@illuma-ai/agents 1.0.81

package/src/scripts/test_code_api.ts

@@ -0,0 +1,361 @@
+// src/scripts/test_code_api.ts
+/**
+ * Direct test of the Code API to verify session file persistence.
+ * This bypasses the LLM and tests the API directly.
+ *
+ * Run with: npx ts-node -r dotenv/config src/scripts/test_code_api.ts
+ */
+import { config } from 'dotenv';
+config();
+
+import fetch, { RequestInit } from 'node-fetch';
+import { HttpsProxyAgent } from 'https-proxy-agent';
+
+const API_KEY = process.env.CODE_EXECUTOR_API_KEY ?? '';
+const BASE_URL =
+  process.env.ILLUMA_CODE_BASEURL ?? 'https://api.illuma.ai/v1';
+const PROXY = process.env.PROXY;
+
+if (!API_KEY) {
+  console.error('CODE_EXECUTOR_API_KEY not set');
+  process.exit(1);
+}
+
+interface FileRef {
+  id: string;
+  name: string;
+  session_id?: string;
+  /** Lineage tracking - present if file was modified from previous session */
+  modified_from?: {
+    id: string;
+    session_id: string;
+  };
+}
+
+interface ExecResult {
+  session_id: string;
+  stdout: string;
+  stderr: string;
+  files?: FileRef[];
+}
+
+interface FileInfo {
+  name: string;
+  metadata: Record<string, string>;
+}
+
+async function makeRequest(
+  endpoint: string,
+  body?: Record<string, unknown>
+): Promise<unknown> {
+  const fetchOptions: RequestInit = {
+    method: body ? 'POST' : 'GET',
+    headers: {
+      'Content-Type': 'application/json',
+      'User-Agent': 'Illuma/1.0',
+      'X-API-Key': API_KEY,
+    },
+  };
+
+  if (body) {
+    fetchOptions.body = JSON.stringify(body);
+  }
+
+  if (PROXY) {
+    fetchOptions.agent = new HttpsProxyAgent(PROXY);
+  }
+
+  console.log(`\n>>> ${body ? 'POST' : 'GET'} ${endpoint}`);
+  if (body) {
+    console.log('Body:', JSON.stringify(body, null, 2));
+  }
+
+  const response = await fetch(endpoint, fetchOptions);
+  const result = await response.json();
+
+  console.log(`<<< Response (${response.status}):`);
+  console.log(JSON.stringify(result, null, 2));
+
+  if (!response.ok) {
+    throw new Error(`HTTP ${response.status}: ${JSON.stringify(result)}`);
+  }
+
+  return result;
+}
+
+async function testCodeAPI(): Promise<void> {
+  console.log('='.repeat(60));
+  console.log('TEST 1: Create a file');
+  console.log('='.repeat(60));
+
+  const createCode = `
+import json
+
+config = {
+    "app_name": "TestApp",
+    "version": "1.0.0",
+    "debug": True
+}
+
+with open("/mnt/data/test_config.json", "w") as f:
+    json.dump(config, f, indent=2)
+
+with open("/mnt/data/test_config.json", "r") as f:
+    print(f.read())
+`;
+
+  const result1 = (await makeRequest(`${BASE_URL}/exec`, {
+    lang: 'py',
+    code: createCode,
+  })) as ExecResult;
+
+  const sessionId = result1.session_id;
+  const files = result1.files ?? [];
+
+  console.log('\n--- Result Summary ---');
+  console.log('session_id:', sessionId);
+  console.log('files:', files);
+  console.log('stdout:', result1.stdout);
+  console.log('stderr:', result1.stderr);
+
+  if (!sessionId || files.length === 0) {
+    console.error('\n❌ No session_id or files returned!');
+    return;
+  }
+
+  // Check if files now include session_id (new API feature)
+  const hasSessionIdInFiles = files.some((f) => f.session_id != null);
+  console.log('\n✅ Files include session_id:', hasSessionIdInFiles);
+
+  console.log('\n' + '='.repeat(60));
+  console.log(
+    'TEST 2: Fetch files IMMEDIATELY (no delay - testing race condition fix)'
+  );
+  console.log('='.repeat(60));
+
+  const filesResult = (await makeRequest(
+    `${BASE_URL}/files/${sessionId}?detail=full`
+  )) as FileInfo[];
+
+  console.log('\n--- Files in session (detail=full) ---');
+  for (const file of filesResult) {
+    console.log('File:', file.name);
+    console.log('  metadata:', file.metadata);
+  }
+
+  if (filesResult.length === 0) {
+    console.log(
+      '\n⚠️  Files endpoint returned empty - race condition may still exist'
+    );
+  } else {
+    console.log('\n✅ Files available immediately!');
+  }
+
+  // Test new normalized detail level
+  console.log('\n' + '='.repeat(60));
+  console.log('TEST 2b: Fetch files with detail=normalized');
+  console.log('='.repeat(60));
+
+  const normalizedResult = (await makeRequest(
+    `${BASE_URL}/files/${sessionId}?detail=normalized`
+  )) as FileRef[];
+
+  console.log('\n--- Files in session (detail=normalized) ---');
+  console.log(JSON.stringify(normalizedResult, null, 2));
+
+  console.log('\n' + '='.repeat(60));
+  console.log(
+    'TEST 3: Read file IMMEDIATELY using files from original response'
+  );
+  console.log('='.repeat(60));
+
+  // Use files directly - if API returns session_id, use that; otherwise add it
+  const fileReferences: FileRef[] = files.map((file) => ({
+    session_id: file.session_id ?? sessionId,
+    id: file.id,
+    name: file.name,
+  }));
+
+  console.log(
+    '\nFile references we will send:',
+    JSON.stringify(fileReferences, null, 2)
+  );
+
+  const readCode = `
+import json
+
+with open("/mnt/data/test_config.json", "r") as f:
+    config = json.load(f)
+    print("Read config:")
+    print(json.dumps(config, indent=2))
+    print("Version:", config.get("version"))
+`;
+
+  const result2 = (await makeRequest(`${BASE_URL}/exec`, {
+    lang: 'py',
+    code: readCode,
+    files: fileReferences,
+  })) as ExecResult;
+
+  console.log('\n--- Result Summary ---');
+  console.log('stdout:', result2.stdout);
+  console.log('stderr:', result2.stderr);
+
+  if (result2.stderr && result2.stderr.includes('FileNotFoundError')) {
+    console.log(
+      '\n❌ File not found! The file reference format might be wrong.'
+    );
+
+    // Try alternative format - just session_id
+    console.log('\n' + '='.repeat(60));
+    console.log('TEST 4: Try with just session_id in request');
+    console.log('='.repeat(60));
+
+    const result3 = (await makeRequest(`${BASE_URL}/exec`, {
+      lang: 'py',
+      code: readCode,
+      session_id: sessionId,
+    })) as ExecResult;
+
+    console.log('\n--- Result Summary ---');
+    console.log('stdout:', result3.stdout);
+    console.log('stderr:', result3.stderr);
+  } else {
+    console.log('\n✅ File read successfully!');
+  }
+
+  // ============================================================
+  // TEST 4: MODIFY the file (same filename) - tests editable files
+  // ============================================================
+  console.log('\n' + '='.repeat(60));
+  console.log('TEST 4: MODIFY file in-place (testing editable files feature)');
+  console.log('='.repeat(60));
+
+  const modifyCode = `
+import json
+
+# Read the existing file
+with open("/mnt/data/test_config.json", "r") as f:
+    config = json.load(f)
+
+print("Original config:")
+print(json.dumps(config, indent=2))
+
+# Modify the config
+config["version"] = "2.0.0"
+config["modified"] = True
+
+# Write BACK to the SAME filename (should work now!)
+with open("/mnt/data/test_config.json", "w") as f:
+    json.dump(config, f, indent=2)
+
+# Verify the write
+with open("/mnt/data/test_config.json", "r") as f:
+    updated = json.load(f)
+
+print("\\nUpdated config:")
+print(json.dumps(updated, indent=2))
+`;
+
+  const result3 = (await makeRequest(`${BASE_URL}/exec`, {
+    lang: 'py',
+    code: modifyCode,
+    files: fileReferences,
+  })) as ExecResult;
+
+  console.log('\n--- Result Summary ---');
+  console.log('stdout:', result3.stdout);
+  console.log('stderr:', result3.stderr);
+  console.log('files:', JSON.stringify(result3.files, null, 2));
+
+  if (result3.stderr && result3.stderr.includes('Permission denied')) {
+    console.log('\n❌ Permission denied - files are still read-only!');
+  } else if (result3.stderr && result3.stderr.includes('Error')) {
+    console.log('\n❌ Error modifying file:', result3.stderr);
+  } else {
+    console.log('\n✅ File modified successfully!');
+
+    // Check for modified_from lineage
+    const modifiedFile = result3.files?.find(
+      (f) => f.name === 'test_config.json'
+    );
+    if (modifiedFile) {
+      console.log('\n--- Modified File Details ---');
+      console.log('  id:', modifiedFile.id);
+      console.log('  name:', modifiedFile.name);
+      console.log('  session_id:', modifiedFile.session_id);
+      if (modifiedFile.modified_from) {
+        console.log(
+          '  modified_from:',
+          JSON.stringify(modifiedFile.modified_from)
+        );
+        console.log(
+          '\n✅ Lineage tracking working! File shows it was modified from previous session.'
+        );
+      } else {
+        console.log(
+          '\n⚠️  No modified_from field - lineage tracking not present'
+        );
+      }
+    } else {
+      console.log('\n⚠️  Modified file not found in response files array');
+    }
+  }
+
+  // ============================================================
+  // TEST 5: Verify modification persists in next execution
+  // ============================================================
+  console.log('\n' + '='.repeat(60));
+  console.log(
+    'TEST 5: Verify modified file can be read in subsequent execution'
+  );
+  console.log('='.repeat(60));
+
+  // Use the new file references from the modify response
+  const newFileRefs: FileRef[] = (result3.files ?? []).map((file) => ({
+    session_id: file.session_id ?? result3.session_id,
+    id: file.id,
+    name: file.name,
+  }));
+
+  if (newFileRefs.length === 0) {
+    console.log(
+      '\n⚠️  No files returned from modification, skipping verification'
+    );
+  } else {
+    console.log(
+      '\nUsing new file references:',
+      JSON.stringify(newFileRefs, null, 2)
+    );
+
+    const verifyCode = `
+import json
+
+with open("/mnt/data/test_config.json", "r") as f:
+    config = json.load(f)
+
+print("Verified config:")
+print(json.dumps(config, indent=2))
+
+if config.get("version") == "2.0.0" and config.get("modified") == True:
+    print("\\n✅ Modification persisted correctly!")
+else:
+    print("\\n❌ Modification did NOT persist!")
+`;
+
+    const result4 = (await makeRequest(`${BASE_URL}/exec`, {
+      lang: 'py',
+      code: verifyCode,
+      files: newFileRefs,
+    })) as ExecResult;
+
+    console.log('\n--- Result Summary ---');
+    console.log('stdout:', result4.stdout);
+    console.log('stderr:', result4.stderr);
+  }
+}
+
+testCodeAPI().catch((err) => {
+  console.error('Error:', err);
+  process.exit(1);
+});

package/src/scripts/thinking-bedrock.ts

@@ -0,0 +1,159 @@
+// src/scripts/thinking-bedrock.ts
+import { config } from 'dotenv';
+config();
+import { HumanMessage, BaseMessage } from '@langchain/core/messages';
+import type { UsageMetadata } from '@langchain/core/messages';
+import * as t from '@/types';
+import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
+import { createCodeExecutionTool } from '@/tools/CodeExecutor';
+import { ToolEndHandler, ModelEndHandler } from '@/events';
+import { GraphEvents, Providers } from '@/common';
+import { getLLMConfig } from '@/utils/llmConfig';
+import { getArgs } from '@/scripts/args';
+import { Run } from '@/run';
+
+const conversationHistory: BaseMessage[] = [];
+let _contentParts: t.MessageContentComplex[] = [];
+const collectedUsage: UsageMetadata[] = [];
+
+async function testBedrockThinking(): Promise<void> {
+  const { userName } = await getArgs();
+  const instructions = `You are a helpful AI assistant for ${userName}. When answering questions, be thorough in your reasoning.`;
+  const { contentParts, aggregateContent } = createContentAggregator();
+  _contentParts = contentParts as t.MessageContentComplex[];
+
+  // Set up event handlers
+  const customHandlers = {
+    [GraphEvents.TOOL_END]: new ToolEndHandler(),
+    [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(collectedUsage),
+    [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
+    [GraphEvents.ON_RUN_STEP_COMPLETED]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP_COMPLETED,
+        data: t.StreamEventData
+      ): void => {
+        console.log('====== ON_RUN_STEP_COMPLETED ======');
+        aggregateContent({
+          event,
+          data: data as unknown as { result: t.ToolEndEvent },
+        });
+      },
+    },
+    [GraphEvents.ON_RUN_STEP]: {
+      handle: (event: GraphEvents.ON_RUN_STEP, data: t.RunStep) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_RUN_STEP_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP_DELTA,
+        data: t.RunStepDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_MESSAGE_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_MESSAGE_DELTA,
+        data: t.MessageDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_REASONING_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_REASONING_DELTA,
+        data: t.ReasoningDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+  };
+
+  const baseLlmConfig = getLLMConfig(Providers.BEDROCK);
+
+  // Enable thinking with token budget for Bedrock
+  const llmConfig = {
+    ...baseLlmConfig,
+    model: 'us.anthropic.claude-3-7-sonnet-20250219-v1:0',
+    maxTokens: 5000,
+    additionalModelRequestFields: {
+      thinking: { type: 'enabled', budget_tokens: 2000 },
+    },
+  };
+
+  const run = await Run.create<t.IState>({
+    runId: 'test-bedrock-thinking-id',
+    graphConfig: {
+      instructions,
+      type: 'standard',
+      tools: [createCodeExecutionTool()],
+      llmConfig,
+    },
+    returnContent: true,
+    customHandlers: customHandlers as t.RunConfig['customHandlers'],
+  });
+
+  const config = {
+    configurable: {
+      thread_id: 'bedrock-thinking-test-thread',
+    },
+    streamMode: 'values',
+    version: 'v2' as const,
+  };
+
+  // Test 1: Regular thinking mode
+  console.log('\n\nTest 1: Bedrock Regular thinking mode');
+  const userMessage1 = `Please print 'hello world' in python`;
+  conversationHistory.push(new HumanMessage(userMessage1));
+
+  console.log('Running first query with Bedrock thinking enabled...');
+  const firstInputs = { messages: [...conversationHistory] };
+  await run.processStream(firstInputs, config);
+
+  // Extract and display thinking blocks
+  const finalMessages = run.getRunMessages();
+  console.log('\n\nFinal messages after Test 1:');
+  console.dir(finalMessages, { depth: null });
+
+  // Test 2: Try multi-turn conversation
+  console.log(
+    '\n\nTest 2: Multi-turn conversation with Bedrock thinking enabled'
+  );
+  const userMessage2 = `Given your previous analysis, what would be the most significant technical challenges in making this transition?`;
+  conversationHistory.push(new HumanMessage(userMessage2));
+
+  console.log('Running second query with Bedrock thinking enabled...');
+  const secondInputs = { messages: [...conversationHistory] };
+  await run.processStream(secondInputs, config);
+
+  // Display thinking blocks for second response
+  const finalMessages2 = run.getRunMessages();
+  console.log('\n\nBedrock thinking feature test completed!');
+  console.dir(finalMessages2, { depth: null });
+
+  console.log('\n\nContent parts:');
+  console.dir(_contentParts, { depth: null });
+}
+
+process.on('unhandledRejection', (reason, promise) => {
+  console.error('Unhandled Rejection at:', promise, 'reason:', reason);
+  console.log('Conversation history:');
+  console.dir(conversationHistory, { depth: null });
+  console.log('Content parts:');
+  console.dir(_contentParts, { depth: null });
+  process.exit(1);
+});
+
+process.on('uncaughtException', (err) => {
+  console.error('Uncaught Exception:', err);
+});
+
+testBedrockThinking().catch((err) => {
+  console.error(err);
+  console.log('Conversation history:');
+  console.dir(conversationHistory, { depth: null });
+  console.log('Content parts:');
+  console.dir(_contentParts, { depth: null });
+  process.exit(1);
+});

package/src/scripts/thinking.ts

@@ -0,0 +1,171 @@
+// src/scripts/test-thinking.ts
+import { config } from 'dotenv';
+config();
+import {
+  HumanMessage,
+  SystemMessage,
+  BaseMessage,
+} from '@langchain/core/messages';
+import type { UsageMetadata } from '@langchain/core/messages';
+import * as t from '@/types';
+import { ChatModelStreamHandler, createContentAggregator } from '@/stream';
+import { createCodeExecutionTool } from '@/tools/CodeExecutor';
+import { ToolEndHandler, ModelEndHandler } from '@/events';
+import { GraphEvents, Providers } from '@/common';
+import { getLLMConfig } from '@/utils/llmConfig';
+import { getArgs } from '@/scripts/args';
+import { Run } from '@/run';
+
+const conversationHistory: BaseMessage[] = [];
+let _contentParts: t.MessageContentComplex[] = [];
+const collectedUsage: UsageMetadata[] = [];
+
+async function testThinking(): Promise<void> {
+  const { userName } = await getArgs();
+  const instructions = `You are a helpful AI assistant for ${userName}. When answering questions, be thorough in your reasoning.`;
+  const { contentParts, aggregateContent } = createContentAggregator();
+  _contentParts = contentParts as t.MessageContentComplex[];
+
+  // Set up event handlers
+  const customHandlers = {
+    [GraphEvents.TOOL_END]: new ToolEndHandler(),
+    [GraphEvents.CHAT_MODEL_END]: new ModelEndHandler(collectedUsage),
+    [GraphEvents.CHAT_MODEL_STREAM]: new ChatModelStreamHandler(),
+    [GraphEvents.ON_RUN_STEP_COMPLETED]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP_COMPLETED,
+        data: t.StreamEventData
+      ): void => {
+        console.log('====== ON_RUN_STEP_COMPLETED ======');
+        aggregateContent({
+          event,
+          data: data as unknown as { result: t.ToolEndEvent },
+        });
+      },
+    },
+    [GraphEvents.ON_RUN_STEP]: {
+      handle: (event: GraphEvents.ON_RUN_STEP, data: t.RunStep) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_RUN_STEP_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_RUN_STEP_DELTA,
+        data: t.RunStepDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_MESSAGE_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_MESSAGE_DELTA,
+        data: t.MessageDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+    [GraphEvents.ON_REASONING_DELTA]: {
+      handle: (
+        event: GraphEvents.ON_REASONING_DELTA,
+        data: t.ReasoningDeltaEvent
+      ) => {
+        aggregateContent({ event, data });
+      },
+    },
+  };
+
+  const baseLlmConfig: t.LLMConfig = getLLMConfig(Providers.ANTHROPIC);
+
+  // Enable thinking with token budget
+  const llmConfig = {
+    ...baseLlmConfig,
+    model: 'claude-3-7-sonnet-latest',
+    thinking: { type: 'enabled', budget_tokens: 2000 },
+  };
+
+  const run = await Run.create<t.IState>({
+    runId: 'test-thinking-id',
+    graphConfig: {
+      instructions,
+      type: 'standard',
+      tools: [createCodeExecutionTool()],
+      llmConfig,
+    },
+    returnContent: true,
+    customHandlers: customHandlers as t.RunConfig['customHandlers'],
+  });
+
+  const config = {
+    configurable: {
+      thread_id: 'thinking-test-thread',
+    },
+    streamMode: 'values',
+    version: 'v2' as const,
+  };
+
+  // Test 1: Regular thinking mode
+  console.log('\n\nTest 1: Regular thinking mode');
+  // const userMessage1 = `What would be the environmental and economic impacts if all cars globally were replaced by electric vehicles overnight?`;
+  const userMessage1 = `Please print 'hello world' in python`;
+  conversationHistory.push(new HumanMessage(userMessage1));
+
+  console.log('Running first query with thinking enabled...');
+  const firstInputs = { messages: [...conversationHistory] };
+  await run.processStream(firstInputs, config);
+
+  // Extract and display thinking blocks
+  const finalMessages = run.getRunMessages();
+
+  // Test 2: Try multi-turn conversation
+  console.log('\n\nTest 2: Multi-turn conversation with thinking enabled');
+  const userMessage2 = `Given your previous analysis, what would be the most significant technical challenges in making this transition?`;
+  conversationHistory.push(new HumanMessage(userMessage2));
+
+  console.log('Running second query with thinking enabled...');
+  const secondInputs = { messages: [...conversationHistory] };
+  await run.processStream(secondInputs, config);
+
+  // Display thinking blocks for second response
+  const finalMessages2 = run.getRunMessages();
+
+  // Test 3: Redacted thinking mode
+  console.log('\n\nTest 3: Redacted thinking mode');
+  const magicString =
+    'ANTHROPIC_MAGIC_STRING_TRIGGER_REDACTED_THINKING_46C9A13E193C177646C7398A98432ECCCE4C1253D5E2D82641AC0E52CC2876CB';
+  const userMessage3 = `${magicString}\n\nExplain how quantum computing works in simple terms.`;
+
+  // Reset conversation for clean test
+  conversationHistory.length = 0;
+  conversationHistory.push(new HumanMessage(userMessage3));
+
+  console.log('Running query with redacted thinking...');
+  const thirdInputs = { messages: [...conversationHistory] };
+  await run.processStream(thirdInputs, config);
+
+  // Display redacted thinking blocks
+  const finalMessages3 = run.getRunMessages();
+  console.log('\n\nThinking feature test completed!');
+  console.dir(finalMessages3, { depth: null });
+}
+
+process.on('unhandledRejection', (reason, promise) => {
+  console.error('Unhandled Rejection at:', promise, 'reason:', reason);
+  console.log('Conversation history:');
+  console.dir(conversationHistory, { depth: null });
+  console.log('Content parts:');
+  console.dir(_contentParts, { depth: null });
+  process.exit(1);
+});
+
+process.on('uncaughtException', (err) => {
+  console.error('Uncaught Exception:', err);
+});
+
+testThinking().catch((err) => {
+  console.error(err);
+  console.log('Conversation history:');
+  console.dir(conversationHistory, { depth: null });
+  console.log('Content parts:');
+  console.dir(_contentParts, { depth: null });
+  process.exit(1);
+});