@illuma-ai/agents 1.1.21 → 1.1.23

package/src/scripts/test-parallel-agent-labeling.js

@@ -1,253 +0,0 @@
-#!/usr/bin/env bun
-import { config } from 'dotenv';
-config();
-import { HumanMessage } from '@langchain/core/messages';
-import { labelContentByAgent, formatAgentMessages } from '@/messages/format';
-import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
-import { Providers, GraphEvents } from '@/common';
-import { ToolEndHandler, ModelEndHandler } from '@/events';
-import { Run } from '@/run';
-const conversationHistory = [];
-/**
- * Test parallel multi-agent system with agent labeling on subsequent runs
- *
- * Graph structure:
- * START -> researcher
- * researcher -> [analyst1, analyst2, analyst3] (fan-out)
- * [analyst1, analyst2, analyst3] -> summarizer (fan-in)
- * summarizer -> END
- */
-async function testParallelWithAgentLabeling() {
-    console.log('Testing Parallel Multi-Agent with Agent Labeling...\n');
-    // Set up content aggregator
-    const { contentParts, aggregateContent } = createContentAggregator();
-    // Define specialized agents
-    const agents = [
-        {
-            agentId: 'researcher',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are a research coordinator. Analyze the request and provide 2-3 sentence coordination brief.`,
-        },
-        {
-            agentId: 'analyst1',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are FINANCIAL ANALYST. Provide 2-3 sentence financial analysis. Start with "FINANCIAL ANALYSIS:"`,
-        },
-        {
-            agentId: 'analyst2',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are TECHNICAL ANALYST. Provide 2-3 sentence technical analysis. Start with "TECHNICAL ANALYSIS:"`,
-        },
-        {
-            agentId: 'analyst3',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are MARKET ANALYST. Provide 2-3 sentence market analysis. Start with "MARKET ANALYSIS:"`,
-        },
-        {
-            agentId: 'summarizer',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are SYNTHESIS EXPERT. Review all analyses and provide 2-3 sentence integrated summary.`,
-        },
-    ];
-    // Define direct edges (fan-out and fan-in)
-    const edges = [
-        {
-            from: 'researcher',
-            to: ['analyst1', 'analyst2', 'analyst3'],
-            description: 'Distribute research to specialist analysts',
-            edgeType: 'sequence',
-        },
-        {
-            from: ['analyst1', 'analyst2', 'analyst3'],
-            to: 'summarizer',
-            description: 'Aggregate analysis results',
-            edgeType: 'sequence',
-            prompt: 'Based on the analyses below, provide an integrated summary:\n\n{results}',
-        },
-    ];
-    // Create custom handlers
-    const customHandlers = {
-        [GraphEvents.TOOL_END]: new ToolEndHandler(),
-        [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(),
-        [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
-        [GraphEvents.ON_RUN_STEP_COMPLETED]: {
-            handle: (event, data) => {
-                aggregateContent({
-                    event,
-                    data: data,
-                });
-            },
-        },
-        [GraphEvents.ON_RUN_STEP]: {
-            handle: (event, data) => {
-                aggregateContent({ event, data: data });
-            },
-        },
-        [GraphEvents.ON_MESSAGE_DELTA]: {
-            handle: (event, data) => {
-                aggregateContent({ event, data: data });
-            },
-        },
-    };
-    try {
-        const query = 'What are the implications of widespread AI adoption?';
-        console.log(`${'='.repeat(80)}`);
-        console.log(`FIRST RUN - USER QUERY: "${query}"`);
-        console.log('='.repeat(80));
-        // Reset conversation
-        conversationHistory.length = 0;
-        conversationHistory.push(new HumanMessage(query));
-        // Create graph
-        const runConfig = {
-            runId: `parallel-test-${Date.now()}`,
-            graphConfig: {
-                type: 'multi-agent',
-                agents,
-                edges,
-            },
-            customHandlers,
-            returnContent: true,
-            skipCleanup: true,
-        };
-        const run = await Run.create(runConfig);
-        console.log('\nProcessing first run with parallel agents...\n');
-        const config = {
-            configurable: {
-                thread_id: 'parallel-agent-labeling-1',
-            },
-            streamMode: 'values',
-            version: 'v2',
-        };
-        const inputs = {
-            messages: conversationHistory,
-        };
-        await run.processStream(inputs, config);
-        const finalMessages = run.getRunMessages();
-        if (finalMessages) {
-            conversationHistory.push(...finalMessages);
-        }
-        // Show agent participation
-        console.log(`\n${'─'.repeat(80)}`);
-        console.log('FIRST RUN - AGENT PARTICIPATION:');
-        console.log('─'.repeat(80));
-        if (run.Graph) {
-            const activeAgents = run.Graph.getActiveAgentIds();
-            console.log(`\nActive agents (${activeAgents.length}):`, activeAgents);
-            const stepsByAgent = run.Graph.getRunStepsByAgent();
-            stepsByAgent.forEach((steps, agentId) => {
-                console.log(`  ${agentId}: ${steps.length} steps`);
-            });
-            console.log(`\nTotal content parts: ${contentParts.length}`);
-        }
-        // =============================================================
-        // SECOND RUN: Test with agent-labeled history
-        // =============================================================
-        console.log(`\n${'='.repeat(80)}`);
-        console.log(`SECOND RUN - Simulating DB Load with Parallel Agent Labeling`);
-        console.log('='.repeat(80));
-        // Simulate DB storage
-        const dbStoredContentParts = [...contentParts];
-        const dbStoredAgentIdMap = Object.fromEntries(run.Graph.getContentPartAgentMap());
-        console.log('\n📦 Simulating DB storage:');
-        console.log(`  - Stored ${dbStoredContentParts.length} content parts`);
-        console.log(`  - Stored agent mappings for ${Object.keys(dbStoredAgentIdMap).length} parts`);
-        // Load and label by agent with labelNonTransferContent option
-        console.log('\n📥 Loading from DB and labeling ALL agent content...');
-        const agentNames = {
-            researcher: 'Researcher',
-            analyst1: 'Financial Analyst',
-            analyst2: 'Technical Analyst',
-            analyst3: 'Market Analyst',
-            summarizer: 'Synthesizer',
-        };
-        const labeledContentParts = labelContentByAgent(dbStoredContentParts.filter((p) => p != null), dbStoredAgentIdMap, agentNames, { labelNonTransferContent: true } // NEW: Label all content
-        );
-        console.log(`  - Labeled ${labeledContentParts.length} content groups by agent`);
-        // Convert to payload
-        const payload = [
-            {
-                role: 'user',
-                content: query,
-            },
-            {
-                role: 'assistant',
-                content: labeledContentParts,
-            },
-        ];
-        // Format using formatAgentMessages
-        console.log('\n🔧 Calling formatAgentMessages...');
-        const { messages: formattedMessages } = formatAgentMessages(payload);
-        console.log(`  - Formatted into ${formattedMessages.length} BaseMessages`);
-        // Show preview
-        console.log('\n👁️  Preview of formatted history:');
-        console.log('─'.repeat(80));
-        for (let i = 0; i < formattedMessages.length; i++) {
-            const msg = formattedMessages[i];
-            const role = msg._getType();
-            const preview = typeof msg.content === 'string'
-                ? msg.content.slice(0, 300)
-                : JSON.stringify(msg.content).slice(0, 300);
-            console.log(`[${i}] ${role}: ${preview}${preview.length >= 300 ? '...' : ''}`);
-            console.log('');
-        }
-        console.log('─'.repeat(80));
-        // Create a second run with labeled history
-        console.log('\n🚀 Starting second run with agent-labeled parallel history...');
-        const followupQuery = 'Which analyst identified the most significant risk?';
-        console.log(`   Followup: "${followupQuery}"`);
-        const secondRunHistory = [
-            ...formattedMessages,
-            new HumanMessage(followupQuery),
-        ];
-        const runConfig2 = {
-            runId: `parallel-test-2-${Date.now()}`,
-            graphConfig: {
-                type: 'multi-agent',
-                agents,
-                edges,
-            },
-            customHandlers,
-            returnContent: true,
-            skipCleanup: true,
-        };
-        const run2 = await Run.create(runConfig2);
-        const inputs2 = {
-            messages: secondRunHistory,
-        };
-        await run2.processStream(inputs2, config);
-        console.log('\n✅ Second run completed successfully!');
-        console.log('   The researcher correctly understood that parallel analysts handled');
-        console.log('   the previous analysis, with clear attribution per agent.');
-        console.log(`\n${'='.repeat(80)}`);
-        console.log('TEST COMPLETE');
-        console.log('='.repeat(80));
-        console.log('\nThis demonstrates that parallel multi-agent patterns work correctly');
-        console.log('with agent labeling, preventing confusion about who said what.');
-    }
-    catch (error) {
-        console.error('Error in parallel agent labeling test:', error);
-    }
-}
-// Run the test
-testParallelWithAgentLabeling();
-//# sourceMappingURL=test-parallel-agent-labeling.js.map

package/src/scripts/test-parallel-handoffs.js

@@ -1,229 +0,0 @@
-import { config } from 'dotenv';
-config();
-import { HumanMessage } from '@langchain/core/messages';
-import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
-import { ToolEndHandler } from '@/events';
-import { Providers, GraphEvents } from '@/common';
-import { sleep } from '@/utils/run';
-import { Run } from '@/run';
-const conversationHistory = [];
-/**
- * Test parallel handoffs - where an LLM calls multiple transfer tools simultaneously
- *
- * Graph structure:
- * coordinator -> [researcher, writer] (via parallel handoff tools)
- *
- * The coordinator agent has two transfer tools:
- * - transfer_to_researcher
- * - transfer_to_writer
- *
- * When given a task that needs both, it should call both tools in parallel.
- */
-async function testParallelHandoffs() {
-    console.log('Testing Parallel Handoffs (LLM calling multiple transfers)...\n');
-    const { contentParts, aggregateContent } = createContentAggregator();
-    const agents = [
-        {
-            agentId: 'coordinator',
-            provider: Providers.OPENAI,
-            clientOptions: {
-                modelName: 'gpt-4o-mini',
-                apiKey: process.env.OPENAI_API_KEY,
-            },
-            instructions: `You are a COORDINATOR agent. Your job is to delegate tasks to specialized agents.
-
-You have access to two transfer tools:
-- transfer_to_researcher: For research and fact-finding tasks
-- transfer_to_writer: For content creation and writing tasks
-
-IMPORTANT: When a task requires BOTH research AND writing, you MUST call BOTH transfer tools SIMULTANEOUSLY in the same response. Do not call them sequentially.
-
-For example, if asked to "research and write about X", call both transfers at once to enable parallel work.
-
-When delegating, provide clear instructions to each agent about what they should do.`,
-        },
-        {
-            agentId: 'researcher',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are a RESEARCHER. When you receive a task:
-1. Provide concise research findings (100-150 words)
-2. Start your response with "📚 RESEARCH FINDINGS:"`,
-        },
-        {
-            agentId: 'writer',
-            provider: Providers.ANTHROPIC,
-            clientOptions: {
-                modelName: 'claude-haiku-4-5',
-                apiKey: process.env.ANTHROPIC_API_KEY,
-            },
-            instructions: `You are a WRITER. When you receive a task:
-1. Provide creative content (100-150 words)
-2. Start your response with "✍️ WRITTEN CONTENT:"`,
-        },
-    ];
-    /**
-     * Create handoff edges from coordinator to both researcher and writer.
-     * These are separate edges so the LLM sees both transfer tools.
-     */
-    const edges = [
-        {
-            from: 'coordinator',
-            to: 'researcher',
-            edgeType: 'transfer',
-            description: 'Transfer to researcher for research and fact-finding tasks',
-            prompt: 'Research task instructions',
-        },
-        {
-            from: 'coordinator',
-            to: 'writer',
-            edgeType: 'transfer',
-            description: 'Transfer to writer for content creation and writing tasks',
-            prompt: 'Writing task instructions',
-        },
-    ];
-    /** Track which agents are active and their timing */
-    const activeAgents = new Set();
-    const agentTimings = {};
-    const startTime = Date.now();
-    const customHandlers = {
-        [GraphEvents.TOOL_END]: new ToolEndHandler(),
-        [GraphEvents.CHAT_MODEL_END]: {
-            handle: (_event, _data, metadata) => {
-                const nodeName = metadata?.langgraph_node;
-                if (nodeName) {
-                    const elapsed = Date.now() - startTime;
-                    agentTimings[nodeName] = agentTimings[nodeName] || {};
-                    agentTimings[nodeName].end = elapsed;
-                    activeAgents.delete(nodeName);
-                    console.log(`\n⏱️  [${nodeName}] COMPLETED at ${elapsed}ms`);
-                }
-            },
-        },
-        [GraphEvents.CHAT_MODEL_START]: {
-            handle: (_event, _data, metadata) => {
-                const nodeName = metadata?.langgraph_node;
-                if (nodeName) {
-                    const elapsed = Date.now() - startTime;
-                    /** Store first start time for parallel overlap calculation */
-                    if (!agentTimings[nodeName]?.start) {
-                        agentTimings[nodeName] = agentTimings[nodeName] || {};
-                        agentTimings[nodeName].start = elapsed;
-                    }
-                    activeAgents.add(nodeName);
-                    console.log(`\n⏱️  [${nodeName}] STARTED at ${elapsed}ms`);
-                    console.log(`   Active agents: ${Array.from(activeAgents).join(', ')}`);
-                }
-            },
-        },
-        [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
-        [GraphEvents.ON_RUN_STEP_COMPLETED]: {
-            handle: (event, data) => {
-                aggregateContent({
-                    event,
-                    data: data,
-                });
-            },
-        },
-        [GraphEvents.ON_RUN_STEP]: {
-            handle: (event, data) => {
-                aggregateContent({ event, data: data });
-            },
-        },
-        [GraphEvents.ON_RUN_STEP_DELTA]: {
-            handle: (event, data) => {
-                aggregateContent({ event, data: data });
-            },
-        },
-        [GraphEvents.ON_MESSAGE_DELTA]: {
-            handle: (event, data) => {
-                aggregateContent({ event, data: data });
-            },
-        },
-    };
-    const runConfig = {
-        runId: `parallel-handoffs-${Date.now()}`,
-        graphConfig: {
-            type: 'multi-agent',
-            agents,
-            edges,
-        },
-        customHandlers,
-        returnContent: true,
-        skipCleanup: true,
-    };
-    try {
-        const run = await Run.create(runConfig);
-        /** Prompt designed to trigger parallel handoffs without confusing language */
-        const userMessage = `Help me with two topics:
-1. The history of the internet
-2. A short poem about technology
-
-I need information on both topics.`;
-        conversationHistory.push(new HumanMessage(userMessage));
-        console.log('User message:', userMessage);
-        console.log('\nInvoking multi-agent graph with parallel handoff request...\n');
-        const config = {
-            configurable: {
-                thread_id: 'parallel-handoffs-test-1',
-            },
-            streamMode: 'values',
-            version: 'v2',
-        };
-        const inputs = {
-            messages: conversationHistory,
-        };
-        await run.processStream(inputs, config);
-        const finalMessages = run.getRunMessages();
-        if (finalMessages) {
-            conversationHistory.push(...finalMessages);
-        }
-        /** Analyze parallel execution */
-        console.log('\n\n========== TIMING SUMMARY ==========');
-        console.log('Available timing keys:', Object.keys(agentTimings));
-        for (const [agent, timing] of Object.entries(agentTimings)) {
-            const duration = timing.end && timing.start ? timing.end - timing.start : 'N/A';
-            console.log(`${agent}: started=${timing.start}ms, ended=${timing.end}ms, duration=${duration}ms`);
-        }
-        /** Check if researcher and writer ran in parallel (handle key variations) */
-        const researcherKey = Object.keys(agentTimings).find((k) => k.includes('researcher'));
-        const writerKey = Object.keys(agentTimings).find((k) => k.includes('writer'));
-        const researcherTiming = researcherKey
-            ? agentTimings[researcherKey]
-            : undefined;
-        const writerTiming = writerKey ? agentTimings[writerKey] : undefined;
-        if (researcherTiming && writerTiming) {
-            const bothStarted = researcherTiming.start && writerTiming.start;
-            const bothEnded = researcherTiming.end && writerTiming.end;
-            if (bothStarted && bothEnded) {
-                const overlap = Math.min(researcherTiming.end, writerTiming.end) -
-                    Math.max(researcherTiming.start, writerTiming.start);
-                if (overlap > 0) {
-                    console.log(`\n✅ PARALLEL HANDOFFS SUCCESSFUL: ${overlap}ms overlap between researcher and writer`);
-                }
-                else {
-                    console.log(`\n⚠️  SEQUENTIAL EXECUTION: researcher and writer did not overlap`);
-                    console.log(`   This may indicate the LLM called transfers sequentially, not in parallel`);
-                }
-            }
-        }
-        else {
-            console.log('\n⚠️  Not all agents were invoked. Check if handoffs occurred.');
-            console.log('   researcher timing:', researcherTiming);
-            console.log('   writer timing:', writerTiming);
-        }
-        console.log('====================================\n');
-        console.log('Final content parts:', contentParts.length, 'parts');
-        console.dir(contentParts, { depth: null });
-        await sleep(3000);
-    }
-    catch (error) {
-        console.error('Error in parallel handoffs test:', error);
-        throw error;
-    }
-}
-testParallelHandoffs();
-//# sourceMappingURL=test-parallel-handoffs.js.map

package/src/scripts/test-thinking-handoff-bedrock.js

@@ -1,132 +0,0 @@
-#!/usr/bin/env bun
-import { config } from 'dotenv';
-config();
-import { HumanMessage } from '@langchain/core/messages';
-import { Run } from '@/run';
-import { ChatModelStreamHandler } from '@/stream';
-import { Providers, GraphEvents } from '@/common';
-import { ToolEndHandler, ModelEndHandler } from '@/events';
-const conversationHistory = [];
-/**
- * Test edge case: switching from OpenAI supervisor (no thinking) to Bedrock specialist (with thinking enabled)
- * This should not throw an error about missing thinking blocks
- */
-async function testBedrockThinkingHandoff() {
-    console.log('Testing OpenAI → Bedrock (with thinking) handoff...\n');
-    // Create custom handlers
-    const customHandlers = {
-        [GraphEvents.TOOL_END]: new ToolEndHandler(),
-        [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(),
-        [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
-        [GraphEvents.TOOL_START]: {
-            handle: (_event, data) => {
-                const toolData = data;
-                if (toolData?.name) {
-                    console.log(`\n🔧 Tool called: ${toolData.name}`);
-                }
-            },
-        },
-    };
-    // Create the graph configuration
-    function createGraphConfig() {
-        console.log('Creating graph with OpenAI supervisor and Bedrock specialist with thinking enabled.\n');
-        const agents = [
-            {
-                agentId: 'supervisor',
-                provider: Providers.OPENAI,
-                clientOptions: {
-                    modelName: 'gpt-4o-mini',
-                    apiKey: process.env.OPENAI_API_KEY,
-                },
-                instructions: `You are a task supervisor. When the user asks about code review, use transfer_to_code_reviewer to hand off to the specialist.`,
-                maxContextTokens: 8000,
-            },
-            {
-                agentId: 'code_reviewer',
-                provider: Providers.BEDROCK,
-                clientOptions: {
-                    region: process.env.BEDROCK_AWS_REGION || 'us-east-1',
-                    model: 'us.anthropic.claude-3-7-sonnet-20250219-v1:0',
-                    credentials: {
-                        accessKeyId: process.env.BEDROCK_AWS_ACCESS_KEY_ID,
-                        secretAccessKey: process.env.BEDROCK_AWS_SECRET_ACCESS_KEY,
-                    },
-                    additionalModelRequestFields: {
-                        thinking: {
-                            type: 'enabled',
-                            budget_tokens: 2000,
-                        },
-                    },
-                },
-                instructions: `You are a code review specialist using Bedrock with extended thinking. Think carefully about the code quality, best practices, and potential issues. Provide thoughtful feedback.`,
-                maxContextTokens: 8000,
-            },
-        ];
-        const edges = [
-            {
-                from: 'supervisor',
-                to: ['code_reviewer'],
-                description: 'Transfer to code review specialist',
-                edgeType: 'transfer',
-            },
-        ];
-        return {
-            runId: `bedrock-thinking-handoff-test-${Date.now()}`,
-            graphConfig: {
-                type: 'multi-agent',
-                agents,
-                edges,
-            },
-            customHandlers,
-            returnContent: true,
-            skipCleanup: true,
-        };
-    }
-    try {
-        // Test query that should trigger a handoff
-        const query = 'Can you review this function and tell me if there are any issues?\n\nfunction add(a, b) { return a + b; }';
-        console.log(`${'='.repeat(60)}`);
-        console.log(`USER QUERY: "${query}"`);
-        console.log('='.repeat(60));
-        // Initialize conversation
-        conversationHistory.push(new HumanMessage(query));
-        // Create and run the graph
-        const runConfig = createGraphConfig();
-        const run = await Run.create(runConfig);
-        const config = {
-            configurable: {
-                thread_id: 'bedrock-thinking-handoff-test-1',
-            },
-            streamMode: 'values',
-            version: 'v2',
-        };
-        console.log('\nProcessing request...\n');
-        // Process with streaming
-        const inputs = {
-            messages: conversationHistory,
-        };
-        await run.processStream(inputs, config);
-        const finalMessages = run.getRunMessages();
-        if (finalMessages) {
-            conversationHistory.push(...finalMessages);
-        }
-        // Success!
-        console.log(`\n${'='.repeat(60)}`);
-        console.log('✅ TEST PASSED');
-        console.log('='.repeat(60));
-        console.log('\nSuccessfully handed off from OpenAI (no thinking) to');
-        console.log('Bedrock with thinking enabled without errors!');
-        console.log('\nThe ensureThinkingBlockInMessages() function correctly');
-        console.log('handled the transition by converting tool sequences to');
-        console.log('HumanMessages before calling the Bedrock API.');
-    }
-    catch (error) {
-        console.error('\n❌ TEST FAILED');
-        console.error('='.repeat(60));
-        console.error('Error:', error);
-        process.exit(1);
-    }
-}
-// Run the test
-testBedrockThinkingHandoff();
-//# sourceMappingURL=test-thinking-handoff-bedrock.js.map