[Condense] Add a button to condense the task context (#3623)

* [Condense] Add a button to condense the task context

* wip

* wip

* wip

* bring back delete size

* account for the system prompt in the context

* update tests to use systemPrompt

* add type

* translations

* nit

* update tests

* filter to the current task

* nit

* refactor

* nit

* non interactive option

* simplify chat summary UI

* changeset

* nit

* fix check-types

* throw

Author: canrobins13

.changeset/tired-dogs-worry.md

@@ -0,0 +1,5 @@
+---
+"roo-cline": patch
+---
+
+Adds a button to intelligently condense the context window

src/core/condense/__tests__/index.test.ts

@@ -97,13 +97,16 @@ describe("summarizeConversation", () => {
 		} as unknown as ApiHandler
 	})
 
+	// Default system prompt for tests
+	const defaultSystemPrompt = "You are a helpful assistant."
+
 	it("should not summarize when there are not enough messages", async () => {
 		const messages: ApiMessage[] = [
 			{ role: "user", content: "Hello", ts: 1 },
 			{ role: "assistant", content: "Hi there", ts: 2 },
 		]
 
-		const result = await summarizeConversation(messages, mockApiHandler)
+		const result = await summarizeConversation(messages, mockApiHandler, defaultSystemPrompt)
 		expect(result.messages).toEqual(messages)
 		expect(result.cost).toBe(0)
 		expect(result.summary).toBe("")
@@ -122,7 +125,7 @@ describe("summarizeConversation", () => {
 			{ role: "user", content: "Tell me more", ts: 7 },
 		]
 
-		const result = await summarizeConversation(messages, mockApiHandler)
+		const result = await summarizeConversation(messages, mockApiHandler, defaultSystemPrompt)
 		expect(result.messages).toEqual(messages)
 		expect(result.cost).toBe(0)
 		expect(result.summary).toBe("")
@@ -141,7 +144,7 @@ describe("summarizeConversation", () => {
 			{ role: "user", content: "Tell me more", ts: 7 },
 		]
 
-		const result = await summarizeConversation(messages, mockApiHandler)
+		const result = await summarizeConversation(messages, mockApiHandler, defaultSystemPrompt)
 
 		// Check that the API was called correctly
 		expect(mockApiHandler.createMessage).toHaveBeenCalled()
@@ -199,7 +202,7 @@ describe("summarizeConversation", () => {
 			return messages.map(({ role, content }: { role: string; content: any }) => ({ role, content }))
 		})
 
-		const result = await summarizeConversation(messages, mockApiHandler)
+		const result = await summarizeConversation(messages, mockApiHandler, defaultSystemPrompt)
 
 		// Should return original messages when summary is empty
 		expect(result.messages).toEqual(messages)
@@ -222,7 +225,7 @@ describe("summarizeConversation", () => {
 			{ role: "user", content: "Tell me more", ts: 7 },
 		]
 
-		await summarizeConversation(messages, mockApiHandler)
+		await summarizeConversation(messages, mockApiHandler, defaultSystemPrompt)
 
 		// Verify the final request message
 		const expectedFinalMessage = {

src/core/condense/index.ts

@@ -57,12 +57,13 @@ export type SummarizeResponse = {
  *
  * @param {ApiMessage[]} messages - The conversation messages
  * @param {ApiHandler} apiHandler - The API handler to use for token counting.
+ * @param {string} systemPrompt - The system prompt for API requests, which should be considered in the context token count
  * @returns {SummarizeResponse} - The result of the summarization operation (see above)
  */
 export async function summarizeConversation(
 	messages: ApiMessage[],
 	apiHandler: ApiHandler,
-	systemPrompt?: string,
+	systemPrompt: string,
 ): Promise<SummarizeResponse> {
 	const response: SummarizeResponse = { messages, cost: 0, summary: "" }
 	const messagesToSummarize = getMessagesSinceLastSummary(messages.slice(0, -N_MESSAGES_TO_KEEP))
@@ -111,10 +112,10 @@ export async function summarizeConversation(
 
 	// Count the tokens in the context for the next API request
 	// We only estimate the tokens in summaryMesage if outputTokens is 0, otherwise we use outputTokens
-	const contextMessages = outputTokens ? [...keepMessages] : [summaryMessage, ...keepMessages]
-	if (systemPrompt) {
-		contextMessages.unshift({ role: "user", content: systemPrompt })
-	}
+	const systemPromptMessage: ApiMessage = { role: "user", content: systemPrompt }
+	const contextMessages = outputTokens
+		? [systemPromptMessage, ...keepMessages]
+		: [systemPromptMessage, summaryMessage, ...keepMessages]
 	const contextBlocks = contextMessages.flatMap((message) =>
 		typeof message.content === "string" ? [{ text: message.content, type: "text" as const }] : message.content,
 	)

src/core/sliding-window/__tests__/sliding-window.test.ts

@@ -248,6 +248,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		// Check the new return type
@@ -276,6 +277,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		expect(result).toEqual({
@@ -302,6 +304,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo1.contextWindow,
 			maxTokens: modelInfo1.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		const result2 = await truncateConversationIfNeeded({
@@ -310,6 +313,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo2.contextWindow,
 			maxTokens: modelInfo2.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		expect(result1.messages).toEqual(result2.messages)
@@ -325,6 +329,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo1.contextWindow,
 			maxTokens: modelInfo1.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		const result4 = await truncateConversationIfNeeded({
@@ -333,6 +338,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo2.contextWindow,
 			maxTokens: modelInfo2.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 
 		expect(result3.messages).toEqual(result4.messages)
@@ -363,6 +369,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(resultWithSmall).toEqual({
 			messages: messagesWithSmallContent,
@@ -392,6 +399,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(resultWithLarge.messages).not.toEqual(messagesWithLargeContent) // Should truncate
 		expect(resultWithLarge.summary).toBe("")
@@ -414,6 +422,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(resultWithVeryLarge.messages).not.toEqual(messagesWithVeryLargeContent) // Should truncate
 		expect(resultWithVeryLarge.summary).toBe("")
@@ -439,6 +448,7 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result).toEqual({
 			messages: expectedResult,
@@ -524,6 +534,7 @@ describe("truncateConversationIfNeeded", () => {
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
 			autoCondenseContext: true,
+			systemPrompt: "System prompt",
 		})
 
 		// Verify summarizeConversation was called
@@ -559,6 +570,7 @@ describe("truncateConversationIfNeeded", () => {
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
 			autoCondenseContext: false,
+			systemPrompt: "System prompt",
 		})
 
 		// Verify summarizeConversation was not called
@@ -612,6 +624,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result1).toEqual({
 			messages: messagesWithSmallContent,
@@ -627,6 +640,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result2.messages).not.toEqual(messagesWithSmallContent)
 		expect(result2.messages.length).toBe(3) // Truncated with 0.5 fraction
@@ -650,6 +664,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result1).toEqual({
 			messages: messagesWithSmallContent,
@@ -665,6 +680,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result2.messages).not.toEqual(messagesWithSmallContent)
 		expect(result2.messages.length).toBe(3) // Truncated with 0.5 fraction
@@ -687,6 +703,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result1.messages).toEqual(messagesWithSmallContent)
 
@@ -697,6 +714,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result2).not.toEqual(messagesWithSmallContent)
 		expect(result2.messages.length).toBe(3) // Truncated with 0.5 fraction
@@ -717,6 +735,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result1.messages).toEqual(messagesWithSmallContent)
 
@@ -727,6 +746,7 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			systemPrompt: "System prompt",
 		})
 		expect(result2).not.toEqual(messagesWithSmallContent)
 		expect(result2.messages.length).toBe(3) // Truncated with 0.5 fraction

src/core/sliding-window/index.ts

@@ -64,7 +64,7 @@ type TruncateOptions = {
 	maxTokens?: number | null
 	apiHandler: ApiHandler
 	autoCondenseContext?: boolean
-	systemPrompt?: string
+	systemPrompt: string
 }
 
 type TruncateResponse = SummarizeResponse & { prevContextTokens: number }