testronaut 1.0.5 → 1.0.7

package/README.md

@@ -20,9 +20,13 @@
 ```
 npm install -g testronaut
 ```
-Or use with npx:
+Then initialize with npx, following the instructions:
 ```
-npx testronaut
+npx testronaut --init
+```
+run the welcome mission:
+```
+npx testronaut welcome.mission.js
 ```
 📁 Directory Structure
 Your project should include a missions/ folder with mission files like:
@@ -98,7 +102,7 @@ testronaut
 Run a specific file:
 
 ```
-testronaut missions/login.mission.js
+testronaut login.mission.js
 ```
 
 🧰 Available Helpers

package/bin/cli.js

@@ -2,12 +2,43 @@
 import fs from 'fs';
 import path from 'path';
 import { fileURLToPath } from 'url';
+import { initializeTestronautProject } from './init.js';
+import { createWelcomeMission } from './createWelcomeMission.js';
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 
 const args = process.argv.slice(2);
 const missionsDir = path.resolve(process.cwd(), 'missions');
 
+const HELP_TEXT = `
+🧑‍🚀 testronaut - Autonomous Agent Mission Runner
+
+Usage:
+  npx testronaut                 Run all missions in the ./missions directory
+  npx testronaut <file>         Run a specific mission file (e.g., login.mission.js)
+
+Options:
+  --init                    Scaffold project folders and a welcome mission
+  --help                    Show this help message
+
+Examples:
+  npx testronaut
+  npx testronaut login.mission.js
+  npx testronaut --init
+`;
+
+
+if (args.includes('--init')) {
+  await initializeTestronautProject();
+  await createWelcomeMission();
+  process.exit(0);
+}
+
+if (args.includes('--help')) {
+  console.log(HELP_TEXT);
+  process.exit(0);
+}
+
 if (!fs.existsSync(missionsDir)) {
   console.error('❌ No `missions` directory found.');
   process.exit(1);

package/bin/createWelcomeMission.js

@@ -0,0 +1,35 @@
+import fs from 'fs';
+import path from 'path';
+
+export const createWelcomeMission = async () => {
+  const missionsDir = path.join(process.cwd(), 'missions');
+  const welcomePath = path.join(missionsDir, 'welcome.mission.js');
+
+  if (fs.existsSync(welcomePath)) {
+    console.log('✅ welcome.mission.js already exists');
+    return;
+  }
+
+  const content = `import { runMissions } from 'testronaut';
+
+const welcomeGoal = \`
+Welcome to Testronaut!
+
+This is your first mission.
+All we ask is that you confirm this setup is working correctly.
+
+To complete this mission, report SUCCESS immediately.
+If you cannot reach this message, report FAILURE.
+\`;
+
+export async function executeMission() {
+  await runMissions({
+    mission: welcomeGoal
+  });
+}
+
+`;
+
+  fs.writeFileSync(welcomePath, content);
+  console.log('🚀 Created missions/welcome.mission.js');
+}

package/bin/init.js

@@ -0,0 +1,77 @@
+import fs from 'fs';
+import path from 'path';
+import prompts from 'prompts';
+import { createWelcomeMission } from './createWelcomeMission.js';
+
+export async function initializeTestronautProject() {
+  const root = process.cwd();
+  const missionsDir = path.join(root, 'missions');
+  const reportsDir = path.join(missionsDir, 'mission_reports');
+  const configPath = path.join(root, 'testronaut-config.json');
+  const envPath = path.join(root, '.env');
+
+  let config = {};
+  if (fs.existsSync(configPath)) {
+    config = JSON.parse(fs.readFileSync(configPath, 'utf-8'));
+  }
+
+  // If not initialized before, walk through setup
+  if (!config.initialized) {
+    // Prompt for LLM
+    const response = await prompts({
+      type: 'select',
+      name: 'llmProvider',
+      message: 'Which LLM provider would you like to use?',
+      choices: [
+        { title: 'OpenAI (GPT-4o)', value: 'openai' },
+        // Add others like Anthropic, Mistral, etc. later
+      ]
+    });
+
+    config.model = response.llmProvider;
+    config.initialized = true;
+    config.outputDir = 'missions/mission_reports';
+    config.projectName = path.basename(root);
+    config.maxTurns = 20;
+    fs.writeFileSync(configPath, JSON.stringify(config, null, 2));
+    console.log('📝 Created testronaut-config.json');
+
+    if (response.llmProvider === 'openai') {
+      // Display .env instructions
+      if (!fs.existsSync(envPath)) {
+        const exampleEnv = `# Add your OpenAI API key below\nOPENAI_API_KEY=sk-...\n`;
+        fs.writeFileSync(envPath, exampleEnv);
+        console.log('🔐 Created .env file');
+      }
+
+      console.log(`
+🔧 Setup Complete!
+
+👉 Please ensure your .env file contains your OpenAI API key like this:
+
+    OPENAI_API_KEY=sk-...
+
+📌 The key must have permissions to call the OpenAI Chat Completions API.
+
+✅ You’re now ready to write missions in \`missions/\` and run your first mission using:
+
+    testronaut welcome.mission.js
+`);
+    }
+
+    createWelcomeMission();
+  } else {
+    console.log('🔁 Already initialized. Skipping setup.');
+  }
+
+  // Create directories if they don't exist
+  if (!fs.existsSync(missionsDir)) {
+    fs.mkdirSync(missionsDir);
+    console.log('📁 Created missions/');
+  }
+
+  if (!fs.existsSync(reportsDir)) {
+    fs.mkdirSync(reportsDir);
+    console.log('📁 Created missions/mission_reports/');
+  }
+}

package/core/turnLoop.js

@@ -43,7 +43,7 @@ const pushDOMAssistant = async (browser, messages, agentMemory, { skipIfLastTool
 
   const domHtml = await CHROME_TOOL_MAP.get_dom(browser, { limit: 100000, exclude: true }, agentMemory);
   await tokenEstimate('gpt-4o', domHtml);
-  fs.writeFileSync(`debug-expanded-${Date.now()}.html`, domHtml);
+  fs.writeFileSync(`missions/mission_reports/debug-expanded-${Date.now()}.html`, domHtml);
 
   messages.push({
     role: 'tool',
@@ -92,6 +92,17 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
 
     let response;
     try {
+      // console.log(
+      //   '🔍 Verifying message structure before OpenAI call:\n',
+      //   messages.map(m => ({
+      //     role: m.role,
+      //     name: m.name,
+      //     tool_call_id: m.tool_call_id,
+      //     tool_calls: m.tool_calls?.map(t => t.id),
+      //     content: m.content?.slice?.(0, 100), // truncate for readability
+      //   }))
+      // );
+      
       response = await openai.chat.completions.create({
         model: 'gpt-4o',
         messages,
@@ -148,6 +159,7 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
         console.log(`[model] → ${fnName}`, args);
         console.log('Calling tool:', fnName);
         let result;
+        let errorMessage = null;
         try {
           console.log("agentMemory: ", agentMemory);
           result = await CHROME_TOOL_MAP[fnName](browser, args, agentMemory);
@@ -165,13 +177,17 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
             result = JSON.stringify(result ?? '');
           }
         } catch (err) {
-          result = `ERROR: ${err.message}`;
+          errorMessage = `ERROR: ${err.message}`;
+          result = errorMessage;
         }
 
+        // 🔍 Diagnostic: log whether this tool call succeeded or failed
+        console.log(`[tool ] ← ${fnName} result:`, errorMessage ? '❌ Failed' : '✅ Success');
+
         const truncated = result.length > 100 ? result.slice(0, 100) + '…' : result;
 
         if (result.length > 100) {
-          const logFile = `tool-result-${Date.now()}.log`;
+          const logFile = `missions/mission_reports/tool-result-${Date.now()}.log`;
           fs.writeFileSync(logFile, result);
           console.log(`[tool ] ← ${truncated} (full output written to ${logFile})`);
         } else {
@@ -197,7 +213,7 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
             focus: [], // or try focusing on ['main'] to keep it light
           }, agentMemory);
           await tokenEstimate('gpt-4o', domHtml);
-          fs.writeFileSync(`debug-expanded-${Date.now()}.html`, domHtml);
+          fs.writeFileSync(`missions/mission_reports/debug-expanded-${Date.now()}.html`, domHtml);
           await pushDOMAssistant(browser, messages, agentMemory, {
             skipIfLastTool: ['get_dom', 'check_text']
           });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "testronaut",
-  "version": "1.0.5",
+  "version": "1.0.7",
   "main": "./runner/testronaut.js",
   "bin": {
     "testronaut": "./bin/cli.js"
@@ -23,6 +23,7 @@
     "cheerio": "^1.1.0",
     "dotenv": "^17.2.0",
     "openai": "^5.10.2",
-    "playwright": "^1.54.1"
+    "playwright": "^1.54.1",
+    "prompts": "^2.4.2"
   }
 }

package/tools/chromeBrowser.js

@@ -90,9 +90,9 @@ export class ChromeBrowser {
       if (innerText.includes(text)) {
         await el.scrollIntoViewIfNeeded().catch(() => {});
         await el.click();
-        await this.page.screenshot({ path: `screenshot_after_click_${Date.now()}.png` });
+        await this.page.screenshot({ path: `missions/mission_reports/screenshot_after_click_${Date.now()}.png` });
         await this.page.waitForTimeout(5000);
-        await this.page.screenshot({ path: `screenshot_after_waiting_click_${Date.now()}.png` });
+        // await this.page.screenshot({ path: `screenshot_after_waiting_click_${Date.now()}.png` });
         return `clicked text "${text}"`;
         
       }
@@ -119,7 +119,7 @@ export class ChromeBrowser {
         console.log(`[click_text] → Clicking visible element with bounding box`, box);
         await el.click();
         await this.page.waitForTimeout(delayMs);
-        await this.page.screenshot({ path: `screenshot_after_click_${Date.now()}.png` });
+        await this.page.screenshot({ path: `missions/mission_reports/screenshot_after_click_${Date.now()}.png` });
         return `clicked visible text "${text}"`;
       }
     }
@@ -149,10 +149,10 @@ export class ChromeBrowser {
     await this.page.waitForTimeout(delayMs);
 
     const expandedDom = await this.page.content();
-    const filename = `expand_menu_dump_${Date.now()}.html`;
+    const filename = `missions/mission_reports/expand_menu_dump_${Date.now()}.html`;
     fs.writeFileSync(filename, expandedDom);
     console.log(`[expand_menu] → Dumped DOM after expand click to ${filename}`);
-    await this.page.screenshot({ path: `screenshot_expand_menu_${Date.now()}.png` });
+    await this.page.screenshot({ path: `missions/mission_reports/screenshot_expand_menu_${Date.now()}.png` });
 
     return `expanded menu using ${selector}`;
   }

package/tools/tokenControl.js

@@ -12,8 +12,8 @@ export const tokenEstimate = async (model, text) => {
 }
 
 export const tokenUseCoolOff = async (totalTokensUsed, turnTimestamps) => {
-  console.log("total tokens used: ", totalTokensUsed);
-  console.log("turnTimestamps: ", turnTimestamps);
+  // console.log("total tokens used: ", totalTokensUsed);
+  // console.log("turnTimestamps: ", turnTimestamps);
   
   if (totalTokensUsed > TOKEN_LIMIT_PER_MIN) {
     const msToWait = await getDynamicBackoffMs(turnTimestamps);

package/tools/turnLoopUtils.js

@@ -34,6 +34,9 @@ export function validateAndInsertMissingToolResponses(messages, options = {}) {
         );
 
         if (!hasResponse) {
+          console.warn(
+            `⚠️ Missing response for tool_call_id: ${expectedId}, function: ${toolCall.function.name}, from assistant message index: ${i}`
+          );
           missingResponses.push(expectedId);
 
           if (insertPlaceholders) {