fix: update system prompt, allow user to switch between chat mode and other modes (chat mode will have minimal amount of context and will help save money)

Author: oleg kusov

src/core/Cline.ts

@@ -405,10 +405,16 @@ export class Cline extends EventEmitter<ClineEvents> {
 			if (!provider) {
 				throw new Error("[Cline.ts ask] Cannot wait for input, provider reference lost.")
 			}
-			await provider.postMessageToWebview({ type: "enableChatInput" }) // <<< Use the new dedicated message type
+			await provider.postMessageToWebview({ type: "enableChatInput" })
 
-			// Wait for handleWebviewAskResponse to set this.askResponse
-			await pWaitFor(() => this.askResponse !== undefined, { interval: 100 })
+			// abort is crusial here to prevent blocking user input on switch mode
+			await pWaitFor(() => this.askResponse !== undefined || this.abort, { interval: 100 })
+
+			if (this.abort) {
+				throw new Error(
+					`[Cline#ask chat_input_wait] Task ${this.taskId}.${this.instanceId} aborted while waiting for user input.`,
+				)
+			}
 
 			console.log("[Cline.ts ask] Resuming after chat input received.")
 			const result = { response: this.askResponse!, text: this.askResponseText, images: this.askResponseImages }
@@ -1064,18 +1070,16 @@ export class Cline extends EventEmitter<ClineEvents> {
 		console.log(`Model: ${this.api.getModel().id}, Context window: ${modelInfo.contextWindow} tokens`)
 
 		// Detailed debug info on each message
-		if (mode === "chat") {
-			console.log(`\n======= MINIMAL MODE DEBUG INFO =======`)
-			console.log(`System prompt: ${systemPrompt}...`)
-			console.log(
-				`\nLast user message content: ${
-					this.apiConversationHistory.length > 0 &&
-					this.apiConversationHistory[this.apiConversationHistory.length - 1].role === "user"
-						? JSON.stringify(this.apiConversationHistory[this.apiConversationHistory.length - 1].content)
-						: "No user message"
-				}...`,
-			)
-		}
+		console.log(`\n======= MINIMAL MODE DEBUG INFO =======`)
+		console.log(`System prompt: ${systemPrompt}...`)
+		console.log(
+			`\nLast user message content: ${
+				this.apiConversationHistory.length > 0 &&
+				this.apiConversationHistory[this.apiConversationHistory.length - 1].role === "user"
+					? JSON.stringify(this.apiConversationHistory[this.apiConversationHistory.length - 1].content)
+					: "No user message"
+			}...`,
+		)
 
 		// If the previous API request's total token usage is close to the context window, truncate the conversation history to free up space for the new request
 		if (previousApiReqIndex >= 0) {
@@ -1955,57 +1959,34 @@ export class Cline extends EventEmitter<ClineEvents> {
 
 				await pWaitFor(() => this.userMessageContentReady)
 
-				// if the model did not tool use, then we need to tell it to either use a tool or attempt_completion
-				const didToolUse = this.assistantMessageContent.some((block) => block.type === "tool_use")
-				if (!didToolUse) {
-					// --- Get provider and mode ---
-					const provider = this.providerRef.deref()
-					const { mode } = (await provider?.getState()) ?? {}
+				// --- Get provider and mode ---
+				const provider = this.providerRef.deref()
+				const { mode } = (await provider?.getState()) ?? {}
 
-					if (mode !== "chat") {
-						// Non-chat mode, no tool use: Add feedback and recurse internally
+				if (mode === "chat") {
+					// Handle chat mode continuation/termination
+					const nextUserContent = await this._handleChatResponse()
+					if (nextUserContent) {
+						return await this.recursivelyMakeClineRequests(nextUserContent, false)
+					} else {
+						didEndLoop = true
+					}
+				} else {
+					// Original logic for non-chat modes
+					const didToolUse = this.assistantMessageContent.some((block) => block.type === "tool_use")
+					if (!didToolUse) {
+						// if the model did not tool use, then we need to tell it to either use a tool or attempt_completion
 						this.userMessageContent.push({
 							type: "text",
 							text: formatResponse.noToolsUsed(),
 						})
 						this.consecutiveMistakeCount++
-						// Continue the loop by recursing with the feedback message
-						// Return the boolean result of this recursive call
-						return await this.recursivelyMakeClineRequests(this.userMessageContent, false)
-					} else {
-						// Chat mode, no tool use: Use ask('chat_input_wait') to pause and get input
-						console.log("[Cline.ts] Chat mode, no tool use. Waiting for next user input via ask.")
-						const { response, text, images } = await this.ask("chat_input_wait")
-
-						if (response === "messageResponse") {
-							// User provided input
-							const nextUserContent: UserContent = []
-							if (text) nextUserContent.push({ type: "text", text })
-							if (images) nextUserContent.push(...formatResponse.imageBlocks(images))
-
-							if (nextUserContent.length > 0) {
-								// Add user input to UI (ask doesn't add messages for chat_input_wait)
-								await this.say("user_feedback", text, images)
-								// Recurse with the new user input
-								return await this.recursivelyMakeClineRequests(nextUserContent, false)
-							} else {
-								// Resumed but no content? End the loop.
-								console.warn("[Cline.ts] Resumed chat wait but received no content.")
-								return true
-							}
-						} else {
-							// User likely cancelled or something went wrong during the ask wait. End the loop.
-							console.log("[Cline.ts] Chat input wait did not receive messageResponse.")
-							return true
-						}
 					}
-				} else {
-					// Tool use detected: tool results are already prepared in this.userMessageContent
-					// by presentAssistantMessage. Continue the loop by recursing.
-					return await this.recursivelyMakeClineRequests(this.userMessageContent, false)
+					const recDidEndLoop = await this.recursivelyMakeClineRequests(this.userMessageContent)
+					didEndLoop = recDidEndLoop
 				}
 			} else {
-				// if there's no assistant_responses, that means we got no text or tool_use content blocks from API which we should assume is an error
+				// if there's no assistant_responses, that means we got no text or tool_use content blocks from API
 				await this.say(
 					"error",
 					"Unexpected API Response: The language model did not provide any assistant messages. This may indicate an issue with the API or the model's output.",
@@ -2014,9 +1995,10 @@ export class Cline extends EventEmitter<ClineEvents> {
 					role: "assistant",
 					content: [{ type: "text", text: "Failure: I did not provide a response." }],
 				})
+				didEndLoop = true
 			}
 
-			return didEndLoop // will always be false for now
+			return didEndLoop
 		} catch (error) {
 			// This should never happen since the only thing that can throw an
 			// error is the attemptApiRequest, which is wrapped in a try catch
@@ -2673,4 +2655,33 @@ export class Cline extends EventEmitter<ClineEvents> {
 	public getToolUsage() {
 		return this.toolUsage
 	}
+
+	// --- Add the new private method here ---
+	private async _handleChatResponse(): Promise<UserContent | null> {
+		// Chat mode: Always wait for user input after assistant's turn.
+		console.log("[Cline.ts] Chat mode. Waiting for next user input via ask.")
+		const { response, text, images } = await this.ask("chat_input_wait")
+
+		if (response === "messageResponse") {
+			// User provided input
+			const nextUserContent: UserContent = []
+			if (text) nextUserContent.push({ type: "text", text })
+			if (images) nextUserContent.push(...formatResponse.imageBlocks(images))
+
+			if (nextUserContent.length > 0) {
+				// Add user input to UI (ask doesn't add messages for chat_input_wait)
+				await this.say("user_feedback", text, images)
+				// Return content for next recursion
+				return nextUserContent
+			} else {
+				// Resumed but no content? End the loop.
+				console.warn("[Cline.ts] Resumed chat wait but received no content.")
+				return null // Signal loop termination
+			}
+		} else {
+			// User likely cancelled or something went wrong during the ask wait. End the loop.
+			console.log("[Cline.ts] Chat input wait did not receive messageResponse.")
+			return null // Signal loop termination
+		}
+	}
 }

src/core/prompts/system.ts

@@ -58,10 +58,23 @@ async function generatePrompt(
 		const globalCustomInstructionsText = globalCustomInstructions || ""
 
 		// Extremely minimal prompt for code-minimal mode
-		return `${roleDefinition}
+		return `
+=======
+
+INSTRUCTIONS BLOCK START
+
+FORGET ALL PREVIOUS INSTRUCTIONS. THESE ARE THE CURRENT INSTRUCTIONS BLOCK
+
+${roleDefinition}
 
 ${customInstructionsText}
-${globalCustomInstructionsText}`
+
+${globalCustomInstructionsText}
+
+INSTRUCTIONS BLOCK END
+
+=======
+`
 	}
 
 	// If diff is disabled, don't pass the diffStrategy
@@ -78,7 +91,14 @@ ${globalCustomInstructionsText}`
 			: Promise.resolve(""),
 	])
 
-	const basePrompt = `${roleDefinition}
+	const basePrompt = `
+=======
+
+INSTRUCTIONS BLOCK START
+
+FORGET ALL PREVIOUS INSTRUCTIONS. THESE ARE THE CURRENT INSTRUCTIONS BLOCK
+	
+${roleDefinition}
 
 ${getSharedToolUseSection()}
 
@@ -107,7 +127,13 @@ ${getSystemInfoSection(cwd)}
 
 ${getObjectiveSection()}
 
-${await addCustomInstructions(promptComponent?.customInstructions || modeConfig.customInstructions || "", globalCustomInstructions || "", cwd, mode, { language: language ?? formatLanguage(vscode.env.language), rooIgnoreInstructions })}`
+${await addCustomInstructions(promptComponent?.customInstructions || modeConfig.customInstructions || "", globalCustomInstructions || "", cwd, mode, { language: language ?? formatLanguage(vscode.env.language), rooIgnoreInstructions })}
+
+INSTRUCTIONS BLOCK END
+
+=======
+
+`
 
 	return basePrompt
 }

src/core/webview/ClineProvider.ts

@@ -787,6 +787,7 @@ export class ClineProvider extends EventEmitter<ClineProviderEvents> implements
 		const cline = this.getCurrentCline()
 
 		if (cline) {
+			// await cline.abortTask(true);
 			telemetryService.captureModeSwitch(cline.taskId, newMode)
 			cline.emit("taskModeSwitched", cline.taskId, newMode)
 		}

src/shared/modes.ts

@@ -61,12 +61,12 @@ export const modes: readonly ModeConfig[] = [
 	},
 	{
 		slug: "chat",
-		name: "💬 Chat (minimal token usage)",
+		name: "💬 Chat",
 		roleDefinition:
-			"You are Roo, a simple chat assistant that only processes files manually attached with '@context' by the user.",
-		groups: ["read", "edit"],
+			"You are Roo, a highly skilled software engineer with extensive knowledge in many programming languages, frameworks, design patterns, and best practices.",
+		groups: [],
 		customInstructions:
-			"This is a direct chat mode. You will ONLY have access to files explicitly attached by the user using '@context'. No automatic context, no auto-loaded files, no workspace awareness. Treat this as a clean environment where only explicitly shared files exist.",
+			"You are currently in Chat mode. Now your answer should be just regular markdown (and code blocks if needed). Don't use special tag symbols (<implementation />, <key_improvements> and etc). It's important! Tool use is disabled in this mode and its very important that you don't use any tools. If you do, you will be penalized. Please, do not use any tools. Focus on providing informative answers, explanations, or engaging in discussion based on the user's messages. Now, lets continue with the conversation.",
 	},
 	{
 		slug: "architect",

webview-ui/src/components/chat/ChatView.tsx

@@ -371,51 +371,24 @@ const ChatViewComponent: React.ForwardRefRenderFunction<ChatViewRef, ChatViewPro
 
 	const handleSendMessage = useCallback(
 		(text: string, images: string[]) => {
+			console.log(
+				`[ChatView.tsx handleSendMessage] Fired. text: "${text}", images: ${images.length}, clineAsk: ${clineAsk}, mode: ${mode}`,
+			)
 			text = text.trim()
-			let messageSent = false // Flag to track if we actually sent a message
 
 			if (text || images.length > 0) {
 				if (messages.length === 0) {
-					// First message starts a new task
+					console.log("[ChatView.tsx handleSendMessage] Condition: First message (newTask).")
 					vscode.postMessage({ type: "newTask", text, images })
-					messageSent = true
-				} else if (clineAsk) {
-					// Handle explicit asks from the backend
-					switch (clineAsk) {
-						case "followup":
-						case "tool":
-						case "browser_action_launch":
-						case "command": // User can provide feedback to a tool or command use.
-						case "command_output": // User can send input to command stdin.
-						case "use_mcp_server":
-						case "completion_result": // If this happens then the user has feedback for the completion result.
-						case "resume_task":
-						case "resume_completed_task":
-						case "mistake_limit_reached":
-							vscode.postMessage({ type: "askResponse", askResponse: "messageResponse", text, images })
-							messageSent = true
-							break
-						// Ensure all relevant ask types that accept messageResponse are included
-					}
-				} else if (mode === "chat") {
-					// Check if specifically in chat mode
-					// If not the first message, no explicit ask, BUT in chat mode: send continuation.
-					vscode.postMessage({ type: "askResponse", askResponse: "messageResponse", text, images })
-					messageSent = true
-				}
-
-				// Only reset the input if we actually sent a message
-				if (messageSent) {
-					handleChatReset()
 				} else {
-					console.warn(
-						"[ChatView.tsx] handleSendMessage called but no condition met to send message. Input not reset.",
-					)
+					vscode.postMessage({ type: "askResponse", askResponse: "messageResponse", text, images })
 				}
+				handleChatReset()
+			} else {
+				console.log("[ChatView.tsx handleSendMessage] No text or images provided. No message sent.")
 			}
 		},
-		// Add 'mode' to dependencies
-		[messages.length, clineAsk, handleChatReset, mode],
+		[messages.length, clineAsk, handleChatReset, mode], // Keep mode dependency for logging
 	)
 
 	const handleSetChatBoxMessage = useCallback(