Add support for interactive command line execution where user input is sent to stdin

src/ClaudeDev.ts

@@ -4,19 +4,19 @@ import * as diff from "diff"
 import { execa } from "execa"
 import fs from "fs/promises"
 import { globby } from "globby"
+import os from "os"
 import osName from "os-name"
+import pWaitFor from "p-wait-for"
 import * as path from "path"
 import { serializeError } from "serialize-error"
-import { DEFAULT_MAX_REQUESTS_PER_TASK } from "./shared/Constants"
-import { Tool, ToolName } from "./shared/Tool"
-import { ClaudeAsk, ClaudeSay, ClaudeSayTool } from "./shared/ExtensionMessage"
 import * as vscode from "vscode"
-import pWaitFor from "p-wait-for"
-import { ClaudeAskResponse } from "./shared/WebviewMessage"
+import { analyzeProject } from "./analyze-project"
 import { ClaudeDevProvider } from "./providers/ClaudeDevProvider"
 import { ClaudeRequestResult } from "./shared/ClaudeRequestResult"
-import os from "os"
-import { analyzeProject } from "./analyze-project"
+import { DEFAULT_MAX_REQUESTS_PER_TASK } from "./shared/Constants"
+import { ClaudeAsk, ClaudeSay, ClaudeSayTool } from "./shared/ExtensionMessage"
+import { Tool, ToolName } from "./shared/Tool"
+import { ClaudeAskResponse } from "./shared/WebviewMessage"
 
 const SYSTEM_PROMPT = `You are Claude Dev, a highly skilled software developer with extensive knowledge in many programming languages, frameworks, design patterns, and best practices.
 
@@ -28,7 +28,7 @@ CAPABILITIES
 - You can debug complex issues and providing detailed explanations, offering architectural insights and design patterns.
 - You have access to tools that let you analyze software projects, execute CLI commands on the user's computer, list files in a directory, read and write files, and ask follow-up questions. These tools help you effectively accomplish a wide range of tasks, such as writing code, making edits or improvements to existing files, understanding the current state of a project, performing system operations, and much more.
     - For example, when asked to make edits or improvements you might use the analyze_project and read_file tools to examine the contents of relevant files, analyze the code and suggest improvements or make necessary edits, then use the write_to_file tool to implement changes.
-- You can use the analyze_project tool to get a comprehensive view of a software project's file structure and important syntactic nodes such as functions, classes, and methods. This can be particularly useful when you need to understand the broader context and relationships between different parts of the code, as well as the overall organization of files and directories.
+- You can use the analyze_project tool to get a comprehensive overview of a software project's file structure and source code definitions. This can be particularly useful when you need to understand the broader context and relationships between different parts of the code, as well as the overall organization of files and directories.
 - The execute_command tool lets you run commands on the user's computer and should be used whenever you feel it can help accomplish the user's task. When you need to execute a CLI command, you must provide a clear explanation of what the command does. Prefer to execute complex CLI commands over creating executable scripts, since they are more flexible and easier to run.
 
 ====
@@ -39,10 +39,9 @@ RULES
 	vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop")
 }
 - Your current working directory is '${process.cwd()}', and you cannot \`cd\` into a different directory to complete a task. You are stuck operating from '${process.cwd()}', so be sure to pass in the appropriate 'path' parameter when using tools that require a path.
-- If you do not know the contents of an existing file you need to edit, use the read_file tool to help you make informed changes. However if you have seen this file before, you should be able to remember its contents.
+- If you do not know the contents of an existing file you need to edit, use the read_file tool to help you make informed changes. However if you have seen this file before, you can assume its contents have not changed since you last read it or wrote to it.
 - When editing files, always provide the complete file content in your response, regardless of the extent of changes. The system handles diff generation automatically.
-- Before using the execute_command tool, you must first think about the System Information context provided by the user to understand their environment and tailor your commands to ensure they are compatible with the user's system.
-- When using the execute_command tool, avoid running servers or executing commands that don't terminate on their own (e.g. Flask web servers, continuous scripts). If a task requires such a process or server, explain in your task completion result why you can't execute it directly and provide clear instructions on how the user can run it themselves.
+- Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system.
 - Try not to use the analyze_project tool more than once since you can refer back to it along with any changes you made to get an adequate understanding of the project. But don't be hesitant to use it in the first place when you know you will be doing a coding task on an existing project. Prefer to use analyze_project over list_files, unless you think list_files is more appropriate for the job i.e. when viewing files on the Desktop.
 - When creating a new project (such as an app, website, or any software project), unless the user specifies otherwise, organize all new files within a dedicated project directory. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser.
 - You must try to use multiple tools in one request when possible. For example if you were to create a website, you would use the write_to_file tool to create the necessary files with their appropriate contents all at once. Or if you wanted to analyze a project, you could use the read_file tool multiple times to look at several key files. This will help you accomplish the user's task more efficiently.
@@ -64,7 +63,7 @@ You accomplish a given task iteratively, breaking it down into clear steps and w
 1. Analyze the user's task and set clear, achievable goals to accomplish it. Prioritize these goals in a logical order.
 2. Work through these goals sequentially, utilizing available tools as necessary. Each goal should correspond to a distinct step in your problem-solving process.
 3. Remember, you have extensive capabilities with access to a wide range of tools that can be used in powerful and clever ways as necessary to accomplish each goal. Before calling a tool, do some analysis within <thinking></thinking> tags. First, think about which of the provided tools is the relevant tool to answer the user's request. Second, go through each of the required parameters of the relevant tool and determine if the user has directly provided or given enough information to infer a value. When deciding if the parameter can be inferred, carefully consider all the context to see if it supports a specific value. If all of the required parameters are present or can be reasonably inferred, close the thinking tag and proceed with the tool call. BUT, if one of the values for a required parameter is missing, DO NOT invoke the function (not even with fillers for the missing params) and instead, ask the user to provide the missing parameters using the ask_followup_question tool. DO NOT ask for more information on optional parameters if it is not provided.
-4. Once you've completed the user's task, you must use the attempt_completion tool to present the result of the task to the user. You may also provide a CLI command to showcase the result of your task; this can be particularly useful for web development tasks, where you can run e.g. \`open index.html\` to show the website you've built. Avoid commands that run indefinitely (like servers). Instead, if such a command is needed, include instructions for the user to run it in the 'result' parameter.
+4. Once you've completed the user's task, you must use the attempt_completion tool to present the result of the task to the user. You may also provide a CLI command to showcase the result of your task; this can be particularly useful for web development tasks, where you can run e.g. \`open index.html\` to show the website you've built.
 5. The user may provide feedback, which you can use to make improvements and try again. But DO NOT continue in pointless back and forth conversations, i.e. don't end your responses with questions or offers for further assistance.
 
 ====
@@ -92,14 +91,14 @@ const tools: Tool[] = [
 	{
 		name: "execute_command",
 		description:
-			"Execute a CLI command on the system. Use this when you need to perform system operations or run specific commands to accomplish any step in the user's task. You must tailor your command to the user's system and provide a clear explanation of what the command does. Do not run servers or commands that don't terminate on their own. Prefer to execute complex CLI commands over creating executable scripts, as they are more flexible and easier to run.",
+			"Execute a CLI command on the system. Use this when you need to perform system operations or run specific commands to accomplish any step in the user's task. You must tailor your command to the user's system and provide a clear explanation of what the command does. Prefer to execute complex CLI commands over creating executable scripts, as they are more flexible and easier to run.",
 		input_schema: {
 			type: "object",
 			properties: {
 				command: {
 					type: "string",
 					description:
-						"The CLI command to execute. This should be valid for the current operating system. Ensure the command is properly formatted and does not contain any harmful instructions. Avoid commands that run indefinitely (like servers) that don't terminate on their own.",
+						"The CLI command to execute. This should be valid for the current operating system. Ensure the command is properly formatted and does not contain any harmful instructions.",
 				},
 			},
 			required: ["command"],
@@ -215,6 +214,7 @@ export class ClaudeDev {
 	private requestCount = 0
 	private askResponse?: ClaudeAskResponse
 	private askResponseText?: string
+	private lastMessageTs?: number
 	private providerRef: WeakRef<ClaudeDevProvider>
 	abort: boolean = false
 
@@ -246,9 +246,14 @@ export class ClaudeDev {
 		}
 		this.askResponse = undefined
 		this.askResponseText = undefined
-		await this.providerRef.deref()?.addClaudeMessage({ ts: Date.now(), type: "ask", ask: type, text: question })
+		const askTs = Date.now()
+		this.lastMessageTs = askTs
+		await this.providerRef.deref()?.addClaudeMessage({ ts: askTs, type: "ask", ask: type, text: question })
 		await this.providerRef.deref()?.postStateToWebview()
-		await pWaitFor(() => this.askResponse !== undefined, { interval: 100 })
+		await pWaitFor(() => this.askResponse !== undefined || this.lastMessageTs !== askTs, { interval: 100 })
+		if (this.lastMessageTs !== askTs) {
+			throw new Error("Current ask promise was ignored") // could happen if we send multiple asks in a row i.e. with command_output. It's important that when we know an ask could fail, it is handled gracefully
+		}
 		const result = { response: this.askResponse!, text: this.askResponseText }
 		this.askResponse = undefined
 		this.askResponseText = undefined
@@ -259,7 +264,9 @@ export class ClaudeDev {
 		if (this.abort) {
 			throw new Error("ClaudeDev instance aborted")
 		}
-		await this.providerRef.deref()?.addClaudeMessage({ ts: Date.now(), type: "say", say: type, text: text })
+		const sayTs = Date.now()
+		this.lastMessageTs = sayTs
+		await this.providerRef.deref()?.addClaudeMessage({ ts: sayTs, type: "say", say: type, text: text })
 		await this.providerRef.deref()?.postStateToWebview()
 	}
 
@@ -511,11 +518,28 @@ export class ClaudeDev {
 		}
 		try {
 			let result = ""
-			// execa by default tries to convery bash into javascript
-			// by using shell: true we use sh on unix or cmd.exe on windows
-			// also worth noting that execa`input` runs commands and the execa() creates a new instance
-			for await (const line of execa({ shell: true })`${command}`) {
-				this.say("command_output", line) // stream output to user in realtime
+			// execa by default tries to convert bash into javascript, so need to specify `shell: true` to use sh on unix or cmd.exe on windows
+			// also worth noting that execa`input` and the execa(command) have nuanced differences like the template literal version handles escaping for you, while with the function call, you need to be more careful about how arguments are passed, especially when using shell: true.
+			// execa returns a promise-like object that is both a promise and a Subprocess that has properties like stdin
+			const subprocess = execa({ shell: true })`${command}`
+
+			for await (const chunk of subprocess) {
+				const line = chunk.toString()
+				// stream output to user in realtime
+				this.ask("command_output", line)
+					.then(({ response, text }) => {
+						// if this ask promise is not ignored, that means the user responded to it somehow either by clicking primary button or by typing text
+						if (response === "yesButtonTapped") {
+							subprocess.kill() // Will result in for loop throwing error, so claude will know command failed
+						} else {
+							// if the user sent some input, we send it to the command stdin
+							// add newline as cli programs expect a newline after each input
+							subprocess.stdin?.write(text + "\n")
+						}
+					})
+					.catch(() => {
+						// this can only happen if this ask promise was ignored, so ignore this error
+					})
 				result += `${line}\n`
 			}
 			// for attemptCompletion, we don't want to return the command output

src/shared/ExtensionMessage.ts

@@ -28,6 +28,7 @@ export type ClaudeAsk =
 	| "request_limit_reached"
 	| "followup"
 	| "command"
+	| "command_output"
 	| "completion_result"
 	| "tool"
 	| "api_req_failed"
@@ -38,7 +39,6 @@ export type ClaudeSay =
 	| "api_req_started"
 	| "api_req_finished"
 	| "text"
-	| "command_output"
 	| "completion_result"
 	| "user_feedback"
 	| "api_req_retried"

webview-ui/src/components/ChatView.tsx

@@ -108,6 +108,13 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 							setPrimaryButtonText("Run Command")
 							setSecondaryButtonText("Reject")
 							break
+						case "command_output":
+							setTextAreaDisabled(false)
+							setClaudeAsk("command_output")
+							setEnableButtons(true)
+							setPrimaryButtonText("Exit Command Early")
+							setSecondaryButtonText(undefined)
+							break
 						case "completion_result":
 							// extension waiting for feedback. but we can just present a new task button
 							setTextAreaDisabled(false)
@@ -131,8 +138,6 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 							break
 						case "text":
 							break
-						case "command_output":
-							break
 						case "completion_result":
 							break
 					}
@@ -168,6 +173,7 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 					case "followup":
 					case "tool":
 					case "command": // user can provide feedback to a tool or command use
+					case "command_output": // user can send input to command stdin
 					case "completion_result": // if this happens then the user has feedback for the completion result
 						vscode.postMessage({ type: "askResponse", askResponse: "textResponse", text })
 						break
@@ -191,6 +197,7 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 			case "request_limit_reached":
 			case "api_req_failed":
 			case "command":
+			case "command_output":
 			case "tool":
 				vscode.postMessage({ type: "askResponse", askResponse: "yesButtonTapped" })
 				break
@@ -315,6 +322,13 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 		return () => clearTimeout(timer)
 	}, [visibleMessages])
 
+	const placeholderText = useMemo(() => {
+		if (messages.at(-1)?.ask === "command_output") {
+			return "Type input for command stdin..."
+		}
+		return task ? "Type a message..." : "Type your task here..."
+	}, [task, messages])
+
 	return (
 		<div
 			style={{
@@ -425,7 +439,7 @@ const ChatView = ({ messages, isHidden, vscodeThemeName, showAnnouncement, hideA
 						//virtuosoRef.current?.scrollToIndex({ index: "LAST", align: "end", behavior: "auto" })
 						virtuosoRef.current?.scrollTo({ top: Number.MAX_SAFE_INTEGER, behavior: "auto" })
 					}
-					placeholder={task ? "Type a message..." : "Type your task here..."}
+					placeholder={placeholderText}
 					maxRows={10}
 					autoFocus={true}
 					style={{

webview-ui/src/utilities/combineCommandSequences.ts

@@ -14,8 +14,8 @@ import { ClaudeMessage } from "@shared/ExtensionMessage"
  * @example
  * const messages: ClaudeMessage[] = [
  *   { type: 'ask', ask: 'command', text: 'ls', ts: 1625097600000 },
- *   { type: 'say', say: 'command_output', text: 'file1.txt', ts: 1625097601000 },
- *   { type: 'say', say: 'command_output', text: 'file2.txt', ts: 1625097602000 }
+ *   { type: 'ask', ask: 'command_output', text: 'file1.txt', ts: 1625097601000 },
+ *   { type: 'ask', ask: 'command_output', text: 'file2.txt', ts: 1625097602000 }
  * ];
  * const result = simpleCombineCommandSequences(messages);
  * // Result: [{ type: 'ask', ask: 'command', text: 'ls\nfile1.txt\nfile2.txt', ts: 1625097600000 }]
@@ -35,7 +35,7 @@ export function combineCommandSequences(messages: ClaudeMessage[]): ClaudeMessag
 					// Stop if we encounter the next command
 					break
 				}
-				if (messages[j].type === "say" && messages[j].say === "command_output") {
+				if (messages[j].type === "ask" && messages[j].ask === "command_output") {
 					if (!didAddOutput) {
 						// Add a newline before the first output
 						combinedText += `\n${COMMAND_OUTPUT_STRING}`
@@ -57,7 +57,7 @@ export function combineCommandSequences(messages: ClaudeMessage[]): ClaudeMessag
 
 	// Second pass: remove command_outputs and replace original commands with combined ones
 	return messages
-		.filter((msg) => !(msg.type === "say" && msg.say === "command_output"))
+		.filter((msg) => !(msg.type === "ask" && msg.ask === "command_output"))
 		.map((msg) => {
 			if (msg.type === "ask" && msg.ask === "command") {
 				const combinedCommand = combinedCommands.find((cmd) => cmd.ts === msg.ts)