineSandboxRewriting = false`. The **fix** now ensures that the `CommandLineSandboxAnalyzer` is only active when `_enableCommandLineSandboxRewriting` is enabled, aligning its behavior with the `CommandLineSandboxRewriter`. This prevents unintended command execution by correctly presenting confirmation prompts when sandbox rewriting is not in effect, improving the reliability of the **terminal command execution flow**.

This extensive commit introduces significant **bug fixes and improvements** to the **terminal output capture and processing** within the **chat agent tools** and **sandbox environments**. It refines the `stripCommandEchoAndPrompt` logic to accurately remove command echoes and prompts, handling various shell integration scenarios, wrapped lines, and platform-specific prompt formats, preventing legitimate output from being stripped. Additionally, it addresses **premature idle detection** in execution strategies, enhances **sandbox failure detection** with heuristic analysis and platform-aware error handling, and ensures correct `execPath` resolution for **remote sandbox environments**. The changes also force `/bin/bash` for the Copilot terminal profile to guarantee **shell integration** functionality and update CI to install necessary sandbox dependencies. This work collectively improves the reliability and cross-platform compatibility of terminal interactions, especially for automated tools consuming terminal output.

This commit **fixes a timing issue** within the **`vscode-api-tests` extension** by modifying the `chat.runInTerminal.test.ts` file. The test now asynchronously polls for the `run_in_terminal` tool to be registered, addressing a race condition where the tool might not be immediately available. This **test fix** improves the **reliability of the test suite** for the `run_in_terminal` feature, preventing intermittent failures caused by asynchronous tool registration. The change ensures the integration tests for this functionality are stable and accurate.

This commit introduces a **fix** to the **terminal chat agent tools**' prompt detection logic, specifically within `strategyHelpers.ts`. It now distinguishes between complete and fragment prompt patterns, preventing the incorrect stripping of command output that might coincidentally end with prompt-like characters. This enhancement ensures that only actual wrapped prompts are removed, improving the accuracy of terminal output parsing. New adversarial tests in `strategyHelpers.test.ts` have been added to verify this correct behavior, addressing previous issues where legitimate output was falsely identified as part of a wrapped prompt.

This commit introduces a **bug fix** and **maintenance** update to the **terminal chat agent tools**, specifically enhancing the robustness of terminal output parsing. It **extends the recognition of bracketed terminal prompts** to correctly identify formats without an `@` symbol, such as `[W007DV9PF9-1:~/path]`, which is common in CI environments like macOS. Additionally, the commit **caps the trailing prompt stripping** logic within `_stripCommandEchoAndPromptOnce` to a maximum of two non-empty lines, preventing the accidental removal of legitimate output. This ensures more accurate processing of terminal command echoes and prompts, with new unit tests added to validate the improved prompt pattern recognition.

This commit **fixes a bug** preventing **shell integration** from working correctly within **Copilot terminal profiles** when `/bin/sh` was the default shell, leading to a loss of exit code detection. It introduces a change in `src/vs/workbench/contrib/terminalContrib/chatAgentTools/browser/tools/runInTerminalTool.ts` to **force `/bin/bash`** over `/bin/sh` for the default terminal profile, mirroring existing platform-specific overrides. This ensures that **Copilot's terminal agent tools** can properly inject shell integration, improving the reliability and functionality of features like exit code detection.

Install bubblewrap and socat in Linux CI pipelines

This commit implements several **maintenance**, **performance**, and **robustness improvements** within the **Terminal Chat Agent Tools** based on PR feedback. It introduces a **performance optimization** in `strategyHelpers.ts` by replacing an inefficient O(n^2) string concatenation with `array.join` in `stripNewLinesAndBuildMapping`, and enhances **security** by stripping sensitive data from debug logs in `stripCommandEchoAndPrompt` functions. A critical **bug fix** is applied in `noneExecuteStrategy.ts`, adding a **5-second timeout** to the cursor movement wait to prevent indefinite hangs during command execution. This work also clarifies descriptions for the `shellIntegrationTimeout` setting, ensuring better understanding of its behavior.

This commit **fixes a unit test** within the **Terminal Chat Agent Tools** contribution by adjusting a mock value. Specifically, it corrects the `cursorY` mock value in the `noneExecuteStrategy.test.ts` file. This is a **maintenance fix** that ensures the reliability and accuracy of the unit tests for the `NoneExecuteStrategy`, preventing potential false negatives due to incorrect test setup. The change improves the robustness of the **Terminal Chat Agent's** testing infrastructure.

This commit delivers a crucial **bug fix** for **terminal output capture** within the **chat agent tools**, addressing problems with premature idle detection and the accurate handling of partial command echoes. It prevents marker jumping by ensuring the `setupRecreatingStartMarker` is properly disposed before sending commands and delays idle detection in `noneExecuteStrategy` until the cursor moves past the command's start line. Additionally, the `findCommandEcho` utility now supports **suffix matching** to correctly identify and strip partial command echoes from wrapped `getOutput()` results, improving reliability for long commands with shell integration. These enhancements significantly improve the accuracy and robustness of **terminal command execution and output parsing**, ensuring a more consistent experience for features relying on precise output capture. Integration tests were also updated to validate these fixes across different platforms and scenarios.

This commit **fixes a critical issue** in **remote environments** where the **sandbox service** incorrectly resolved the executable path (`execPath`), leading to failures outside of production builds. To rectify this, the `IRemoteAgentEnvironment` interface and its associated **remote agent environment services** are **enhanced** to expose and transmit the server's actual `process.execPath` to the client. The `terminalSandboxService` is then updated to leverage this new `remoteEnv.execPath` instead of its previous hardcoded logic. This **bug fix** ensures **correct sandbox executable path resolution** for features like chat agent tools, improving reliability across all remote setups.

This commit **fixes** a **test** within the **VS Code API test suite** by ensuring proper setup for **chat history** verification. Specifically, it modifies `extensions/vscode-api-tests/src/singlefolder-tests/chat.test.ts` to explicitly open a new chat session before testing participant and slash command history. This **test improvement** prevents potential flakiness or failures by guaranteeing the correct initial state, thereby enhancing the **reliability** of the **chat feature's API tests**.