provideChatSummary() uses advanced summarizer (#1189)

Author: joshspicer

src/extension/prompt/node/summarizer.ts

@@ -9,55 +9,13 @@ import { ChatFetchResponseType, ChatLocation } from '../../../platform/chat/comm
 import { IEndpointProvider } from '../../../platform/endpoint/common/endpointProvider';
 import { ILogService } from '../../../platform/log/common/logService';
 import { IInstantiationService } from '../../../util/vs/platform/instantiation/common/instantiation';
+import { ConversationHistorySummarizationPrompt } from '../../prompts/node/agent/summarizedConversationHistory';
+import { renderPromptElement } from '../../prompts/node/base/promptRenderer';
+import { ChatVariablesCollection } from '../common/chatVariablesCollection';
 import { TurnStatus } from '../common/conversation';
+import { IBuildPromptContext } from '../common/intents';
 import { addHistoryToConversation } from './chatParticipantRequestHandler';
 
-// Simple prompt template for summarization
-// Consider adopting the more sophisticated summarizedConversationHistory.tsx
-class SummaryPrompt {
-	render() {
-		return {
-			messages: [
-				{
-					role: 'system' as const,
-					content: `You are an expert at summarizing chat conversations.
-
-You will be provided:
-
-- A series of user/assistant message pairs in chronological order
-- A final user message indicating the user's intent.
-
-Your task is to:
-
-- Create a detailed summary of the conversation that captures the user's intent and key information.
-
-Keep in mind:
-
-- The user is iterating on a feature specification, bug fix, or other common programming task.
-- There may be relevant code snippets or files referenced in the conversation.
-- The user is collaborating with the assistant to refine their ideas and solutions, course-correcting the assistant as needed.
-- The user will provide feedback on the assistant's suggestions and may request changes or improvements.
-- Disregard messages that the user has indicated are incorrect, irrelevant, or unhelpful.
-- Preserve relevant and actionable context and key information.
-- If the conversation is long or discusses several tasks, keep the summary focused on the task indicated by the user's intent.
-- Always prefer decisions in later messages over earlier ones.
-
-Structure your summary using the following format:
-
-TITLE: A brief title for the summary
-USER INTENT: The user's goal or intent for the conversation
-TASK DESCRIPTION: Main technical goals and user requirements
-EXISTING: What has already been accomplished. Include file paths and other direct references.
-PENDING: What still needs to be done. Include file paths and other direct references.
-CODE STATE: A list of all files discussed or modified. Provide code snippets or diffs that illustrate important context.
-RELEVANT CODE/DOCUMENTATION SNIPPETS: Key code or documentation snippets from referenced files or discussions.
-OTHER NOTES: Any additional context or information that may be relevant.`
-				}
-			]
-		};
-	}
-}
-
 export class ChatSummarizerProvider implements vscode.ChatSummarizer {
 
 	constructor(
@@ -77,43 +35,40 @@ export class ChatSummarizerProvider implements vscode.ChatSummarizer {
 		}
 
 		const endpoint = await this.endpointProvider.getChatEndpoint('gpt-4o-mini');
-		const summaryPrompt = new SummaryPrompt();
-		const { messages: systemMessages } = summaryPrompt.render();
-
-		// Condense each turn into a single user message containing both request and response
-		const conversationContent = turns
-			.filter(turn => turn.request?.message)
-			.map(turn => {
-				const userMsg = turn.request?.message || '';
-				const assistantMsg = turn.responseMessage?.message || '';
-				if (assistantMsg) {
-					return `User: ${userMsg}\n\nAssistant: ${assistantMsg}`;
-				} else {
-					return `User: ${userMsg}`;
-				}
-			})
-			.join('\n\n---\n\n');
-
-		const conversationMessages: Raw.ChatMessage[] = [
-			{
-				role: Raw.ChatRole.User,
-				content: [{
-					type: Raw.ChatCompletionContentPartKind.Text,
-					text: `Here is the conversation to summarize:\n\n${conversationContent}`
-				}]
-			}
-		];
+		const promptContext: IBuildPromptContext = {
+			requestId: 'chat-summary',
+			query: '',
+			history: turns,
+			chatVariables: new ChatVariablesCollection(),
+			isContinuation: false,
+			toolCallRounds: undefined,
+			toolCallResults: undefined,
+		};
 
-		const allMessages: Raw.ChatMessage[] = [
-			{
-				role: Raw.ChatRole.System,
-				content: [{
-					type: Raw.ChatCompletionContentPartKind.Text,
-					text: systemMessages[0].content as string
-				}]
-			},
-			...conversationMessages
-		];
+		let allMessages: Raw.ChatMessage[];
+		try {
+			const rendered = await renderPromptElement(
+				this.instantiationService,
+				endpoint,
+				ConversationHistorySummarizationPrompt,
+				{
+					priority: 0,
+					endpoint,
+					location: ChatLocation.Panel,
+					promptContext,
+					maxToolResultLength: 2000,
+					triggerSummarize: false,
+					simpleMode: false,
+					maxSummaryTokens: 7_000,
+				},
+				undefined,
+				token
+			);
+			allMessages = rendered.messages;
+		} catch (err) {
+			this.logService.error(`Failed to render conversation summarization prompt: ${err instanceof Error ? err.message : String(err)}`);
+			return '';
+		}
 
 		const response = await endpoint.makeChatRequest(
 			'summarize',

src/extension/prompts/node/agent/summarizedConversationHistory.tsx

@@ -320,6 +320,8 @@ export interface SummarizedAgentHistoryProps extends BasePromptElementProps {
 	readonly enableCacheBreakpoints?: boolean;
 	readonly workingNotebook?: NotebookDocument;
 	readonly maxToolResultLength: number;
+	/** Optional hard cap on summary tokens; effective budget = min(prompt sizing tokenBudget, this value) */
+	readonly maxSummaryTokens?: number;
 }
 
 /**
@@ -524,9 +526,13 @@ class ConversationHistorySummarizer {
 		}
 
 		const summarySize = await this.sizing.countTokens(response.value);
-		if (summarySize > this.sizing.tokenBudget) {
+		const effectiveBudget =
+			!!this.props.maxSummaryTokens
+				? Math.min(this.sizing.tokenBudget, this.props.maxSummaryTokens)
+				: this.sizing.tokenBudget;
+		if (summarySize > effectiveBudget) {
 			this.sendSummarizationTelemetry('too_large', response.requestId, this.props.endpoint.model, mode, elapsedTime, response.usage);
-			this.logInfo(`Summary too large: ${summarySize} tokens`, mode);
+			this.logInfo(`Summary too large: ${summarySize} tokens (effective budget ${effectiveBudget})`, mode);
 			throw new Error('Summary too large');
 		}