summaryrefslogtreecommitdiff
path: root/packages/server/src
diff options
context:
space:
mode:
Diffstat (limited to 'packages/server/src')
-rw-r--r--packages/server/src/core/gemini-client.ts18
-rw-r--r--packages/server/src/core/turn.ts134
-rw-r--r--packages/server/src/tools/edit.ts95
-rw-r--r--packages/server/src/tools/glob.ts8
-rw-r--r--packages/server/src/tools/grep.ts8
-rw-r--r--packages/server/src/tools/ls.ts8
-rw-r--r--packages/server/src/tools/read-file.ts15
-rw-r--r--packages/server/src/tools/terminal.ts959
-rw-r--r--packages/server/src/tools/tools.ts45
-rw-r--r--packages/server/src/tools/web-fetch.ts8
-rw-r--r--packages/server/src/tools/write-file.ts68
-rw-r--r--packages/server/src/utils/BackgroundTerminalAnalyzer.ts473
12 files changed, 1653 insertions, 186 deletions
diff --git a/packages/server/src/core/gemini-client.ts b/packages/server/src/core/gemini-client.ts
index d78a0559..b9b44534 100644
--- a/packages/server/src/core/gemini-client.ts
+++ b/packages/server/src/core/gemini-client.ts
@@ -18,15 +18,7 @@ import {
import { CoreSystemPrompt } from './prompts.js';
import process from 'node:process';
import { getFolderStructure } from '../utils/getFolderStructure.js';
-import { Turn, ServerTool, GeminiEventType } from './turn.js';
-
-// Import the ServerGeminiStreamEvent type
-type ServerGeminiStreamEvent =
- | { type: GeminiEventType.Content; value: string }
- | {
- type: GeminiEventType.ToolCallRequest;
- value: { callId: string; name: string; args: Record<string, unknown> };
- };
+import { Turn, ServerTool, ServerGeminiStreamEvent } from './turn.js';
export class GeminiClient {
private ai: GoogleGenAI;
@@ -112,6 +104,14 @@ export class GeminiClient {
for await (const event of resultStream) {
yield event;
}
+
+ const confirmations = turn.getConfirmationDetails();
+ if (confirmations.length > 0) {
+ break;
+ }
+
+ // What do we do when we have both function responses and confirmations?
+
const fnResponses = turn.getFunctionResponses();
if (fnResponses.length > 0) {
request = fnResponses;
diff --git a/packages/server/src/core/turn.ts b/packages/server/src/core/turn.ts
index 3d8c8c76..0a1c594c 100644
--- a/packages/server/src/core/turn.ts
+++ b/packages/server/src/core/turn.ts
@@ -13,7 +13,11 @@ import {
FunctionDeclaration,
} from '@google/genai';
// Removed UI type imports
-import { ToolResult } from '../tools/tools.js'; // Keep ToolResult for now
+import {
+ ToolCallConfirmationDetails,
+ ToolResult,
+ ToolResultDisplay,
+} from '../tools/tools.js'; // Keep ToolResult for now
// Removed gemini-stream import (types defined locally)
// --- Types for Server Logic ---
@@ -25,7 +29,7 @@ interface ServerToolExecutionOutcome {
args: Record<string, unknown>; // Use unknown for broader compatibility
result?: ToolResult;
error?: Error;
- // Confirmation details are handled by CLI, not server logic
+ confirmationDetails: ToolCallConfirmationDetails | undefined;
}
// Define a structure for tools passed to the server
@@ -34,6 +38,9 @@ export interface ServerTool {
schema: FunctionDeclaration; // Schema is needed
// The execute method signature might differ slightly or be wrapped
execute(params: Record<string, unknown>): Promise<ToolResult>;
+ shouldConfirmExecute(
+ params: Record<string, unknown>,
+ ): Promise<ToolCallConfirmationDetails | false>;
// validation and description might be handled differently or passed
}
@@ -41,17 +48,36 @@ export interface ServerTool {
export enum GeminiEventType {
Content = 'content',
ToolCallRequest = 'tool_call_request',
+ ToolCallResponse = 'tool_call_response',
+ ToolCallConfirmation = 'tool_call_confirmation',
}
-interface ToolCallRequestInfo {
+export interface ToolCallRequestInfo {
callId: string;
name: string;
args: Record<string, unknown>;
}
-type ServerGeminiStreamEvent =
+export interface ToolCallResponseInfo {
+ callId: string;
+ responsePart: Part;
+ resultDisplay: ToolResultDisplay | undefined;
+ error: Error | undefined;
+}
+
+export interface ServerToolCallConfirmationDetails {
+ request: ToolCallRequestInfo;
+ details: ToolCallConfirmationDetails;
+}
+
+export type ServerGeminiStreamEvent =
| { type: GeminiEventType.Content; value: string }
- | { type: GeminiEventType.ToolCallRequest; value: ToolCallRequestInfo };
+ | { type: GeminiEventType.ToolCallRequest; value: ToolCallRequestInfo }
+ | { type: GeminiEventType.ToolCallResponse; value: ToolCallResponseInfo }
+ | {
+ type: GeminiEventType.ToolCallConfirmation;
+ value: ServerToolCallConfirmationDetails;
+ };
// --- Turn Class (Refactored for Server) ---
@@ -65,6 +91,7 @@ export class Turn {
args: Record<string, unknown>; // Use unknown
}>;
private fnResponses: Part[];
+ private confirmationDetails: ToolCallConfirmationDetails[];
private debugResponses: GenerateContentResponse[];
constructor(chat: Chat, availableTools: ServerTool[]) {
@@ -72,6 +99,7 @@ export class Turn {
this.availableTools = new Map(availableTools.map((t) => [t.name, t]));
this.pendingToolCalls = [];
this.fnResponses = [];
+ this.confirmationDetails = [];
this.debugResponses = [];
}
@@ -113,19 +141,31 @@ export class Turn {
error: new Error(
`Tool "${pendingToolCall.name}" not found or not provided to Turn.`,
),
+ confirmationDetails: undefined,
};
}
- // No confirmation logic in the server Turn
+
try {
- // TODO: Add validation step if needed (tool.validateParams?)
- const result = await tool.execute(pendingToolCall.args);
- return { ...pendingToolCall, result };
+ const confirmationDetails = await tool.shouldConfirmExecute(
+ pendingToolCall.args,
+ );
+ if (confirmationDetails) {
+ return { ...pendingToolCall, confirmationDetails };
+ } else {
+ const result = await tool.execute(pendingToolCall.args);
+ return {
+ ...pendingToolCall,
+ result,
+ confirmationDetails: undefined,
+ };
+ }
} catch (execError: unknown) {
return {
...pendingToolCall,
error: new Error(
`Tool execution failed: ${execError instanceof Error ? execError.message : String(execError)}`,
),
+ confirmationDetails: undefined,
};
}
},
@@ -133,9 +173,37 @@ export class Turn {
const outcomes = await Promise.all(toolPromises);
// Process outcomes and prepare function responses
- this.fnResponses = this.buildFunctionResponses(outcomes);
this.pendingToolCalls = []; // Clear pending calls for this turn
+ for (let i = 0; i < outcomes.length; i++) {
+ const outcome = outcomes[i];
+ if (outcome.confirmationDetails) {
+ this.confirmationDetails.push(outcome.confirmationDetails);
+ const serverConfirmationetails: ServerToolCallConfirmationDetails = {
+ request: {
+ callId: outcome.callId,
+ name: outcome.name,
+ args: outcome.args,
+ },
+ details: outcome.confirmationDetails,
+ };
+ yield {
+ type: GeminiEventType.ToolCallConfirmation,
+ value: serverConfirmationetails,
+ };
+ } else {
+ const responsePart = this.buildFunctionResponse(outcome);
+ this.fnResponses.push(responsePart);
+ const responseInfo: ToolCallResponseInfo = {
+ callId: outcome.callId,
+ responsePart,
+ resultDisplay: outcome.result?.returnDisplay,
+ error: outcome.error,
+ };
+ yield { type: GeminiEventType.ToolCallResponse, value: responseInfo };
+ }
+ }
+
// If there were function responses, the caller (GeminiService) will loop
// and call run() again with these responses.
// If no function responses, the turn ends here.
@@ -160,31 +228,27 @@ export class Turn {
}
// Builds the Part array expected by the Google GenAI API
- private buildFunctionResponses(
- outcomes: ServerToolExecutionOutcome[],
- ): Part[] {
- return outcomes.map((outcome): Part => {
- const { name, result, error } = outcome;
- let fnResponsePayload: Record<string, unknown>;
+ private buildFunctionResponse(outcome: ServerToolExecutionOutcome): Part {
+ const { name, result, error } = outcome;
+ let fnResponsePayload: Record<string, unknown>;
- if (error) {
- // Format error for the LLM
- const errorMessage = error?.message || String(error);
- fnResponsePayload = { error: `Tool execution failed: ${errorMessage}` };
- console.error(`[Server Turn] Error executing tool ${name}:`, error);
- } else {
- // Pass successful tool result (content meant for LLM)
- fnResponsePayload = { output: result?.llmContent ?? '' }; // Default to empty string if no content
- }
+ if (error) {
+ // Format error for the LLM
+ const errorMessage = error?.message || String(error);
+ fnResponsePayload = { error: `Tool execution failed: ${errorMessage}` };
+ console.error(`[Server Turn] Error executing tool ${name}:`, error);
+ } else {
+ // Pass successful tool result (content meant for LLM)
+ fnResponsePayload = { output: result?.llmContent ?? '' }; // Default to empty string if no content
+ }
- return {
- functionResponse: {
- name,
- id: outcome.callId,
- response: fnResponsePayload,
- },
- };
- });
+ return {
+ functionResponse: {
+ name,
+ id: outcome.callId,
+ response: fnResponsePayload,
+ },
+ };
}
private abortError(): Error {
@@ -193,6 +257,10 @@ export class Turn {
return error; // Return instead of throw, let caller handle
}
+ getConfirmationDetails(): ToolCallConfirmationDetails[] {
+ return this.confirmationDetails;
+ }
+
// Allows the service layer to get the responses needed for the next API call
getFunctionResponses(): Part[] {
return this.fnResponses;
diff --git a/packages/server/src/tools/edit.ts b/packages/server/src/tools/edit.ts
index 67c5a37b..5dbeaf41 100644
--- a/packages/server/src/tools/edit.ts
+++ b/packages/server/src/tools/edit.ts
@@ -7,7 +7,14 @@
import fs from 'fs';
import path from 'path';
import * as Diff from 'diff';
-import { BaseTool, ToolResult, ToolResultDisplay } from './tools.js';
+import {
+ BaseTool,
+ ToolCallConfirmationDetails,
+ ToolConfirmationOutcome,
+ ToolEditConfirmationDetails,
+ ToolResult,
+ ToolResultDisplay,
+} from './tools.js';
import { SchemaValidator } from '../utils/schemaValidator.js';
import { makeRelative, shortenPath } from '../utils/paths.js';
import { isNodeError } from '../utils/errors.js';
@@ -48,8 +55,9 @@ interface CalculatedEdit {
/**
* Implementation of the Edit tool logic (moved from CLI)
*/
-export class EditLogic extends BaseTool<EditToolParams, ToolResult> {
+export class EditTool extends BaseTool<EditToolParams, ToolResult> {
static readonly Name = 'replace'; // Keep static name
+ private shouldAlwaysEdit = false;
private readonly rootDirectory: string;
@@ -61,9 +69,9 @@ export class EditLogic extends BaseTool<EditToolParams, ToolResult> {
// Note: The description here mentions other tools like ReadFileTool/WriteFileTool
// by name. This might need updating if those tool names change.
super(
- EditLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ EditTool.Name,
+ 'Edit',
+ 'Replaces a SINGLE, UNIQUE occurrence of text within a file. Requires providing significant context around the change to ensure uniqueness. For moving/renaming files, use the Bash tool with `mv`. For replacing entire file contents or creating new files use the WriteFile tool. Always use the ReadFile tool to examine the file before using this tool.',
{
properties: {
file_path: {
@@ -225,7 +233,82 @@ export class EditLogic extends BaseTool<EditToolParams, ToolResult> {
};
}
- // Removed shouldConfirmExecute - Confirmation is handled by the CLI wrapper
+ /**
+ * Handles the confirmation prompt for the Edit tool in the CLI.
+ * It needs to calculate the diff to show the user.
+ */
+ async shouldConfirmExecute(
+ params: EditToolParams,
+ ): Promise<ToolCallConfirmationDetails | false> {
+ if (this.shouldAlwaysEdit) {
+ return false;
+ }
+ const validationError = this.validateToolParams(params);
+ if (validationError) {
+ console.error(
+ `[EditTool Wrapper] Attempted confirmation with invalid parameters: ${validationError}`,
+ );
+ return false;
+ }
+ let currentContent: string | null = null;
+ let fileExists = false;
+ let newContent = '';
+ try {
+ currentContent = fs.readFileSync(params.file_path, 'utf8');
+ fileExists = true;
+ } catch (err: unknown) {
+ if (isNodeError(err) && err.code === 'ENOENT') {
+ fileExists = false;
+ } else {
+ console.error(`Error reading file for confirmation diff: ${err}`);
+ return false;
+ }
+ }
+ if (params.old_string === '' && !fileExists) {
+ newContent = params.new_string;
+ } else if (!fileExists) {
+ return false;
+ } else if (currentContent !== null) {
+ const occurrences = this.countOccurrences(
+ currentContent,
+ params.old_string,
+ );
+ const expectedReplacements =
+ params.expected_replacements === undefined
+ ? 1
+ : params.expected_replacements;
+ if (occurrences === 0 || occurrences !== expectedReplacements) {
+ return false;
+ }
+ newContent = this.replaceAll(
+ currentContent,
+ params.old_string,
+ params.new_string,
+ );
+ } else {
+ return false;
+ }
+ const fileName = path.basename(params.file_path);
+ const fileDiff = Diff.createPatch(
+ fileName,
+ currentContent ?? '',
+ newContent,
+ 'Current',
+ 'Proposed',
+ { context: 3 },
+ );
+ const confirmationDetails: ToolEditConfirmationDetails = {
+ title: `Confirm Edit: ${shortenPath(makeRelative(params.file_path, this.rootDirectory))}`,
+ fileName,
+ fileDiff,
+ onConfirm: async (outcome: ToolConfirmationOutcome) => {
+ if (outcome === ToolConfirmationOutcome.ProceedAlways) {
+ this.shouldAlwaysEdit = true;
+ }
+ },
+ };
+ return confirmationDetails;
+ }
getDescription(params: EditToolParams): string {
const relativePath = makeRelative(params.file_path, this.rootDirectory);
diff --git a/packages/server/src/tools/glob.ts b/packages/server/src/tools/glob.ts
index e81858c8..f51456c3 100644
--- a/packages/server/src/tools/glob.ts
+++ b/packages/server/src/tools/glob.ts
@@ -29,7 +29,7 @@ export interface GlobToolParams {
/**
* Implementation of the Glob tool logic (moved from CLI)
*/
-export class GlobLogic extends BaseTool<GlobToolParams, ToolResult> {
+export class GlobTool extends BaseTool<GlobToolParams, ToolResult> {
static readonly Name = 'glob'; // Keep static name
/**
@@ -43,9 +43,9 @@ export class GlobLogic extends BaseTool<GlobToolParams, ToolResult> {
*/
constructor(rootDirectory: string) {
super(
- GlobLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ GlobTool.Name,
+ 'FindFiles', // Display name handled by CLI wrapper
+ 'Efficiently finds files matching specific glob patterns (e.g., `src/**/*.ts`, `**/*.md`), returning absolute paths sorted by modification time (newest first). Ideal for quickly locating files based on their name or path structure, especially in large codebases.', // Description handled by CLI wrapper
{
properties: {
pattern: {
diff --git a/packages/server/src/tools/grep.ts b/packages/server/src/tools/grep.ts
index b0d4637c..1873a794 100644
--- a/packages/server/src/tools/grep.ts
+++ b/packages/server/src/tools/grep.ts
@@ -51,7 +51,7 @@ interface GrepMatch {
/**
* Implementation of the Grep tool logic (moved from CLI)
*/
-export class GrepLogic extends BaseTool<GrepToolParams, ToolResult> {
+export class GrepTool extends BaseTool<GrepToolParams, ToolResult> {
static readonly Name = 'search_file_content'; // Keep static name
private rootDirectory: string;
@@ -62,9 +62,9 @@ export class GrepLogic extends BaseTool<GrepToolParams, ToolResult> {
*/
constructor(rootDirectory: string) {
super(
- GrepLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ GrepTool.Name,
+ 'SearchText',
+ 'Searches for a regular expression pattern within the content of files in a specified directory (or current working directory). Can filter files by a glob pattern. Returns the lines containing matches, along with their file paths and line numbers.',
{
properties: {
pattern: {
diff --git a/packages/server/src/tools/ls.ts b/packages/server/src/tools/ls.ts
index 0e856e80..a646dd22 100644
--- a/packages/server/src/tools/ls.ts
+++ b/packages/server/src/tools/ls.ts
@@ -58,7 +58,7 @@ export interface FileEntry {
/**
* Implementation of the LS tool logic
*/
-export class LSLogic extends BaseTool<LSToolParams, ToolResult> {
+export class LSTool extends BaseTool<LSToolParams, ToolResult> {
static readonly Name = 'list_directory';
/**
@@ -73,9 +73,9 @@ export class LSLogic extends BaseTool<LSToolParams, ToolResult> {
*/
constructor(rootDirectory: string) {
super(
- LSLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ LSTool.Name,
+ 'ReadFolder',
+ 'Lists the names of files and subdirectories directly within a specified directory path. Can optionally ignore entries matching provided glob patterns.',
{
properties: {
path: {
diff --git a/packages/server/src/tools/read-file.ts b/packages/server/src/tools/read-file.ts
index 52856e42..6cd70302 100644
--- a/packages/server/src/tools/read-file.ts
+++ b/packages/server/src/tools/read-file.ts
@@ -33,7 +33,7 @@ export interface ReadFileToolParams {
/**
* Implementation of the ReadFile tool logic
*/
-export class ReadFileLogic extends BaseTool<ReadFileToolParams, ToolResult> {
+export class ReadFileTool extends BaseTool<ReadFileToolParams, ToolResult> {
static readonly Name: string = 'read_file';
private static readonly DEFAULT_MAX_LINES = 2000;
private static readonly MAX_LINE_LENGTH = 2000;
@@ -41,9 +41,9 @@ export class ReadFileLogic extends BaseTool<ReadFileToolParams, ToolResult> {
constructor(rootDirectory: string) {
super(
- ReadFileLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ ReadFileTool.Name,
+ 'ReadFile',
+ 'Reads and returns the content of a specified file from the local filesystem. Handles large files by allowing reading specific line ranges.',
{
properties: {
path: {
@@ -236,16 +236,15 @@ export class ReadFileLogic extends BaseTool<ReadFileToolParams, ToolResult> {
const startLine = params.offset || 0;
const endLine = params.limit
? startLine + params.limit
- : Math.min(startLine + ReadFileLogic.DEFAULT_MAX_LINES, lines.length);
+ : Math.min(startLine + ReadFileTool.DEFAULT_MAX_LINES, lines.length);
const selectedLines = lines.slice(startLine, endLine);
let truncated = false;
const formattedLines = selectedLines.map((line) => {
let processedLine = line;
- if (line.length > ReadFileLogic.MAX_LINE_LENGTH) {
+ if (line.length > ReadFileTool.MAX_LINE_LENGTH) {
processedLine =
- line.substring(0, ReadFileLogic.MAX_LINE_LENGTH) +
- '... [truncated]';
+ line.substring(0, ReadFileTool.MAX_LINE_LENGTH) + '... [truncated]';
truncated = true;
}
diff --git a/packages/server/src/tools/terminal.ts b/packages/server/src/tools/terminal.ts
index 6366106c..eab170ab 100644
--- a/packages/server/src/tools/terminal.ts
+++ b/packages/server/src/tools/terminal.ts
@@ -4,18 +4,42 @@
* SPDX-License-Identifier: Apache-2.0
*/
-import { spawn, SpawnOptions } from 'child_process';
+import {
+ spawn,
+ SpawnOptions,
+ ChildProcessWithoutNullStreams,
+} from 'child_process';
import path from 'path';
-import { BaseTool, ToolResult } from './tools.js';
-import { SchemaValidator } from '../utils/schemaValidator.js';
-import { getErrorMessage } from '../utils/errors.js';
+import os from 'os';
+import crypto from 'crypto';
+import { promises as fs } from 'fs';
+import {
+ SchemaValidator,
+ getErrorMessage,
+ isNodeError,
+ Config,
+} from '@gemini-code/server';
+import {
+ BaseTool,
+ ToolCallConfirmationDetails,
+ ToolConfirmationOutcome,
+ ToolExecuteConfirmationDetails,
+ ToolResult,
+} from './tools.js';
+import { BackgroundTerminalAnalyzer } from '../utils/BackgroundTerminalAnalyzer.js';
export interface TerminalToolParams {
command: string;
+ description?: string;
+ timeout?: number;
+ runInBackground?: boolean;
}
const MAX_OUTPUT_LENGTH = 10000;
-const DEFAULT_EXEC_TIMEOUT_MS = 5 * 60 * 1000;
+const DEFAULT_TIMEOUT_MS = 30 * 60 * 1000;
+const MAX_TIMEOUT_OVERRIDE_MS = 10 * 60 * 1000;
+const BACKGROUND_LAUNCH_TIMEOUT_MS = 15 * 1000;
+const BACKGROUND_POLL_TIMEOUT_MS = 30000;
const BANNED_COMMAND_ROOTS = [
'alias',
@@ -85,41 +109,197 @@ const BANNED_COMMAND_ROOTS = [
'open',
];
-/**
- * Simplified implementation of the Terminal tool logic for single command execution.
- */
-export class TerminalLogic extends BaseTool<TerminalToolParams, ToolResult> {
- static readonly Name = 'execute_bash_command';
+interface QueuedCommand {
+ params: TerminalToolParams;
+ resolve: (result: ToolResult) => void;
+ reject: (error: Error) => void;
+ confirmationDetails: ToolExecuteConfirmationDetails | false;
+}
+
+export class TerminalTool extends BaseTool<TerminalToolParams, ToolResult> {
+ static Name: string = 'execute_bash_command';
private readonly rootDirectory: string;
+ private readonly outputLimit: number;
+ private bashProcess: ChildProcessWithoutNullStreams | null = null;
+ private currentCwd: string;
+ private isExecuting: boolean = false;
+ private commandQueue: QueuedCommand[] = [];
+ private currentCommandCleanup: (() => void) | null = null;
+ private shouldAlwaysExecuteCommands: Map<string, boolean> = new Map();
+ private shellReady: Promise<void>;
+ private resolveShellReady: (() => void) | undefined;
+ private rejectShellReady: ((reason?: unknown) => void) | undefined;
+ private readonly backgroundTerminalAnalyzer: BackgroundTerminalAnalyzer;
+ private readonly config: Config;
- constructor(rootDirectory: string) {
- super(
- TerminalLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
- {
- type: 'object',
- properties: {
- command: {
- description: `The exact bash command or sequence of commands (using ';' or '&&') to execute. Must adhere to usage guidelines. Example: 'npm install && npm run build'`,
- type: 'string',
- },
+ constructor(
+ rootDirectory: string,
+ config: Config,
+ outputLimit: number = MAX_OUTPUT_LENGTH,
+ ) {
+ const toolDisplayName = 'Terminal';
+ const toolDescription = `Executes one or more bash commands sequentially in a secure and persistent interactive shell session. Can run commands in the foreground (waiting for completion) or background (returning after launch, with subsequent status polling).
+
+Core Functionality:
+* Starts in project root: '${path.basename(rootDirectory)}'. Current Directory starts as: ${rootDirectory} (will update based on 'cd' commands).
+* Persistent State: Environment variables and the current working directory (\`pwd\`) persist between calls to this tool.
+* **Execution Modes:**
+ * **Foreground (default):** Waits for the command to complete. Captures stdout, stderr, and exit code. Output is truncated if it exceeds ${outputLimit} characters.
+ * **Background (\`runInBackground: true\`):** Appends \`&\` to the command and redirects its output to temporary files. Returns *after* the command is launched, providing the Process ID (PID) and launch status. Subsequently, the tool **polls** for the background process status for up to ${BACKGROUND_POLL_TIMEOUT_MS / 1000} seconds. Once the process finishes or polling times out, the tool reads the captured stdout/stderr from the temporary files, runs an internal LLM analysis on the output, cleans up the files, and returns the final status, captured output, and analysis.
+* Timeout: Optional timeout per 'execute' call (default: ${DEFAULT_TIMEOUT_MS / 60000} min, max override: ${MAX_TIMEOUT_OVERRIDE_MS / 60000} min for foreground). Background *launch* has a fixed shorter timeout (${BACKGROUND_LAUNCH_TIMEOUT_MS / 1000}s) for the launch attempt itself. Background *polling* has its own timeout (${BACKGROUND_POLL_TIMEOUT_MS / 1000}s). Timeout attempts SIGINT for foreground commands.
+
+Usage Guidance & Restrictions:
+
+1. **Directory/File Verification (IMPORTANT):**
+ * BEFORE executing commands that create files or directories (e.g., \`mkdir foo/bar\`, \`touch new/file.txt\`, \`git clone ...\`), use the dedicated File System tool (e.g., 'list_directory') to verify the target parent directory exists and is the correct location.
+ * Example: Before running \`mkdir foo/bar\`, first use the File System tool to check that \`foo\` exists in the current directory (\`${rootDirectory}\` initially, check current CWD if it changed).
+
+2. **Use Specialized Tools (CRITICAL):**
+ * Do NOT use this tool for filesystem searching (\`find\`, \`grep\`). Use the dedicated Search tool instead.
+ * Do NOT use this tool for reading files (\`cat\`, \`head\`, \`tail\`, \`less\`, \`more\`). Use the dedicated File Reader tool instead.
+ * Do NOT use this tool for listing files (\`ls\`). Use the dedicated File System tool ('list_directory') instead. Relying on this tool's output for directory structure is unreliable due to potential truncation and lack of structured data.
+
+3. **Security & Banned Commands:**
+ * Certain commands are banned for security (e.g., network: ${BANNED_COMMAND_ROOTS.filter((c) => ['curl', 'wget', 'ssh'].includes(c)).join(', ')}; session: ${BANNED_COMMAND_ROOTS.filter((c) => ['exit', 'export', 'kill'].includes(c)).join(', ')}; etc.). The full list is extensive.
+ * If you attempt a banned command, this tool will return an error explaining the restriction. You MUST relay this error clearly to the user.
+
+4. **Command Execution Notes:**
+ * Chain multiple commands using shell operators like ';' or '&&'. Do NOT use newlines within the 'command' parameter string itself (newlines are fine inside quoted arguments).
+ * The shell's current working directory is tracked internally. While \`cd\` is permitted if the user explicitly asks or it's necessary for a workflow, **strongly prefer** using absolute paths or paths relative to the *known* current working directory to avoid errors. Check the '(Executed in: ...)' part of the previous command's output for the CWD.
+ * Good example (if CWD is /workspace/project): \`pytest tests/unit\` or \`ls /workspace/project/data\`
+ * Less preferred: \`cd tests && pytest unit\` (only use if necessary or requested)
+
+5. **Background Tasks (\`runInBackground: true\`):**
+ * Use this for commands that are intended to run continuously (e.g., \`node server.js\`, \`npm start\`).
+ * The tool initially returns success if the process *launches* successfully, along with its PID.
+ * **Polling & Final Result:** The tool then monitors the process. The *final* result (delivered after polling completes or times out) will include:
+ * The final status (completed or timed out).
+ * The complete stdout and stderr captured in temporary files (truncated if necessary).
+ * An LLM-generated analysis/summary of the output.
+ * The initial exit code (usually 0) signifies successful *launching*; the final status indicates completion or timeout after polling.
+
+Use this tool for running build steps (\`npm install\`, \`make\`), linters (\`eslint .\`), test runners (\`pytest\`, \`jest\`), code formatters (\`prettier --write .\`), package managers (\`pip install\`), version control operations (\`git status\`, \`git diff\`), starting background servers/services (\`node server.js --runInBackground true\`), or other safe, standard command-line operations within the project workspace.`;
+ const toolParameterSchema = {
+ type: 'object',
+ properties: {
+ command: {
+ description: `The exact bash command or sequence of commands (using ';' or '&&') to execute. Must adhere to usage guidelines. Example: 'npm install && npm run build'`,
+ type: 'string',
+ },
+ description: {
+ description: `Optional: A brief, user-centric explanation of what the command does and why it's being run. Used for logging and confirmation prompts. Example: 'Install project dependencies'`,
+ type: 'string',
+ },
+ timeout: {
+ description: `Optional execution time limit in milliseconds for FOREGROUND commands. Max ${MAX_TIMEOUT_OVERRIDE_MS}ms (${MAX_TIMEOUT_OVERRIDE_MS / 60000} min). Defaults to ${DEFAULT_TIMEOUT_MS}ms (${DEFAULT_TIMEOUT_MS / 60000} min) if not specified or invalid. Ignored if 'runInBackground' is true.`,
+ type: 'number',
+ },
+ runInBackground: {
+ description: `If true, execute the command in the background using '&'. Defaults to false. Use for servers or long tasks.`,
+ type: 'boolean',
},
- required: ['command'],
},
+ required: ['command'],
+ };
+ super(
+ TerminalTool.Name,
+ toolDisplayName,
+ toolDescription,
+ toolParameterSchema,
);
+ this.config = config;
this.rootDirectory = path.resolve(rootDirectory);
+ this.currentCwd = this.rootDirectory;
+ this.outputLimit = outputLimit;
+ this.shellReady = new Promise((resolve, reject) => {
+ this.resolveShellReady = resolve;
+ this.rejectShellReady = reject;
+ });
+ this.backgroundTerminalAnalyzer = new BackgroundTerminalAnalyzer(config);
+ this.initializeShell();
+ }
+
+ private initializeShell() {
+ if (this.bashProcess) {
+ try {
+ this.bashProcess.kill();
+ } catch {
+ /* Ignore */
+ }
+ }
+ const spawnOptions: SpawnOptions = {
+ cwd: this.rootDirectory,
+ shell: true,
+ env: { ...process.env },
+ stdio: ['pipe', 'pipe', 'pipe'],
+ };
+ try {
+ const bashPath = os.platform() === 'win32' ? 'bash.exe' : 'bash';
+ this.bashProcess = spawn(
+ bashPath,
+ ['-s'],
+ spawnOptions,
+ ) as ChildProcessWithoutNullStreams;
+ this.currentCwd = this.rootDirectory;
+ this.bashProcess.on('error', (err) => {
+ console.error('Persistent Bash Error:', err);
+ this.rejectShellReady?.(err);
+ this.bashProcess = null;
+ this.isExecuting = false;
+ this.clearQueue(
+ new Error(`Persistent bash process failed to start: ${err.message}`),
+ );
+ });
+ this.bashProcess.on('close', (code, signal) => {
+ this.bashProcess = null;
+ this.isExecuting = false;
+ this.rejectShellReady?.(
+ new Error(
+ `Persistent bash process exited (code: ${code}, signal: ${signal})`,
+ ),
+ );
+ this.shellReady = new Promise((resolve, reject) => {
+ this.resolveShellReady = resolve;
+ this.rejectShellReady = reject;
+ });
+ this.clearQueue(
+ new Error(
+ `Persistent bash process exited unexpectedly (code: ${code}, signal: ${signal}). State is lost. Queued commands cancelled.`,
+ ),
+ );
+ if (signal !== 'SIGINT') {
+ setTimeout(() => this.initializeShell(), 1000);
+ }
+ });
+ setTimeout(() => {
+ if (this.bashProcess && !this.bashProcess.killed) {
+ this.resolveShellReady?.();
+ } else if (!this.bashProcess) {
+ // Error likely handled
+ } else {
+ this.rejectShellReady?.(
+ new Error('Shell killed during initialization'),
+ );
+ }
+ }, 1000);
+ } catch (error: unknown) {
+ console.error('Failed to spawn persistent bash:', error);
+ this.rejectShellReady?.(error);
+ this.bashProcess = null;
+ this.clearQueue(
+ new Error(`Failed to spawn persistent bash: ${getErrorMessage(error)}`),
+ );
+ }
}
- validateParams(params: TerminalToolParams): string | null {
+ validateToolParams(params: TerminalToolParams): string | null {
if (
- this.schema.parameters &&
!SchemaValidator.validate(
- this.schema.parameters as Record<string, unknown>,
+ this.parameterSchema as Record<string, unknown>,
params,
)
) {
- return "Parameters failed schema validation (expecting only 'command').";
+ return `Parameters failed schema validation.`;
}
const commandOriginal = params.command.trim();
if (!commandOriginal) {
@@ -137,120 +317,685 @@ export class TerminalLogic extends BaseTool<TerminalToolParams, ToolResult> {
return `Command contains a banned keyword: '${cleanPart}'. Banned list includes network tools, session control, etc.`;
}
}
+ if (
+ params.timeout !== undefined &&
+ (typeof params.timeout !== 'number' || params.timeout <= 0)
+ ) {
+ return 'Timeout must be a positive number of milliseconds.';
+ }
return null;
}
getDescription(params: TerminalToolParams): string {
- return params.command;
+ return params.description || params.command;
}
- async execute(
+ async shouldConfirmExecute(
params: TerminalToolParams,
- executionCwd?: string,
- timeout: number = DEFAULT_EXEC_TIMEOUT_MS,
- ): Promise<ToolResult> {
- const validationError = this.validateParams(params);
+ ): Promise<ToolCallConfirmationDetails | false> {
+ const rootCommand =
+ params.command
+ .trim()
+ .split(/[\s;&&|]+/)[0]
+ ?.split(/[/\\]/)
+ .pop() || 'unknown';
+ if (this.shouldAlwaysExecuteCommands.get(rootCommand)) {
+ return false;
+ }
+ const description = this.getDescription(params);
+ const confirmationDetails: ToolExecuteConfirmationDetails = {
+ title: 'Confirm Shell Command',
+ command: params.command,
+ rootCommand,
+ description: `Execute in '${this.currentCwd}':\n${description}`,
+ onConfirm: async (outcome: ToolConfirmationOutcome) => {
+ if (outcome === ToolConfirmationOutcome.ProceedAlways) {
+ this.shouldAlwaysExecuteCommands.set(rootCommand, true);
+ }
+ },
+ };
+ return confirmationDetails;
+ }
+
+ async execute(params: TerminalToolParams): Promise<ToolResult> {
+ const validationError = this.validateToolParams(params);
if (validationError) {
return {
llmContent: `Command rejected: ${params.command}\nReason: ${validationError}`,
returnDisplay: `Error: ${validationError}`,
};
}
-
- const cwd = executionCwd ? path.resolve(executionCwd) : this.rootDirectory;
- if (!cwd.startsWith(this.rootDirectory) && cwd !== this.rootDirectory) {
- const message = `Execution CWD validation failed: Attempted path "${cwd}" resolves outside the allowed root directory "${this.rootDirectory}".`;
- return {
- llmContent: `Command rejected: ${params.command}\nReason: ${message}`,
- returnDisplay: `Error: ${message}`,
- };
- }
-
return new Promise((resolve) => {
- const spawnOptions: SpawnOptions = {
- cwd,
- shell: true,
- env: { ...process.env },
- stdio: 'pipe',
- windowsHide: true,
- timeout: timeout,
+ const queuedItem: QueuedCommand = {
+ params,
+ resolve,
+ reject: (error) =>
+ resolve({
+ llmContent: `Internal tool error for command: ${params.command}\nError: ${error.message}`,
+ returnDisplay: `Internal Tool Error: ${error.message}`,
+ }),
+ confirmationDetails: false,
};
- let stdout = '';
- let stderr = '';
- let processError: Error | null = null;
- let timedOut = false;
+ this.commandQueue.push(queuedItem);
+ setImmediate(() => this.triggerQueueProcessing());
+ });
+ }
- try {
- const child = spawn(params.command, spawnOptions);
- child.stdout!.on('data', (data) => {
- stdout += data.toString();
- if (stdout.length > MAX_OUTPUT_LENGTH) {
- stdout = this.truncateOutput(stdout);
- child.stdout!.pause();
+ private async triggerQueueProcessing(): Promise<void> {
+ if (this.isExecuting || this.commandQueue.length === 0) {
+ return;
+ }
+ this.isExecuting = true;
+ const { params, resolve, reject } = this.commandQueue.shift()!;
+ try {
+ await this.shellReady;
+ if (!this.bashProcess || this.bashProcess.killed) {
+ throw new Error(
+ 'Persistent bash process is not available or was killed.',
+ );
+ }
+ const result = await this.executeCommandInShell(params);
+ resolve(result);
+ } catch (error: unknown) {
+ console.error(`Error executing command "${params.command}":`, error);
+ if (error instanceof Error) {
+ reject(error);
+ } else {
+ reject(new Error('Unknown error occurred: ' + JSON.stringify(error)));
+ }
+ } finally {
+ this.isExecuting = false;
+ setImmediate(() => this.triggerQueueProcessing());
+ }
+ }
+
+ private executeCommandInShell(
+ params: TerminalToolParams,
+ ): Promise<ToolResult> {
+ let tempStdoutPath: string | null = null;
+ let tempStderrPath: string | null = null;
+ let originalResolve: (value: ToolResult | PromiseLike<ToolResult>) => void;
+ let originalReject: (reason?: unknown) => void;
+ const promise = new Promise<ToolResult>((resolve, reject) => {
+ originalResolve = resolve;
+ originalReject = reject;
+ if (!this.bashProcess) {
+ return reject(
+ new Error('Bash process is not running. Cannot execute command.'),
+ );
+ }
+ const isBackgroundTask = params.runInBackground ?? false;
+ const commandUUID = crypto.randomUUID();
+ const startDelimiter = `::START_CMD_${commandUUID}::`;
+ const endDelimiter = `::END_CMD_${commandUUID}::`;
+ const exitCodeDelimiter = `::EXIT_CODE_${commandUUID}::`;
+ const pidDelimiter = `::PID_${commandUUID}::`;
+ if (isBackgroundTask) {
+ try {
+ const tempDir = os.tmpdir();
+ tempStdoutPath = path.join(tempDir, `term_out_${commandUUID}.log`);
+ tempStderrPath = path.join(tempDir, `term_err_${commandUUID}.log`);
+ } catch (err: unknown) {
+ return reject(
+ new Error(
+ `Failed to determine temporary directory: ${getErrorMessage(err)}`,
+ ),
+ );
+ }
+ }
+ let stdoutBuffer = '';
+ let stderrBuffer = '';
+ let commandOutputStarted = false;
+ let exitCode: number | null = null;
+ let backgroundPid: number | null = null;
+ let receivedEndDelimiter = false;
+ const effectiveTimeout = isBackgroundTask
+ ? BACKGROUND_LAUNCH_TIMEOUT_MS
+ : Math.min(
+ params.timeout ?? DEFAULT_TIMEOUT_MS,
+ MAX_TIMEOUT_OVERRIDE_MS,
+ );
+ let onStdoutData: ((data: Buffer) => void) | null = null;
+ let onStderrData: ((data: Buffer) => void) | null = null;
+ let launchTimeoutId: NodeJS.Timeout | null = null;
+ launchTimeoutId = setTimeout(() => {
+ const timeoutMessage = isBackgroundTask
+ ? `Background command launch timed out after ${effectiveTimeout}ms.`
+ : `Command timed out after ${effectiveTimeout}ms.`;
+ if (!isBackgroundTask && this.bashProcess && !this.bashProcess.killed) {
+ try {
+ this.bashProcess.stdin.write('\x03');
+ } catch (e: unknown) {
+ console.error('Error writing SIGINT on timeout:', e);
}
+ }
+ const listenersToClean = { onStdoutData, onStderrData };
+ cleanupListeners(listenersToClean);
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ this.cleanupTempFiles(tempStdoutPath, tempStderrPath).catch((err) => {
+ console.warn(
+ `Error cleaning up temp files on timeout: ${err.message}`,
+ );
+ });
+ }
+ originalResolve({
+ llmContent: `Command execution failed: ${timeoutMessage}\nCommand: ${params.command}\nExecuted in: ${this.currentCwd}\n${isBackgroundTask ? 'Mode: Background Launch' : `Mode: Foreground\nTimeout Limit: ${effectiveTimeout}ms`}\nPartial Stdout (Launch):\n${this.truncateOutput(stdoutBuffer)}\nPartial Stderr (Launch):\n${this.truncateOutput(stderrBuffer)}\nNote: ${isBackgroundTask ? 'Launch failed or took too long.' : 'Attempted interrupt (SIGINT). Shell state might be unpredictable if command ignored interrupt.'}`,
+ returnDisplay: `Timeout: ${timeoutMessage}`,
});
- child.stderr!.on('data', (data) => {
- stderr += data.toString();
- if (stderr.length > MAX_OUTPUT_LENGTH) {
- stderr = this.truncateOutput(stderr);
- child.stderr!.pause();
+ }, effectiveTimeout);
+ const processDataChunk = (chunk: string, isStderr: boolean): boolean => {
+ let dataToProcess = chunk;
+ if (!commandOutputStarted) {
+ const startIndex = dataToProcess.indexOf(startDelimiter);
+ if (startIndex !== -1) {
+ commandOutputStarted = true;
+ dataToProcess = dataToProcess.substring(
+ startIndex + startDelimiter.length,
+ );
+ } else {
+ return false;
}
- });
- child.on('error', (err) => {
- processError = err;
+ }
+ const pidIndex = dataToProcess.indexOf(pidDelimiter);
+ if (pidIndex !== -1) {
+ const pidMatch = dataToProcess
+ .substring(pidIndex + pidDelimiter.length)
+ .match(/^(\d+)/);
+ if (pidMatch?.[1]) {
+ backgroundPid = parseInt(pidMatch[1], 10);
+ const pidEndIndex =
+ pidIndex + pidDelimiter.length + pidMatch[1].length;
+ const beforePid = dataToProcess.substring(0, pidIndex);
+ if (isStderr) stderrBuffer += beforePid;
+ else stdoutBuffer += beforePid;
+ dataToProcess = dataToProcess.substring(pidEndIndex);
+ } else {
+ const beforePid = dataToProcess.substring(0, pidIndex);
+ if (isStderr) stderrBuffer += beforePid;
+ else stdoutBuffer += beforePid;
+ dataToProcess = dataToProcess.substring(
+ pidIndex + pidDelimiter.length,
+ );
+ }
+ }
+ const exitCodeIndex = dataToProcess.indexOf(exitCodeDelimiter);
+ if (exitCodeIndex !== -1) {
+ const exitCodeMatch = dataToProcess
+ .substring(exitCodeIndex + exitCodeDelimiter.length)
+ .match(/^(\d+)/);
+ if (exitCodeMatch?.[1]) {
+ exitCode = parseInt(exitCodeMatch[1], 10);
+ const beforeExitCode = dataToProcess.substring(0, exitCodeIndex);
+ if (isStderr) stderrBuffer += beforeExitCode;
+ else stdoutBuffer += beforeExitCode;
+ dataToProcess = dataToProcess.substring(
+ exitCodeIndex +
+ exitCodeDelimiter.length +
+ exitCodeMatch[1].length,
+ );
+ } else {
+ const beforeExitCode = dataToProcess.substring(0, exitCodeIndex);
+ if (isStderr) stderrBuffer += beforeExitCode;
+ else stdoutBuffer += beforeExitCode;
+ dataToProcess = dataToProcess.substring(
+ exitCodeIndex + exitCodeDelimiter.length,
+ );
+ }
+ }
+ const endDelimiterIndex = dataToProcess.indexOf(endDelimiter);
+ if (endDelimiterIndex !== -1) {
+ receivedEndDelimiter = true;
+ const beforeEndDelimiter = dataToProcess.substring(
+ 0,
+ endDelimiterIndex,
+ );
+ if (isStderr) stderrBuffer += beforeEndDelimiter;
+ else stdoutBuffer += beforeEndDelimiter;
+ const afterEndDelimiter = dataToProcess.substring(
+ endDelimiterIndex + endDelimiter.length,
+ );
+ const exitCodeEchoMatch = afterEndDelimiter.match(/^(\d+)/);
+ dataToProcess = exitCodeEchoMatch
+ ? afterEndDelimiter.substring(exitCodeEchoMatch[1].length)
+ : afterEndDelimiter;
+ }
+ if (dataToProcess.length > 0) {
+ if (isStderr) stderrBuffer += dataToProcess;
+ else stdoutBuffer += dataToProcess;
+ }
+ if (receivedEndDelimiter && exitCode !== null) {
+ setImmediate(cleanupAndResolve);
+ return true;
+ }
+ return false;
+ };
+ onStdoutData = (data: Buffer) => processDataChunk(data.toString(), false);
+ onStderrData = (data: Buffer) => processDataChunk(data.toString(), true);
+ const cleanupListeners = (listeners?: {
+ onStdoutData: ((data: Buffer) => void) | null;
+ onStderrData: ((data: Buffer) => void) | null;
+ }) => {
+ if (launchTimeoutId) clearTimeout(launchTimeoutId);
+ launchTimeoutId = null;
+ const stdoutListener = listeners?.onStdoutData ?? onStdoutData;
+ const stderrListener = listeners?.onStderrData ?? onStderrData;
+ if (this.bashProcess && !this.bashProcess.killed) {
+ if (stdoutListener)
+ this.bashProcess.stdout.removeListener('data', stdoutListener);
+ if (stderrListener)
+ this.bashProcess.stderr.removeListener('data', stderrListener);
+ }
+ if (this.currentCommandCleanup === cleanupListeners) {
+ this.currentCommandCleanup = null;
+ }
+ onStdoutData = null;
+ onStderrData = null;
+ };
+ this.currentCommandCleanup = cleanupListeners;
+ const cleanupAndResolve = async () => {
+ if (
+ !this.currentCommandCleanup ||
+ this.currentCommandCleanup !== cleanupListeners
+ ) {
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ this.cleanupTempFiles(tempStdoutPath, tempStderrPath).catch(
+ (err) => {
+ console.warn(
+ `Error cleaning up temp files for superseded command: ${err.message}`,
+ );
+ },
+ );
+ }
+ return;
+ }
+ const launchStdout = this.truncateOutput(stdoutBuffer);
+ const launchStderr = this.truncateOutput(stderrBuffer);
+ const listenersToClean = { onStdoutData, onStderrData };
+ cleanupListeners(listenersToClean);
+ if (exitCode === null) {
console.error(
- `TerminalLogic spawn error for "${params.command}":`,
- err,
+ `CRITICAL: Command "${params.command}" (background: ${isBackgroundTask}) finished delimiter processing but exitCode is null.`,
);
- });
- child.on('close', (code, signal) => {
- const exitCode = code ?? (signal ? -1 : -2);
- if (signal === 'SIGTERM' || signal === 'SIGKILL') {
- if (child.killed && timeout > 0) timedOut = true;
+ const errorMode = isBackgroundTask
+ ? 'Background Launch'
+ : 'Foreground';
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ await this.cleanupTempFiles(tempStdoutPath, tempStderrPath);
+ }
+ originalResolve({
+ llmContent: `Command: ${params.command}\nExecuted in: ${this.currentCwd}\nMode: ${errorMode}\nExit Code: -2 (Internal Error: Exit code not captured)\nStdout (during setup):\n${launchStdout}\nStderr (during setup):\n${launchStderr}`,
+ returnDisplay:
+ `Internal Error: Failed to capture command exit code.\n${launchStdout}\nStderr: ${launchStderr}`.trim(),
+ });
+ return;
+ }
+ let cwdUpdateError = '';
+ if (!isBackgroundTask) {
+ const mightChangeCwd = params.command.trim().startsWith('cd ');
+ if (exitCode === 0 || mightChangeCwd) {
+ try {
+ const latestCwd = await this.getCurrentShellCwd();
+ if (this.currentCwd !== latestCwd) {
+ this.currentCwd = latestCwd;
+ }
+ } catch (e: unknown) {
+ if (exitCode === 0) {
+ cwdUpdateError = `\nWarning: Failed to verify/update current working directory after command: ${getErrorMessage(e)}`;
+ console.error(
+ 'Failed to update CWD after successful command:',
+ e,
+ );
+ }
+ }
}
- const finalStdout = this.truncateOutput(stdout);
- const finalStderr = this.truncateOutput(stderr);
- let llmContent = `Command: ${params.command}\nExecuted in: ${cwd}\nExit Code: ${exitCode}\n`;
- if (timedOut) llmContent += `Status: Timed Out after ${timeout}ms\n`;
- if (processError)
- llmContent += `Process Error: ${processError.message}\n`;
- llmContent += `Stdout:\n${finalStdout}\nStderr:\n${finalStderr}`;
- let displayOutput = finalStderr.trim() || finalStdout.trim();
- if (timedOut)
- displayOutput = `Timeout: ${displayOutput || 'No output before timeout'}`;
- else if (exitCode !== 0 && !displayOutput)
- displayOutput = `Failed (Exit Code: ${exitCode})`;
- else if (exitCode === 0 && !displayOutput)
+ }
+ if (isBackgroundTask) {
+ const launchSuccess = exitCode === 0;
+ const pidString =
+ backgroundPid !== null ? backgroundPid.toString() : 'Not Captured';
+ if (
+ launchSuccess &&
+ backgroundPid !== null &&
+ tempStdoutPath &&
+ tempStderrPath
+ ) {
+ this.inspectBackgroundProcess(
+ backgroundPid,
+ params.command,
+ this.currentCwd,
+ launchStdout,
+ launchStderr,
+ tempStdoutPath,
+ tempStderrPath,
+ originalResolve,
+ );
+ } else {
+ const reason =
+ backgroundPid === null
+ ? 'PID not captured'
+ : `Launch failed (Exit Code: ${exitCode})`;
+ const displayMessage = `Failed to launch process in background (${reason})`;
+ console.error(
+ `Background launch failed for command: ${params.command}. Reason: ${reason}`,
+ );
+ if (tempStdoutPath && tempStderrPath) {
+ await this.cleanupTempFiles(tempStdoutPath, tempStderrPath);
+ }
+ originalResolve({
+ llmContent: `Background Command Launch Failed: ${params.command}\nExecuted in: ${this.currentCwd}\nReason: ${reason}\nPID: ${pidString}\nExit Code (Launch): ${exitCode}\nStdout (During Launch):\n${launchStdout}\nStderr (During Launch):\n${launchStderr}`,
+ returnDisplay: displayMessage,
+ });
+ }
+ } else {
+ let displayOutput = '';
+ const stdoutTrimmed = launchStdout.trim();
+ const stderrTrimmed = launchStderr.trim();
+ if (stderrTrimmed) {
+ displayOutput = stderrTrimmed;
+ } else if (stdoutTrimmed) {
+ displayOutput = stdoutTrimmed;
+ }
+ if (exitCode !== 0 && !displayOutput) {
+ displayOutput = `Failed with exit code: ${exitCode}`;
+ } else if (exitCode === 0 && !displayOutput) {
displayOutput = `Success (no output)`;
- resolve({
- llmContent,
+ }
+ originalResolve({
+ llmContent: `Command: ${params.command}\nExecuted in: ${this.currentCwd}\nExit Code: ${exitCode}\nStdout:\n${launchStdout}\nStderr:\n${launchStderr}${cwdUpdateError}`,
returnDisplay: displayOutput.trim() || `Exit Code: ${exitCode}`,
});
- });
- } catch (spawnError: unknown) {
- const errMsg = getErrorMessage(spawnError);
+ }
+ };
+ if (!this.bashProcess || this.bashProcess.killed) {
console.error(
- `TerminalLogic failed to spawn "${params.command}":`,
- spawnError,
+ 'Bash process lost or killed before listeners could be attached.',
+ );
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ this.cleanupTempFiles(tempStdoutPath, tempStderrPath).catch((err) => {
+ console.warn(
+ `Error cleaning up temp files on attach failure: ${err.message}`,
+ );
+ });
+ }
+ return originalReject(
+ new Error(
+ 'Bash process lost or killed before listeners could be attached.',
+ ),
+ );
+ }
+ if (onStdoutData) this.bashProcess.stdout.on('data', onStdoutData);
+ if (onStderrData) this.bashProcess.stderr.on('data', onStderrData);
+ let commandToWrite: string;
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ commandToWrite = `echo "${startDelimiter}"; { { ${params.command} > "${tempStdoutPath}" 2> "${tempStderrPath}"; } & } 2>/dev/null; __LAST_PID=$!; echo "${pidDelimiter}$__LAST_PID" >&2; echo "${exitCodeDelimiter}$?" >&2; echo "${endDelimiter}$?" >&1\n`;
+ } else if (!isBackgroundTask) {
+ commandToWrite = `echo "${startDelimiter}"; ${params.command}; __EXIT_CODE=$?; echo "${exitCodeDelimiter}$__EXIT_CODE" >&2; echo "${endDelimiter}$__EXIT_CODE" >&1\n`;
+ } else {
+ return originalReject(
+ new Error(
+ 'Internal setup error: Missing temporary file paths for background execution.',
+ ),
+ );
+ }
+ try {
+ if (this.bashProcess?.stdin?.writable) {
+ this.bashProcess.stdin.write(commandToWrite, (err) => {
+ if (err) {
+ console.error(
+ `Error writing command "${params.command}" to bash stdin (callback):`,
+ err,
+ );
+ const listenersToClean = { onStdoutData, onStderrData };
+ cleanupListeners(listenersToClean);
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ this.cleanupTempFiles(tempStdoutPath, tempStderrPath).catch(
+ (e) => console.warn(`Cleanup failed: ${e.message}`),
+ );
+ }
+ originalReject(
+ new Error(
+ `Shell stdin write error: ${err.message}. Command likely did not execute.`,
+ ),
+ );
+ }
+ });
+ } else {
+ throw new Error(
+ 'Shell stdin is not writable or process closed when attempting to write command.',
+ );
+ }
+ } catch (e: unknown) {
+ console.error(
+ `Error writing command "${params.command}" to bash stdin (sync):`,
+ e,
+ );
+ const listenersToClean = { onStdoutData, onStderrData };
+ cleanupListeners(listenersToClean);
+ if (isBackgroundTask && tempStdoutPath && tempStderrPath) {
+ this.cleanupTempFiles(tempStdoutPath, tempStderrPath).catch((err) =>
+ console.warn(`Cleanup failed: ${err.message}`),
+ );
+ }
+ originalReject(
+ new Error(
+ `Shell stdin write exception: ${getErrorMessage(e)}. Command likely did not execute.`,
+ ),
);
- resolve({
- llmContent: `Failed to start command: ${params.command}\nError: ${errMsg}`,
- returnDisplay: `Error spawning command: ${errMsg}`,
- });
}
});
+ return promise;
}
- private truncateOutput(
- output: string,
- limit: number = MAX_OUTPUT_LENGTH,
- ): string {
- if (output.length > limit) {
+ private async inspectBackgroundProcess(
+ pid: number,
+ command: string,
+ cwd: string,
+ initialStdout: string,
+ initialStderr: string,
+ tempStdoutPath: string,
+ tempStderrPath: string,
+ resolve: (value: ToolResult | PromiseLike<ToolResult>) => void,
+ ): Promise<void> {
+ let finalStdout = '';
+ let finalStderr = '';
+ let llmAnalysis = '';
+ let fileReadError = '';
+ try {
+ const { status, summary } = await this.backgroundTerminalAnalyzer.analyze(
+ pid,
+ tempStdoutPath,
+ tempStderrPath,
+ command,
+ );
+ if (status === 'Unknown') llmAnalysis = `LLM analysis failed: ${summary}`;
+ else llmAnalysis = summary;
+ } catch (llmerror: unknown) {
+ console.error(
+ `LLM analysis failed for PID ${pid} command "${command}":`,
+ llmerror,
+ );
+ llmAnalysis = `LLM analysis failed: ${getErrorMessage(llmerror)}`;
+ }
+ try {
+ finalStdout = await fs.readFile(tempStdoutPath, 'utf-8');
+ finalStderr = await fs.readFile(tempStderrPath, 'utf-8');
+ } catch (err: unknown) {
+ console.error(`Error reading temp output files for PID ${pid}:`, err);
+ fileReadError = `\nWarning: Failed to read temporary output files (${getErrorMessage(err)}). Final output may be incomplete.`;
+ }
+ await this.cleanupTempFiles(tempStdoutPath, tempStderrPath);
+ const truncatedFinalStdout = this.truncateOutput(finalStdout);
+ const truncatedFinalStderr = this.truncateOutput(finalStderr);
+ resolve({
+ llmContent: `Background Command: ${command}\nLaunched in: ${cwd}\nPID: ${pid}\n--- LLM Analysis ---\n${llmAnalysis}\n--- Final Stdout (from ${path.basename(tempStdoutPath)}) ---\n${truncatedFinalStdout}\n--- Final Stderr (from ${path.basename(tempStderrPath)}) ---\n${truncatedFinalStderr}\n--- Launch Stdout ---\n${initialStdout}\n--- Launch Stderr ---\n${initialStderr}${fileReadError}`,
+ returnDisplay: `(PID: ${pid}): ${this.truncateOutput(llmAnalysis, 200)}`,
+ });
+ }
+
+ private async cleanupTempFiles(
+ stdoutPath: string | null,
+ stderrPath: string | null,
+ ): Promise<void> {
+ const unlinkQuietly = async (filePath: string | null) => {
+ if (!filePath) return;
+ try {
+ await fs.unlink(filePath);
+ } catch (err: unknown) {
+ if (!isNodeError(err) || err.code !== 'ENOENT') {
+ console.warn(
+ `Failed to delete temporary file '${filePath}': ${getErrorMessage(err)}`,
+ );
+ }
+ }
+ };
+ await Promise.all([unlinkQuietly(stdoutPath), unlinkQuietly(stderrPath)]);
+ }
+
+ private getCurrentShellCwd(): Promise<string> {
+ return new Promise((resolve, reject) => {
+ if (
+ !this.bashProcess ||
+ !this.bashProcess.stdin?.writable ||
+ this.bashProcess.killed
+ ) {
+ return reject(
+ new Error(
+ 'Shell not running, stdin not writable, or killed for PWD check',
+ ),
+ );
+ }
+ const pwdUuid = crypto.randomUUID();
+ const pwdDelimiter = `::PWD_${pwdUuid}::`;
+ let pwdOutput = '';
+ let onPwdData: ((data: Buffer) => void) | null = null;
+ let onPwdError: ((data: Buffer) => void) | null = null;
+ let pwdTimeoutId: NodeJS.Timeout | null = null;
+ let finished = false;
+ const cleanupPwdListeners = (err?: Error) => {
+ if (finished) return;
+ finished = true;
+ if (pwdTimeoutId) clearTimeout(pwdTimeoutId);
+ pwdTimeoutId = null;
+ const stdoutListener = onPwdData;
+ const stderrListener = onPwdError;
+ onPwdData = null;
+ onPwdError = null;
+ if (this.bashProcess && !this.bashProcess.killed) {
+ if (stdoutListener)
+ this.bashProcess.stdout.removeListener('data', stdoutListener);
+ if (stderrListener)
+ this.bashProcess.stderr.removeListener('data', stderrListener);
+ }
+ if (err) {
+ reject(err);
+ } else {
+ resolve(pwdOutput.trim());
+ }
+ };
+ onPwdData = (data: Buffer) => {
+ if (!onPwdData) return;
+ const dataStr = data.toString();
+ const delimiterIndex = dataStr.indexOf(pwdDelimiter);
+ if (delimiterIndex !== -1) {
+ pwdOutput += dataStr.substring(0, delimiterIndex);
+ cleanupPwdListeners();
+ } else {
+ pwdOutput += dataStr;
+ }
+ };
+ onPwdError = (data: Buffer) => {
+ if (!onPwdError) return;
+ const dataStr = data.toString();
+ console.error(`Error during PWD check: ${dataStr}`);
+ cleanupPwdListeners(
+ new Error(
+ `Stderr received during pwd check: ${this.truncateOutput(dataStr, 100)}`,
+ ),
+ );
+ };
+ this.bashProcess.stdout.on('data', onPwdData);
+ this.bashProcess.stderr.on('data', onPwdError);
+ pwdTimeoutId = setTimeout(() => {
+ cleanupPwdListeners(new Error('Timeout waiting for pwd response'));
+ }, 5000);
+ try {
+ const pwdCommand = `printf "%s" "$PWD"; printf "${pwdDelimiter}";\n`;
+ if (this.bashProcess?.stdin?.writable) {
+ this.bashProcess.stdin.write(pwdCommand, (err) => {
+ if (err) {
+ console.error('Error writing pwd command (callback):', err);
+ cleanupPwdListeners(
+ new Error(`Failed to write pwd command: ${err.message}`),
+ );
+ }
+ });
+ } else {
+ throw new Error('Shell stdin not writable for pwd command.');
+ }
+ } catch (e: unknown) {
+ console.error('Exception writing pwd command:', e);
+ cleanupPwdListeners(
+ new Error(`Exception writing pwd command: ${getErrorMessage(e)}`),
+ );
+ }
+ });
+ }
+
+ private truncateOutput(output: string, limit?: number): string {
+ const effectiveLimit = limit ?? this.outputLimit;
+ if (output.length > effectiveLimit) {
return (
- output.substring(0, limit) +
- `\n... [Output truncated at ${limit} characters]`
+ output.substring(0, effectiveLimit) +
+ `\n... [Output truncated at ${effectiveLimit} characters]`
);
}
return output;
}
+
+ private clearQueue(error: Error) {
+ const queue = this.commandQueue;
+ this.commandQueue = [];
+ queue.forEach(({ resolve, params }) =>
+ resolve({
+ llmContent: `Command cancelled: ${params.command}\nReason: ${error.message}`,
+ returnDisplay: `Command Cancelled: ${error.message}`,
+ }),
+ );
+ }
+
+ destroy() {
+ this.rejectShellReady?.(
+ new Error('BashTool destroyed during initialization or operation.'),
+ );
+ this.rejectShellReady = undefined;
+ this.resolveShellReady = undefined;
+ this.clearQueue(new Error('BashTool is being destroyed.'));
+ try {
+ this.currentCommandCleanup?.();
+ } catch (e) {
+ console.warn('Error during current command cleanup:', e);
+ }
+ if (this.bashProcess) {
+ const proc = this.bashProcess;
+ const pid = proc.pid;
+ this.bashProcess = null;
+ proc.stdout?.removeAllListeners();
+ proc.stderr?.removeAllListeners();
+ proc.removeAllListeners('error');
+ proc.removeAllListeners('close');
+ proc.stdin?.end();
+ try {
+ proc.kill('SIGTERM');
+ setTimeout(() => {
+ if (!proc.killed) {
+ proc.kill('SIGKILL');
+ }
+ }, 500);
+ } catch (e: unknown) {
+ console.warn(
+ `Error trying to kill bash process PID: ${pid}: ${getErrorMessage(e)}`,
+ );
+ }
+ }
+ }
}
diff --git a/packages/server/src/tools/tools.ts b/packages/server/src/tools/tools.ts
index 4851f164..ed7c017a 100644
--- a/packages/server/src/tools/tools.ts
+++ b/packages/server/src/tools/tools.ts
@@ -49,7 +49,14 @@ export interface Tool<
*/
getDescription(params: TParams): string;
- // Removed shouldConfirmExecute as it's UI-specific
+ /**
+ * Determines if the tool should prompt for confirmation before execution
+ * @param params Parameters for the tool execution
+ * @returns Whether execute should be confirmed.
+ */
+ shouldConfirmExecute(
+ params: TParams,
+ ): Promise<ToolCallConfirmationDetails | false>;
/**
* Executes the tool with the given parameters
@@ -115,7 +122,17 @@ export abstract class BaseTool<
return JSON.stringify(params);
}
- // Removed shouldConfirmExecute as it's UI-specific
+ /**
+ * Determines if the tool should prompt for confirmation before execution
+ * @param params Parameters for the tool execution
+ * @returns Whether or not execute should be confirmed by the user.
+ */
+ shouldConfirmExecute(
+ // eslint-disable-next-line @typescript-eslint/no-unused-vars
+ params: TParams,
+ ): Promise<ToolCallConfirmationDetails | false> {
+ return Promise.resolve(false);
+ }
/**
* Abstract method to execute the tool with the given parameters
@@ -148,3 +165,27 @@ export type ToolResultDisplay = string | FileDiff;
export interface FileDiff {
fileDiff: string;
}
+
+export interface ToolCallConfirmationDetails {
+ title: string;
+ onConfirm: (outcome: ToolConfirmationOutcome) => Promise<void>;
+}
+
+export interface ToolEditConfirmationDetails
+ extends ToolCallConfirmationDetails {
+ fileName: string;
+ fileDiff: string;
+}
+
+export interface ToolExecuteConfirmationDetails
+ extends ToolCallConfirmationDetails {
+ command: string;
+ rootCommand: string;
+ description: string;
+}
+
+export enum ToolConfirmationOutcome {
+ ProceedOnce,
+ ProceedAlways,
+ Cancel,
+}
diff --git a/packages/server/src/tools/web-fetch.ts b/packages/server/src/tools/web-fetch.ts
index 29e33fbe..415dc033 100644
--- a/packages/server/src/tools/web-fetch.ts
+++ b/packages/server/src/tools/web-fetch.ts
@@ -21,14 +21,14 @@ export interface WebFetchToolParams {
/**
* Implementation of the WebFetch tool logic (moved from CLI)
*/
-export class WebFetchLogic extends BaseTool<WebFetchToolParams, ToolResult> {
+export class WebFetchTool extends BaseTool<WebFetchToolParams, ToolResult> {
static readonly Name: string = 'web_fetch';
constructor() {
super(
- WebFetchLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ WebFetchTool.Name,
+ 'WebFetch',
+ 'Fetches text content from a given URL. Handles potential network errors and non-success HTTP status codes.',
{
properties: {
url: {
diff --git a/packages/server/src/tools/write-file.ts b/packages/server/src/tools/write-file.ts
index ce723061..814efa86 100644
--- a/packages/server/src/tools/write-file.ts
+++ b/packages/server/src/tools/write-file.ts
@@ -7,7 +7,14 @@
import fs from 'fs';
import path from 'path';
import * as Diff from 'diff'; // Keep for result generation
-import { BaseTool, ToolResult, FileDiff } from './tools.js'; // Updated import (Removed ToolResultDisplay)
+import {
+ BaseTool,
+ ToolResult,
+ FileDiff,
+ ToolEditConfirmationDetails,
+ ToolConfirmationOutcome,
+ ToolCallConfirmationDetails,
+} from './tools.js'; // Updated import (Removed ToolResultDisplay)
import { SchemaValidator } from '../utils/schemaValidator.js'; // Updated import
import { makeRelative, shortenPath } from '../utils/paths.js'; // Updated import
import { isNodeError } from '../utils/errors.js'; // Import isNodeError
@@ -30,16 +37,17 @@ export interface WriteFileToolParams {
/**
* Implementation of the WriteFile tool logic (moved from CLI)
*/
-export class WriteFileLogic extends BaseTool<WriteFileToolParams, ToolResult> {
+export class WriteFileTool extends BaseTool<WriteFileToolParams, ToolResult> {
static readonly Name: string = 'write_file';
+ private shouldAlwaysWrite = false;
private readonly rootDirectory: string;
constructor(rootDirectory: string) {
super(
- WriteFileLogic.Name,
- '', // Display name handled by CLI wrapper
- '', // Description handled by CLI wrapper
+ WriteFileTool.Name,
+ 'WriteFile',
+ 'Writes content to a specified file in the local filesystem.',
{
properties: {
file_path: {
@@ -98,6 +106,56 @@ export class WriteFileLogic extends BaseTool<WriteFileToolParams, ToolResult> {
return `Writing to ${shortenPath(relativePath)}`;
}
+ /**
+ * Handles the confirmation prompt for the WriteFile tool in the CLI.
+ */
+ async shouldConfirmExecute(
+ params: WriteFileToolParams,
+ ): Promise<ToolCallConfirmationDetails | false> {
+ if (this.shouldAlwaysWrite) {
+ return false;
+ }
+
+ const validationError = this.validateToolParams(params);
+ if (validationError) {
+ console.error(
+ `[WriteFile Wrapper] Attempted confirmation with invalid parameters: ${validationError}`,
+ );
+ return false;
+ }
+
+ const relativePath = makeRelative(params.file_path, this.rootDirectory);
+ const fileName = path.basename(params.file_path);
+
+ let currentContent = '';
+ try {
+ currentContent = fs.readFileSync(params.file_path, 'utf8');
+ } catch {
+ // File might not exist, that's okay for write/create
+ }
+
+ const fileDiff = Diff.createPatch(
+ fileName,
+ currentContent,
+ params.content,
+ 'Current',
+ 'Proposed',
+ { context: 3 },
+ );
+
+ const confirmationDetails: ToolEditConfirmationDetails = {
+ title: `Confirm Write: ${shortenPath(relativePath)}`,
+ fileName,
+ fileDiff,
+ onConfirm: async (outcome: ToolConfirmationOutcome) => {
+ if (outcome === ToolConfirmationOutcome.ProceedAlways) {
+ this.shouldAlwaysWrite = true;
+ }
+ },
+ };
+ return confirmationDetails;
+ }
+
async execute(params: WriteFileToolParams): Promise<ToolResult> {
const validationError = this.validateParams(params);
if (validationError) {
diff --git a/packages/server/src/utils/BackgroundTerminalAnalyzer.ts b/packages/server/src/utils/BackgroundTerminalAnalyzer.ts
new file mode 100644
index 00000000..31cd91c6
--- /dev/null
+++ b/packages/server/src/utils/BackgroundTerminalAnalyzer.ts
@@ -0,0 +1,473 @@
+/**
+ * @license
+ * Copyright 2025 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { Content, SchemaUnion, Type } from '@google/genai';
+import {
+ Config,
+ getErrorMessage,
+ isNodeError,
+ GeminiClient,
+} from '@gemini-code/server';
+import { promises as fs } from 'fs';
+import { exec as _exec } from 'child_process';
+import { promisify } from 'util';
+
+// Define the AnalysisStatus type alias
+type AnalysisStatus =
+ | 'Running'
+ | 'SuccessReported'
+ | 'ErrorReported'
+ | 'Unknown'
+ | 'AnalysisFailed';
+
+// Promisify child_process.exec for easier async/await usage
+const execAsync = promisify(_exec);
+
+// Define the expected interface for the AI client dependency
+export interface AiClient {
+ generateJson(
+ prompt: Content[], // Keep flexible or define a stricter prompt structure type
+ schema: SchemaUnion,
+ // eslint-disable-next-line @typescript-eslint/no-explicit-any
+ ): Promise<any>; // Ideally, specify the expected JSON structure TAnalysisResult | TAnalysisFailure
+}
+
+// Identifier for the background process (e.g., PID)
+// Using `unknown` allows more flexibility than `object` while still being type-safe
+export type ProcessHandle = number | string | unknown;
+
+// Represents the structure expected from a successful LLM analysis call
+export interface AnalysisResult {
+ summary: string;
+ inferredStatus: 'Running' | 'SuccessReported' | 'ErrorReported' | 'Unknown';
+}
+
+// Represents the structure returned when the LLM analysis itself fails
+export interface AnalysisFailure {
+ error: string;
+ inferredStatus: 'AnalysisFailed';
+}
+
+// Type guard to check if the result is a failure object
+function isAnalysisFailure(
+ result: AnalysisResult | AnalysisFailure,
+): result is AnalysisFailure {
+ return (result as AnalysisFailure).inferredStatus === 'AnalysisFailed';
+}
+
+// Represents the final outcome after polling is complete (or failed/timed out)
+export interface FinalAnalysisOutcome {
+ status: string; // e.g., 'Completed_SuccessReported', 'TimedOut_Running', 'AnalysisFailed'
+ summary: string; // Final summary or error message
+}
+
+export class BackgroundTerminalAnalyzer {
+ private geminiClient: GeminiClient | null = null;
+ private readonly maxOutputAnalysisLength = 20000;
+ private pollIntervalMs: number;
+ private maxAttempts: number;
+ private initialDelayMs: number;
+
+ constructor(
+ config: Config, // Accept Config object
+ options: {
+ pollIntervalMs?: number;
+ maxAttempts?: number;
+ initialDelayMs?: number;
+ } = {},
+ ) {
+ try {
+ // Initialize Gemini client using config
+ this.geminiClient = new GeminiClient(
+ config.getApiKey(),
+ config.getModel(),
+ );
+ } catch (error) {
+ console.error(
+ 'Failed to initialize GeminiClient in BackgroundTerminalAnalyzer:',
+ error,
+ );
+ // Set client to null so analyzeOutput handles it
+ this.geminiClient = null;
+ }
+ this.pollIntervalMs = options.pollIntervalMs ?? 5000; // Default 5 seconds
+ this.maxAttempts = options.maxAttempts ?? 6; // Default 6 attempts (approx 30s total)
+ this.initialDelayMs = options.initialDelayMs ?? 500; // Default 0.5s initial delay
+ }
+
+ /**
+ * Polls the output of a background process using an LLM
+ * until a conclusive status is determined or timeout occurs.
+ * @param pid The handle/identifier of the background process (typically PID number).
+ * @param tempStdoutFilePath Path to the temporary file capturing stdout.
+ * @param tempStderrFilePath Path to the temporary file capturing stderr.
+ * @param command The command string that was executed (for context in prompts).
+ * @returns A promise resolving to the final analysis outcome.
+ */
+ async analyze(
+ pid: ProcessHandle,
+ tempStdoutFilePath: string,
+ tempStderrFilePath: string,
+ command: string,
+ ): Promise<FinalAnalysisOutcome> {
+ // --- Validate PID ---
+ if (typeof pid !== 'number' || !Number.isInteger(pid) || pid <= 0) {
+ console.error(
+ `BackgroundTerminalAnalyzer: Invalid or non-numeric PID provided (${pid}). Analysis cannot proceed.`,
+ );
+ return {
+ status: 'AnalysisFailed',
+ summary: 'Invalid PID provided for analysis.',
+ };
+ }
+
+ // --- Initial Delay ---
+ // Wait briefly before the first check to allow the process to initialize
+ // and potentially write initial output.
+ await new Promise((resolve) => setTimeout(resolve, this.initialDelayMs));
+
+ let attempts = 0;
+ let lastAnalysisResult: AnalysisResult | AnalysisFailure | null = null;
+
+ while (attempts < this.maxAttempts) {
+ attempts++;
+ let currentStdout = '';
+ let currentStderr = '';
+
+ // --- Robust File Reading ---
+ try {
+ currentStdout = await fs.readFile(tempStdoutFilePath, 'utf-8');
+ } catch (error: unknown) {
+ // If file doesn't exist yet or isn't readable, treat as empty, but log warning
+ if (!isNodeError(error) || error.code !== 'ENOENT') {
+ console.warn(
+ `Attempt ${attempts}: Failed to read stdout file ${tempStdoutFilePath}: ${getErrorMessage(error)}`,
+ );
+ }
+ }
+ try {
+ currentStderr = await fs.readFile(tempStderrFilePath, 'utf-8');
+ } catch (error: unknown) {
+ if (!isNodeError(error) || error.code !== 'ENOENT') {
+ console.warn(
+ `Attempt ${attempts}: Failed to read stderr file ${tempStderrFilePath}: ${getErrorMessage(error)}`,
+ );
+ }
+ }
+
+ // --- Process Status Check ---
+ let isRunning = false;
+ try {
+ // Check if process is running *before* the final analysis if it seems to have ended
+ isRunning = await this.isProcessRunning(pid);
+ if (!isRunning) {
+ // Reread files one last time in case output was written just before exit
+ try {
+ currentStdout = await fs.readFile(tempStdoutFilePath, 'utf-8');
+ } catch {
+ /* ignore */
+ }
+ try {
+ currentStderr = await fs.readFile(tempStderrFilePath, 'utf-8');
+ } catch {
+ /* ignore */
+ }
+
+ lastAnalysisResult = await this.performLlmAnalysis(
+ currentStdout,
+ currentStderr,
+ command,
+ pid,
+ );
+
+ if (isAnalysisFailure(lastAnalysisResult)) {
+ return {
+ status: 'Completed_AnalysisFailed',
+ summary: `Process ended. Final analysis failed: ${lastAnalysisResult.error}`,
+ };
+ }
+ // Append ProcessEnded to the status determined by the final analysis
+ return {
+ status: 'Completed_' + lastAnalysisResult.inferredStatus,
+ summary: `Process ended. Final analysis summary: ${lastAnalysisResult.summary}`,
+ };
+ }
+ } catch (procCheckError: unknown) {
+ // Log the error but allow polling to continue, as log analysis might still be useful
+ console.warn(
+ `Could not check process status for PID ${pid} on attempt ${attempts}: ${getErrorMessage(procCheckError)}`,
+ );
+ // Decide if you want to bail out here or continue analysis based on logs only
+ // For now, we continue.
+ }
+
+ // --- LLM Analysis ---
+ lastAnalysisResult = await this.performLlmAnalysis(
+ currentStdout,
+ currentStderr,
+ command,
+ pid,
+ );
+
+ if (isAnalysisFailure(lastAnalysisResult)) {
+ console.error(
+ `LLM Analysis failed for PID ${pid} on attempt ${attempts}:`,
+ lastAnalysisResult.error,
+ );
+ // Stop polling on analysis failure, returning the specific failure status
+ return {
+ status: lastAnalysisResult.inferredStatus,
+ summary: lastAnalysisResult.error,
+ };
+ }
+
+ // --- Exit Conditions ---
+ if (
+ lastAnalysisResult.inferredStatus === 'SuccessReported' ||
+ lastAnalysisResult.inferredStatus === 'ErrorReported'
+ ) {
+ return {
+ status: lastAnalysisResult.inferredStatus,
+ summary: lastAnalysisResult.summary,
+ };
+ }
+
+ // Heuristic: If the process seems stable and 'Running' after several checks,
+ // return that status without waiting for the full timeout. Adjust threshold as needed.
+ const runningExitThreshold = Math.floor(this.maxAttempts / 3) + 1; // e.g., exit after attempt 4 if maxAttempts is 6
+ if (
+ attempts >= runningExitThreshold &&
+ lastAnalysisResult.inferredStatus === 'Running'
+ ) {
+ return {
+ status: lastAnalysisResult.inferredStatus,
+ summary: lastAnalysisResult.summary,
+ };
+ }
+
+ // --- Wait before next poll ---
+ if (attempts < this.maxAttempts) {
+ await new Promise((resolve) =>
+ setTimeout(resolve, this.pollIntervalMs),
+ );
+ }
+ } // End while loop
+
+ // --- Timeout Condition ---
+ console.warn(
+ `Polling timed out for PID ${pid} after ${this.maxAttempts} attempts.`,
+ );
+
+ // Determine final status based on the last successful analysis (if any)
+ const finalStatus =
+ lastAnalysisResult && !isAnalysisFailure(lastAnalysisResult)
+ ? `TimedOut_${lastAnalysisResult.inferredStatus}` // e.g., TimedOut_Running
+ : 'TimedOut_AnalysisFailed'; // If last attempt failed or no analysis succeeded
+
+ const finalSummary =
+ lastAnalysisResult && !isAnalysisFailure(lastAnalysisResult)
+ ? `Polling timed out after ${this.maxAttempts} attempts. Last known summary: ${lastAnalysisResult.summary}`
+ : lastAnalysisResult && isAnalysisFailure(lastAnalysisResult)
+ ? `Polling timed out; last analysis attempt failed: ${lastAnalysisResult}`
+ : `Polling timed out after ${this.maxAttempts} attempts without any successful analysis.`;
+
+ return { status: finalStatus, summary: finalSummary };
+ }
+
+ // --- Actual Implementation of isProcessRunning ---
+ /**
+ * Checks if the background process is still running using OS-specific methods.
+ * @param pid Process handle/identifier (expects a number for standard checks).
+ * @returns True if running, false otherwise.
+ * @throws Error if the check itself fails critically (e.g., command not found, permissions).
+ */
+ private async isProcessRunning(pid: ProcessHandle): Promise<boolean> {
+ if (typeof pid !== 'number' || !Number.isInteger(pid) || pid <= 0) {
+ console.warn(
+ `isProcessRunning: Invalid PID provided (${pid}). Assuming not running.`,
+ );
+ return false;
+ }
+
+ try {
+ if (process.platform === 'win32') {
+ // Windows: Use tasklist command
+ const command = `tasklist /FI "PID eq ${pid}" /NH`; // /NH for no header
+ const { stdout } = await execAsync(command);
+ // Check if the output contains the process information (it will have the image name if found)
+ return stdout.toLowerCase().includes('.exe'); // A simple check, adjust if needed
+ } else {
+ // Linux/macOS/Unix-like: Use kill -0 signal
+ // process.kill sends signal 0 to check existence without killing
+ process.kill(pid, 0);
+ return true; // If no error is thrown, process exists
+ }
+ } catch (error: unknown) {
+ if (isNodeError(error) && error.code === 'ESRCH') {
+ // ESRCH: Standard error code for "No such process" on Unix-like systems
+ return false;
+ } else if (
+ process.platform === 'win32' &&
+ getErrorMessage(error).includes('No tasks are running')
+ ) {
+ // tasklist specific error when PID doesn't exist
+ return false;
+ } else {
+ // Other errors (e.g., EPERM - permission denied) mean we couldn't determine status.
+ // Re-throwing might be appropriate depending on desired behavior.
+ // Here, we log it and cautiously return true, assuming it *might* still be running.
+ console.warn(
+ `isProcessRunning(${pid}) encountered error: ${getErrorMessage(error)}. Assuming process might still exist.`,
+ );
+ // Or you could throw the error: throw new Error(`Failed to check process status for PID ${pid}: ${error.message}`);
+ return true; // Cautious assumption
+ }
+ }
+ }
+
+ // --- LLM Analysis Method (largely unchanged but added validation robustness) ---
+ private async performLlmAnalysis(
+ stdoutContent: string,
+ stderrContent: string,
+ command: string,
+ pid: number,
+ ): Promise<AnalysisResult | AnalysisFailure> {
+ if (!this.geminiClient) {
+ return {
+ error: '[Analysis unavailable: Gemini client not initialized]',
+ inferredStatus: 'AnalysisFailed',
+ };
+ }
+
+ const truncatedStdout =
+ stdoutContent.substring(0, this.maxOutputAnalysisLength) +
+ (stdoutContent.length > this.maxOutputAnalysisLength
+ ? '... [truncated]'
+ : '');
+ const truncatedStderr =
+ stderrContent.substring(0, this.maxOutputAnalysisLength) +
+ (stderrContent.length > this.maxOutputAnalysisLength
+ ? '... [truncated]'
+ : '');
+
+ const analysisPrompt = `**Analyze Background Process Logs**
+
+**Context:** A command (\`${command}\`) was executed in the background. You are analyzing the standard output (stdout) and standard error (stderr) collected so far to understand its progress and outcome. This analysis will be used to inform a user about what the command did.
+
+**Input:**
+* **Command:** \`${command}\`
+* **Stdout:**
+ \`\`\`
+ ${truncatedStdout}
+ \`\`\`
+* **Stderr:**
+ \`\`\`
+ ${truncatedStderr}
+ \`\`\`
+
+**Task:**
+
+Based *only* on the provided stdout and stderr:
+
+1. **Interpret and Summarize:** Do *not* simply repeat the logs. Analyze the content and provide a concise summary describing the significant actions, results, progress, or errors reported by the command. If logs are empty, state that no output was captured. Summaries should be formatted as markdown. Focus on the most recent or conclusive information if logs are long.
+2. **Infer Current Status:** Based *only* on the log content, infer the likely status of the command's execution as reflected *in the logs*. Choose the most appropriate status from the options defined in the schema (\`Running\`, \`SuccessReported\`, \`ErrorReported\`, \`Unknown\`). For example:
+ * If logs show ongoing activity or progress messages without clear completion or error signals, use \`Running\`.
+ * If logs contain explicit messages indicating successful completion or the final expected output of a successful run, use \`SuccessReported\`.
+ * If logs contain error messages, stack traces, or failure indications, use \`ErrorReported\`.
+ * If the logs provide insufficient information to determine a clear status (e.g., empty logs, vague messages), use \`Unknown\`.
+ * If dealing with a node server, the second the port has been shown the server is considered booted, use \`SuccessReported\`.
+ * *Note: This status reflects the log content, not necessarily the absolute real-time state of the OS process.*
+3. **Format Output:** Return the results as a JSON object adhering strictly to the following schema:
+
+ \`\`\`json
+ ${JSON.stringify(
+ {
+ // Generate the schema JSON string for the prompt context
+ type: 'object',
+ properties: {
+ summary: {
+ type: 'string',
+ description:
+ 'Concise markdown summary (1-3 sentences) of log interpretation.',
+ },
+ inferredStatus: {
+ type: 'string',
+ enum: ['Running', 'SuccessReported', 'ErrorReported', 'Unknown'],
+ description:
+ 'Status inferred from logs: Running, SuccessReported, ErrorReported, Unknown',
+ },
+ },
+ required: ['summary', 'inferredStatus'],
+ },
+ null,
+ 2,
+ )}
+ \`\`\`
+
+**Instructions:**
+* The \`summary\` must be an interpretation of the logs, focusing on key outcomes or activities. Prioritize recent events if logs are extensive.
+* The \`inferredStatus\` should reflect the most likely state *deduced purely from the log text provided*. Ensure it is one of the specified enum values.`;
+
+ const schema: SchemaUnion = {
+ type: Type.OBJECT,
+ properties: {
+ summary: {
+ type: Type.STRING,
+ description:
+ 'Concise markdown summary (1-3 sentences) of log interpretation.',
+ },
+ inferredStatus: {
+ type: Type.STRING,
+ description:
+ 'Status inferred from logs: Running, SuccessReported, ErrorReported, Unknown',
+ enum: ['Running', 'SuccessReported', 'ErrorReported', 'Unknown'],
+ },
+ },
+ required: ['summary', 'inferredStatus'],
+ };
+
+ try {
+ const resultJson = await this.geminiClient.generateJson(
+ [{ role: 'user', parts: [{ text: analysisPrompt }] }],
+ schema,
+ );
+
+ // Validate and construct the AnalysisResult object
+ const summary =
+ typeof resultJson?.summary === 'string'
+ ? resultJson.summary
+ : '[Summary unavailable]';
+
+ // Define valid statuses using the AnalysisStatus type (ensure it's defined above)
+ const validStatuses: Array<Exclude<AnalysisStatus, 'AnalysisFailed'>> = [
+ 'Running',
+ 'SuccessReported',
+ 'ErrorReported',
+ 'Unknown',
+ ];
+
+ // Cast the unknown value to string before checking with includes
+ const statusString = resultJson?.inferredStatus as string;
+ const inferredStatus = validStatuses.includes(
+ statusString as Exclude<AnalysisStatus, 'AnalysisFailed'>,
+ )
+ ? (statusString as Exclude<AnalysisStatus, 'AnalysisFailed'>)
+ : 'Unknown';
+
+ // Explicitly construct the object matching AnalysisResult type
+ const analysisResult: AnalysisResult = { summary, inferredStatus };
+ return analysisResult;
+ } catch (error: unknown) {
+ console.error(`LLM Analysis Request Failed for PID ${pid}:`, error);
+ // Return the AnalysisFailure type
+ const analysisFailure: AnalysisFailure = {
+ error: `[Analysis failed: ${getErrorMessage(error)}]`,
+ inferredStatus: 'AnalysisFailed', // This matches the AnalysisStatus type
+ };
+ return analysisFailure;
+ }
+ }
+}
generated by cgit v1.2.3 (git 2.51.0) at 2025-11-05 18:32:01 +0000