feat(core): refactor shell execution to use node-pty (#6491)

Co-authored-by: Jacob Richman <jacob314@gmail.com>

packages/core/src/tools/shell.test.ts

@@ -56,6 +56,7 @@ describe('ShellTool', () => {
       getSummarizeToolOutputConfig: vi.fn().mockReturnValue(undefined),
       getWorkspaceContext: () => createMockWorkspaceContext('.'),
       getGeminiClient: vi.fn(),
+      getShouldUseNodePtyShell: vi.fn().mockReturnValue(false),
     } as unknown as Config;
 
     shellTool = new ShellTool(mockConfig);
@@ -123,13 +124,12 @@ describe('ShellTool', () => {
       const fullResult: ShellExecutionResult = {
         rawOutput: Buffer.from(result.output || ''),
         output: 'Success',
-        stdout: 'Success',
-        stderr: '',
         exitCode: 0,
         signal: null,
         error: null,
         aborted: false,
         pid: 12345,
+        executionMethod: 'child_process',
         ...result,
       };
       resolveExecutionPromise(fullResult);
@@ -152,6 +152,9 @@ describe('ShellTool', () => {
         expect.any(String),
         expect.any(Function),
         mockAbortSignal,
+        false,
+        undefined,
+        undefined,
       );
       expect(result.llmContent).toContain('Background PIDs: 54322');
       expect(vi.mocked(fs.unlinkSync)).toHaveBeenCalledWith(tmpFile);
@@ -164,13 +167,12 @@ describe('ShellTool', () => {
       resolveShellExecution({
         rawOutput: Buffer.from(''),
         output: '',
-        stdout: '',
-        stderr: '',
         exitCode: 0,
         signal: null,
         error: null,
         aborted: false,
         pid: 12345,
+        executionMethod: 'child_process',
       });
       await promise;
       expect(mockShellExecutionService).toHaveBeenCalledWith(
@@ -178,6 +180,9 @@ describe('ShellTool', () => {
         expect.any(String),
         expect.any(Function),
         mockAbortSignal,
+        false,
+        undefined,
+        undefined,
       );
     });
 
@@ -189,16 +194,14 @@ describe('ShellTool', () => {
         error,
         exitCode: 1,
         output: 'err',
-        stderr: 'err',
         rawOutput: Buffer.from('err'),
-        stdout: '',
         signal: null,
         aborted: false,
         pid: 12345,
+        executionMethod: 'child_process',
       });
 
       const result = await promise;
-      // The final llmContent should contain the user's command, not the wrapper
       expect(result.llmContent).toContain('Error: wrapped command failed');
       expect(result.llmContent).not.toContain('pgrep');
     });
@@ -231,13 +234,12 @@ describe('ShellTool', () => {
       resolveExecutionPromise({
         output: 'long output',
         rawOutput: Buffer.from('long output'),
-        stdout: 'long output',
-        stderr: '',
         exitCode: 0,
         signal: null,
         error: null,
         aborted: false,
         pid: 12345,
+        executionMethod: 'child_process',
       });
 
       const result = await promise;
@@ -283,7 +285,6 @@ describe('ShellTool', () => {
         // First chunk, should be throttled.
         mockShellOutputCallback({
           type: 'data',
-          stream: 'stdout',
           chunk: 'hello ',
         });
         expect(updateOutputMock).not.toHaveBeenCalled();
@@ -294,24 +295,22 @@ describe('ShellTool', () => {
         // Send a second chunk. THIS event triggers the update with the CUMULATIVE content.
         mockShellOutputCallback({
           type: 'data',
-          stream: 'stderr',
-          chunk: 'world',
+          chunk: 'hello world',
         });
 
         // It should have been called once now with the combined output.
         expect(updateOutputMock).toHaveBeenCalledOnce();
-        expect(updateOutputMock).toHaveBeenCalledWith('hello \nworld');
+        expect(updateOutputMock).toHaveBeenCalledWith('hello world');
 
         resolveExecutionPromise({
           rawOutput: Buffer.from(''),
           output: '',
-          stdout: '',
-          stderr: '',
           exitCode: 0,
           signal: null,
           error: null,
           aborted: false,
           pid: 12345,
+          executionMethod: 'child_process',
         });
         await promise;
       });
@@ -350,13 +349,12 @@ describe('ShellTool', () => {
         resolveExecutionPromise({
           rawOutput: Buffer.from(''),
           output: '',
-          stdout: '',
-          stderr: '',
           exitCode: 0,
           signal: null,
           error: null,
           aborted: false,
           pid: 12345,
+          executionMethod: 'child_process',
         });
         await promise;
       });

packages/core/src/tools/shell.ts

@@ -96,6 +96,8 @@ class ShellToolInvocation extends BaseToolInvocation<
   async execute(
     signal: AbortSignal,
     updateOutput?: (output: string) => void,
+    terminalColumns?: number,
+    terminalRows?: number,
   ): Promise<ToolResult> {
     const strippedCommand = stripShellWrapper(this.params.command);
 
@@ -128,13 +130,11 @@ class ShellToolInvocation extends BaseToolInvocation<
         this.params.directory || '',
       );
 
-      let cumulativeStdout = '';
-      let cumulativeStderr = '';
-
+      let cumulativeOutput = '';
       let lastUpdateTime = Date.now();
       let isBinaryStream = false;
 
-      const { result: resultPromise } = ShellExecutionService.execute(
+      const { result: resultPromise } = await ShellExecutionService.execute(
         commandToExecute,
         cwd,
         (event: ShellOutputEvent) => {
@@ -147,15 +147,9 @@ class ShellToolInvocation extends BaseToolInvocation<
 
           switch (event.type) {
             case 'data':
-              if (isBinaryStream) break; // Don't process text if we are in binary mode
-              if (event.stream === 'stdout') {
-                cumulativeStdout += event.chunk;
-              } else {
-                cumulativeStderr += event.chunk;
-              }
-              currentDisplayOutput =
-                cumulativeStdout +
-                (cumulativeStderr ? `\n${cumulativeStderr}` : '');
+              if (isBinaryStream) break;
+              cumulativeOutput = event.chunk;
+              currentDisplayOutput = cumulativeOutput;
               if (Date.now() - lastUpdateTime > OUTPUT_UPDATE_INTERVAL_MS) {
                 shouldUpdate = true;
               }
@@ -186,6 +180,9 @@ class ShellToolInvocation extends BaseToolInvocation<
           }
         },
         signal,
+        this.config.getShouldUseNodePtyShell(),
+        terminalColumns,
+        terminalRows,
       );
 
       const result = await resultPromise;
@@ -217,7 +214,7 @@ class ShellToolInvocation extends BaseToolInvocation<
       if (result.aborted) {
         llmContent = 'Command was cancelled by user before it could complete.';
         if (result.output.trim()) {
-          llmContent += ` Below is the output (on stdout and stderr) before it was cancelled:\n${result.output}`;
+          llmContent += ` Below is the output before it was cancelled:\n${result.output}`;
         } else {
           llmContent += ' There was no output before it was cancelled.';
         }
@@ -231,8 +228,7 @@ class ShellToolInvocation extends BaseToolInvocation<
         llmContent = [
           `Command: ${this.params.command}`,
           `Directory: ${this.params.directory || '(root)'}`,
-          `Stdout: ${result.stdout || '(empty)'}`,
-          `Stderr: ${result.stderr || '(empty)'}`,
+          `Output: ${result.output || '(empty)'}`,
           `Error: ${finalError}`, // Use the cleaned error string.
           `Exit Code: ${result.exitCode ?? '(none)'}`,
           `Signal: ${result.signal ?? '(none)'}`,