@librechat/agents 3.0.0-rc7 → 3.0.0-rc9

package/src/scripts/test-tools-before-handoff.ts

@@ -0,0 +1,233 @@
+import { config } from 'dotenv';
+config();
+
+import { TavilySearch } from '@langchain/tavily';
+import { HumanMessage, BaseMessage } from '@langchain/core/messages';
+import { Run } from '@/run';
+import { Providers, GraphEvents } from '@/common';
+import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
+import { ToolEndHandler, ModelEndHandler } from '@/events';
+import type * as t from '@/types';
+
+const conversationHistory: BaseMessage[] = [];
+
+/**
+ * Test edge case: Agent performs 2 web searches before handing off
+ *
+ * This tests how the system behaves when an agent with handoff capabilities
+ * uses tools before transferring control to another agent.
+ */
+async function testToolsBeforeHandoff() {
+  console.log('Testing Tools Before Handoff Edge Case...\n');
+
+  // Set up content aggregator
+  const { contentParts, aggregateContent } = createContentAggregator();
+
+  // Track tool calls and handoffs
+  let toolCallCount = 0;
+  let handoffOccurred = false;
+
+  // Create custom handlers
+  const customHandlers = {
+    [GraphEvents.TOOL_END]: new ToolEndHandler(undefined, (name?: string) => {
+      console.log(`\n✅ Tool completed: ${name}`);
+      return true;
+    }),
+    [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(),
+    [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
+    [GraphEvents.ON_RUN_STEP]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP,
+        data: t.StreamEventData
+      ): void => {
+        const runStepData = data as any;
+        if (runStepData?.name) {
+          console.log(`\n[${runStepData.name}] Processing...`);
+        }
+        aggregateContent({ event, data: data as t.RunStep });
+      },
+    },
+    [GraphEvents.ON_RUN_STEP_COMPLETED]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP_COMPLETED,
+        data: t.StreamEventData
+      ): void => {
+        aggregateContent({
+          event,
+          data: data as unknown as { result: t.ToolEndEvent },
+        });
+      },
+    },
+    [GraphEvents.ON_MESSAGE_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_MESSAGE_DELTA,
+        data: t.StreamEventData
+      ): void => {
+        // console.log('====== ON_MESSAGE_DELTA ======');
+        console.dir(data, { depth: null });
+        aggregateContent({ event, data: data as t.MessageDeltaEvent });
+      },
+    },
+    [GraphEvents.TOOL_START]: {
+      handle: (
+        _event: string,
+        data: t.StreamEventData,
+        metadata?: Record<string, unknown>
+      ): void => {
+        const toolData = data as any;
+        console.log(`\n🔧 Tool started:`);
+        console.dir({ toolData, metadata }, { depth: null });
+
+        if (toolData?.output?.name === 'tavily_search_results_json') {
+          toolCallCount++;
+          console.log(`📊 Search #${toolCallCount} initiated`);
+        } else if (toolData?.output?.name?.includes('transfer_to_')) {
+          handoffOccurred = true;
+          const specialist = toolData.name.replace('transfer_to_', '');
+          console.log(`\n🔀 Handoff initiated to: ${specialist}`);
+        }
+      },
+    },
+  };
+
+  // Create the graph with research agent and report writer
+  function createGraphWithToolsAndHandoff(): t.RunConfig {
+    const agents: t.AgentInputs[] = [
+      {
+        agentId: 'research_coordinator',
+        provider: Providers.OPENAI,
+        clientOptions: {
+          modelName: 'gpt-4.1-mini',
+          apiKey: process.env.OPENAI_API_KEY,
+        },
+        tools: [new TavilySearch({ maxResults: 3 })],
+        instructions: `You are a Research Coordinator with access to web search and a report writer specialist.
+        
+        Your workflow MUST follow these steps IN ORDER:
+        1. FIRST: Write an initial response acknowledging the request and outlining your research plan
+           - Explain what aspects you'll investigate
+           - Describe your search strategy
+        2. SECOND: Conduct exactly 2 web searches to gather comprehensive information
+           - Search 1: Get general information about the topic
+           - Search 2: Get specific details, recent updates, or complementary data
+           - Note: Even if your searches are unsuccessful, you MUST still proceed to handoff after EXACTLY 2 searches
+        3. FINALLY: After completing both searches, transfer to the report writer
+           - Provide the report writer with a summary of your findings
+        
+        CRITICAL: You MUST write your initial response before ANY tool use. Then complete both searches before handoff.`,
+        maxContextTokens: 8000,
+      },
+      {
+        agentId: 'report_writer',
+        provider: Providers.OPENAI,
+        clientOptions: {
+          modelName: 'gpt-5-mini',
+          apiKey: process.env.OPENAI_API_KEY,
+        },
+        instructions: `You are a Report Writer specialist. Your role is to:
+        1. Receive research findings from the Research Coordinator
+        2. Create a well-structured, comprehensive report
+        3. Include all key findings from the research
+        4. Format the report with clear sections and bullet points
+        5. Add a brief executive summary at the beginning
+        
+        Focus on clarity, completeness, and professional presentation.`,
+        maxContextTokens: 8000,
+      },
+    ];
+
+    // Create edge from research coordinator to report writer
+    const edges: t.GraphEdge[] = [
+      {
+        from: 'research_coordinator',
+        to: 'report_writer',
+        description: 'Transfer to report writer after completing research',
+        edgeType: 'handoff',
+      },
+    ];
+
+    return {
+      runId: `tools-before-handoff-${Date.now()}`,
+      graphConfig: {
+        type: 'multi-agent',
+        agents,
+        edges,
+      },
+      customHandlers,
+      returnContent: true,
+    };
+  }
+
+  try {
+    // Single test query that requires research before report writing
+    const query = `Research the latest developments in quantum computing from 2025, 
+    including major breakthroughs and commercial applications. 
+    I need a comprehensive report with recent findings.`;
+
+    console.log('='.repeat(60));
+    console.log(`USER QUERY: "${query}"`);
+    console.log('='.repeat(60));
+
+    // Create the graph
+    const runConfig = createGraphWithToolsAndHandoff();
+    const run = await Run.create(runConfig);
+
+    console.log('\nExpected behavior:');
+    console.log('1. Research Coordinator writes initial response/plan');
+    console.log('2. Research Coordinator performs 2 web searches');
+    console.log('3. Research Coordinator hands off to Report Writer');
+    console.log('4. Report Writer creates final report\n');
+
+    // Process with streaming
+    conversationHistory.push(new HumanMessage(query));
+    const inputs = {
+      messages: conversationHistory,
+    };
+
+    const config = {
+      configurable: {
+        thread_id: 'tools-handoff-test-1',
+      },
+      streamMode: 'values',
+      version: 'v2' as const,
+    };
+
+    const finalContentParts = await run.processStream(inputs, config);
+    const finalMessages = run.getRunMessages();
+
+    if (finalMessages) {
+      conversationHistory.push(...finalMessages);
+    }
+
+    // Show results summary
+    console.log(`\n${'─'.repeat(60)}`);
+    console.log('EDGE CASE TEST RESULTS:');
+    console.log('─'.repeat(60));
+    console.log(`Tool calls before handoff: ${toolCallCount}`);
+    console.log(`Expected tool calls: 2`);
+    console.log(`Handoff occurred: ${handoffOccurred ? 'Yes ✅' : 'No ❌'}`);
+    console.log(
+      `Test status: ${toolCallCount === 2 && handoffOccurred ? 'PASSED ✅' : 'FAILED ❌'}`
+    );
+    console.log('─'.repeat(60));
+
+    // Display conversation history
+    console.log('\nConversation History:');
+    console.log('─'.repeat(60));
+    conversationHistory.forEach((msg, idx) => {
+      const role = msg.constructor.name.replace('Message', '');
+      console.log(`\n[${idx}] ${role}:`);
+      if (typeof msg.content === 'string') {
+        console.log(
+          msg.content.substring(0, 200) +
+            (msg.content.length > 200 ? '...' : '')
+        );
+      }
+    });
+  } catch (error) {
+    console.error('Error in tools-before-handoff test:', error);
+  }
+}
+
+// Run the test
+testToolsBeforeHandoff();

package/dist/types/scripts/abort.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/ant_web_search.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/args.d.ts

@@ -1,7 +0,0 @@
-import { Providers } from '@/common';
-export declare function getArgs(): Promise<{
-    userName: string;
-    location: string;
-    provider: Providers;
-    currentDate: string;
-}>;

package/dist/types/scripts/caching.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/cli.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/cli2.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/cli3.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/cli4.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/cli5.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/code_exec.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/code_exec_files.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/code_exec_simple.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/content.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/empty_input.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/handoff-test.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/image.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/memory.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/multi-agent-conditional.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/multi-agent-parallel.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/multi-agent-sequence.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/multi-agent-supervisor.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/multi-agent-test.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/search.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/simple.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/stream.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/test-custom-prompt-key.d.ts

@@ -1,2 +0,0 @@
-#!/usr/bin/env bun
-export {};

package/dist/types/scripts/test-handoff-input.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/test-multi-agent-list-handoff.d.ts

@@ -1,2 +0,0 @@
-#!/usr/bin/env bun
-export {};

package/dist/types/scripts/thinking.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/scripts/tools.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/types/specs/spec.utils.d.ts

@@ -1 +0,0 @@
-export declare function capitalizeFirstLetter(string: string): string;

package/src/scripts/multi-agent-example-output.md

@@ -1,110 +0,0 @@
-# Multi-Agent Test Scripts - Example Output
-
-## multi-agent-test.ts
-
-```
-Testing Multi-Agent Handoff System...
-
-Invoking multi-agent graph...
-
-====== ON_RUN_STEP ======
-{ name: 'flight_assistant', type: 'agent', ... }
-[flight_assistant] Starting...
-
-====== CHAT_MODEL_STREAM ======
-I'll help you book a flight from Boston to New York...
-
-====== TOOL_START ======
-{ name: 'transfer_to_hotel_assistant', ... }
-
-====== ON_RUN_STEP_COMPLETED ======
-[flight_assistant] Completed
-
-====== ON_RUN_STEP ======
-{ name: 'hotel_assistant', type: 'agent', ... }
-[hotel_assistant] Starting...
-
-====== CHAT_MODEL_STREAM ======
-Great! I'll help you find a hotel near Times Square...
-
-Final content parts:
-[
-  { type: 'text', text: "I'll help you book a flight..." },
-  { type: 'tool_use', name: 'transfer_to_hotel_assistant', ... },
-  { type: 'text', text: "Great! I'll help you find a hotel..." }
-]
-```
-
-## multi-agent-parallel.ts
-
-```
-Testing Parallel Multi-Agent System (Fan-in/Fan-out)...
-
-Invoking parallel multi-agent graph...
-
-====== ON_RUN_STEP ======
-[researcher] Starting analysis...
-
-====== ON_RUN_STEP ======
-[analyst1] Starting analysis...
-[analyst2] Starting analysis...
-[analyst3] Starting analysis...
-
-====== CHAT_MODEL_STREAM ======
-[analyst1] From a financial perspective...
-[analyst2] From a technical perspective...
-[analyst3] From a market perspective...
-
-====== ON_RUN_STEP_COMPLETED ======
-[analyst1] Completed analysis
-[analyst2] Completed analysis
-[analyst3] Completed analysis
-
-====== ON_RUN_STEP ======
-[summarizer] Starting analysis...
-
-Final content parts: 5 parts
-```
-
-## multi-agent-conditional.ts
-
-```
-Testing Conditional Multi-Agent System...
-
---- Processing: "How do I implement a binary search tree in Python?" ---
-
-====== ON_RUN_STEP ======
-{ name: 'classifier', type: 'agent', ... }
-
-====== ON_RUN_STEP ======
-{ name: 'technical_expert', type: 'agent', ... }
-Routing to: technical_expert
-
-====== CHAT_MODEL_STREAM ======
-To implement a binary search tree in Python, you'll need to create a Node class...
-
-Expert used: technical_expert
-Content parts: 2
----
-
---- Processing: "What are the key strategies for market expansion?" ---
-
-Routing to: business_expert
-...
-
---- Processing: "What is the capital of France?" ---
-
-Routing to: general_assistant
-...
-```
-
-## Key Features of Updated Scripts
-
-1. **Proper Event Handling**: All GraphEvents are properly handled with custom handlers
-2. **Content Aggregation**: Using `createContentAggregator()` to collect all content parts
-3. **Stream Output**: Real-time token streaming via `ChatModelStreamHandler`
-4. **Debug Logging**: Comprehensive event logging for debugging multi-agent flows
-5. **Conversation History**: Proper tracking of messages across agent interactions
-6. **Return Content**: `returnContent: true` ensures content parts are returned
-
-The scripts now follow the same patterns as `simple.ts` and `tools.ts`, providing consistent behavior and comprehensive event handling across all multi-agent scenarios.