Merge pull request #4999 from microsoft/dev/bhavyau/fix-tool-reference-summarization-0.43

Fix: strip tool_search messages from summarization to prevent tool_reference errors

Author: bhavyaus

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

@@ -14,6 +14,7 @@ import { IHistoricalTurn, ISessionTranscriptService } from '../../../../platform
 import { ConfigKey, IConfigurationService } from '../../../../platform/configuration/common/configurationService';
 import { isAnthropicFamily, isGeminiFamily } from '../../../../platform/endpoint/common/chatModelCapabilities';
 import { ILogService } from '../../../../platform/log/common/logService';
+import { CUSTOM_TOOL_SEARCH_NAME } from '../../../../platform/networking/common/anthropic';
 import { IChatEndpoint } from '../../../../platform/networking/common/networking';
 import { APIUsage } from '../../../../platform/networking/common/openai';
 import { IPromptPathRepresentationService } from '../../../../platform/prompts/common/promptPathRepresentationService';
@@ -714,6 +715,16 @@ class ConversationHistorySummarizer {
 			stripCacheBreakpoints(summarizationPrompt);
 
 			let messages = ToolCallingLoop.stripInternalToolCallIds(summarizationPrompt);
+
+			// Strip custom client-side tool search (tool_search) tool_use/tool_result
+			// pairs. The summarization call uses ChatLocation.Other but
+			// createMessagesRequestBody still converts tool_search results to
+			// tool_reference blocks (customToolSearchEnabled isn't gated by location).
+			// Without tool search enabled in the request, Anthropic rejects them.
+			if (isAnthropicFamily(endpoint)) {
+				messages = stripToolSearchMessages(messages);
+			}
+
 			// Gemini strictly requires every function_call to have a matching function_response.
 			// When prompt-tsx prunes tool result messages due to token budget, orphaned tool_calls
 			// can remain, causing a 400 INVALID_ARGUMENT error. Strip them for Gemini models.
@@ -914,6 +925,44 @@ function stripCacheBreakpoints(messages: ChatMessage[]): void {
 	});
 }
 
+/**
+ * Strip custom client-side tool search (tool_search) tool_use and tool_result
+ * messages from the conversation. The summarization call uses ChatLocation.Other
+ * but createMessagesRequestBody still converts tool_search results to
+ * tool_reference blocks (customToolSearchEnabled isn't gated by location).
+ * Without tool search enabled in the request, Anthropic rejects tool_reference
+ * content blocks with: "Input tag 'tool_reference' found using 'type' does not
+ * match any of the expected tags".
+ */
+export function stripToolSearchMessages(messages: ChatMessage[]): ChatMessage[] {
+	const toolSearchIds = new Set<string>();
+	for (const message of messages) {
+		if (message.role === Raw.ChatRole.Assistant && message.toolCalls) {
+			for (const tc of message.toolCalls) {
+				if (tc.function.name === CUSTOM_TOOL_SEARCH_NAME) {
+					toolSearchIds.add(tc.id);
+				}
+			}
+		}
+	}
+
+	if (toolSearchIds.size === 0) {
+		return messages;
+	}
+
+	return messages.map(message => {
+		if (message.role === Raw.ChatRole.Assistant && message.toolCalls) {
+			const filteredToolCalls = message.toolCalls.filter(tc => !toolSearchIds.has(tc.id));
+			if (filteredToolCalls.length !== message.toolCalls.length) {
+				return { ...message, toolCalls: filteredToolCalls.length > 0 ? filteredToolCalls : undefined };
+			}
+		} else if (message.role === Raw.ChatRole.Tool && message.toolCallId && toolSearchIds.has(message.toolCallId)) {
+			return undefined;
+		}
+		return message;
+	}).filter((m): m is ChatMessage => m !== undefined);
+}
+
 export interface ISummarizedConversationHistoryInfo {
 	readonly props: SummarizedAgentHistoryProps;
 	readonly summarizedToolCallRoundId: string;

src/extension/prompts/node/agent/test/summarization.spec.tsx

@@ -30,7 +30,7 @@ import { ToolName } from '../../../../tools/common/toolNames';
 import { PromptRenderer } from '../../base/promptRenderer';
 import { AgentPrompt, AgentPromptProps } from '../agentPrompt';
 import { PromptRegistry } from '../promptRegistry';
-import { ConversationHistorySummarizationPrompt, extractInlineSummary, InlineSummarizationRequestedMetadata, SummarizedConversationHistory, SummarizedConversationHistoryMetadata, SummarizedConversationHistoryPropsBuilder } from '../summarizedConversationHistory';
+import { ConversationHistorySummarizationPrompt, extractInlineSummary, InlineSummarizationRequestedMetadata, stripToolSearchMessages, SummarizedConversationHistory, SummarizedConversationHistoryMetadata, SummarizedConversationHistoryPropsBuilder } from '../summarizedConversationHistory';
 
 suite('Agent Summarization', () => {
 	let accessor: ITestingServicesAccessor;
@@ -737,3 +737,102 @@ suite('Inline Summarization Prompt', () => {
 		expect(inlineMeta).toBeDefined();
 	});
 });
+
+suite('stripToolSearchMessages', () => {
+	function makeAssistantMessage(toolCalls: { id: string; name: string }[], text = 'response'): Raw.ChatMessage {
+		return {
+			role: Raw.ChatRole.Assistant,
+			content: [{ type: Raw.ChatCompletionContentPartKind.Text, text }],
+			toolCalls: toolCalls.map(tc => ({
+				type: 'function' as const,
+				id: tc.id,
+				function: { name: tc.name, arguments: '{}' },
+			})),
+		};
+	}
+
+	function makeToolResult(toolCallId: string, text = 'result'): Raw.ChatMessage {
+		return {
+			role: Raw.ChatRole.Tool,
+			content: [{ type: Raw.ChatCompletionContentPartKind.Text, text }],
+			toolCallId,
+		};
+	}
+
+	function makeUserMessage(text = 'hello'): Raw.ChatMessage {
+		return {
+			role: Raw.ChatRole.User,
+			content: [{ type: Raw.ChatCompletionContentPartKind.Text, text }],
+		};
+	}
+
+	test('returns messages unchanged when no tool_search calls present', () => {
+		const messages = [
+			makeUserMessage(),
+			makeAssistantMessage([{ id: 'tc1', name: 'read_file' }]),
+			makeToolResult('tc1'),
+		];
+		const result = stripToolSearchMessages(messages);
+		expect(result).toBe(messages);
+	});
+
+	test('strips custom tool_search tool_use and tool_result', () => {
+		const messages = [
+			makeUserMessage(),
+			makeAssistantMessage([
+				{ id: 'tc1', name: 'read_file' },
+				{ id: 'tc2', name: 'tool_search' },
+			]),
+			makeToolResult('tc1'),
+			makeToolResult('tc2', '["read_file", "edit_file"]'),
+		];
+		const result = stripToolSearchMessages(messages);
+		expect(result).toHaveLength(3);
+		const assistant = result[1];
+		expect(assistant.role).toBe(Raw.ChatRole.Assistant);
+		if (assistant.role === Raw.ChatRole.Assistant) {
+			expect(assistant.toolCalls).toHaveLength(1);
+			expect(assistant.toolCalls![0].id).toBe('tc1');
+		}
+		expect(result.find(m => m.role === Raw.ChatRole.Tool && m.toolCallId === 'tc2')).toBeUndefined();
+	});
+
+	test('removes toolCalls property when all tool calls are tool_search', () => {
+		const messages = [
+			makeUserMessage(),
+			makeAssistantMessage([{ id: 'tc1', name: 'tool_search' }]),
+			makeToolResult('tc1'),
+		];
+		const result = stripToolSearchMessages(messages);
+		expect(result).toHaveLength(2);
+		const assistant = result[1];
+		if (assistant.role === Raw.ChatRole.Assistant) {
+			expect(assistant.toolCalls).toBeUndefined();
+		}
+	});
+
+	test('does not strip server-side tool_search_tool_regex', () => {
+		const messages = [
+			makeUserMessage(),
+			makeAssistantMessage([{ id: 'tc1', name: 'tool_search_tool_regex' }]),
+			makeToolResult('tc1'),
+		];
+		const result = stripToolSearchMessages(messages);
+		expect(result).toBe(messages);
+	});
+
+	test('preserves non-tool messages', () => {
+		const messages = [
+			makeUserMessage('first'),
+			makeAssistantMessage([{ id: 'tc1', name: 'tool_search' }]),
+			makeToolResult('tc1'),
+			makeUserMessage('second'),
+			makeAssistantMessage([{ id: 'tc2', name: 'edit_file' }]),
+			makeToolResult('tc2'),
+		];
+		const result = stripToolSearchMessages(messages);
+		expect(result).toHaveLength(5);
+		expect(result[0].content[0]).toEqual({ type: Raw.ChatCompletionContentPartKind.Text, text: 'first' });
+		expect(result[2].content[0]).toEqual({ type: Raw.ChatCompletionContentPartKind.Text, text: 'second' });
+	});
+});