testronaut 1.0.9 → 1.0.10

package/core/agent.js

@@ -6,10 +6,13 @@ export async function runAgent(goals, missionName, maxTurns = 20, ) {
   const browser = new ChromeBrowser();
   await browser.start();
   const startTime = Date.now();
+  let result;
 
   try {
+    const missionResults = [];
     for (const goal of goals) {
       // console.log(`\n🚀 Running Agent with goal:\n${goal}\n`);
+      const steps = [];
       const messages = [
         {
           role: 'system',
@@ -29,31 +32,55 @@ export async function runAgent(goals, missionName, maxTurns = 20, ) {
         { role: 'user', content: goal.goal },
       ];
 
-      const success = await turnLoop(browser, messages, maxTurns);
-      if (!success) {
-        console.log('🛑 Agent stopped due to failed goal.\n');
-        const endTime = Date.now();
-        return {
-          missionName: missionName,
-          status: 'failed',
-          steps: [ /* tool steps or key moments */ ],
-          startTime: startTime,
-          endTime: endTime,
+      result = await turnLoop(
+        browser, 
+        messages, 
+        maxTurns, 
+        0, /* currentTurn */
+        0, /* retryCount */
+        null, /* currentStep */
+        {
+          steps,          // 👈 pass in
+          missionName,    // 👈 pass in (for tagging)
         }
-        ;
+      );
+      
+      missionResults.push({
+        missionName,
+        status: result.success ? 'passed' : 'failed',
+        steps: JSON.parse(JSON.stringify(steps)), // 👈 deep-clone to avoid refs
+        startTime: Date.now() - 1, // set real values if you have them
+        endTime: Date.now(),
+      });
+
+      if (!result.success) {
+        console.log('🛑 Agent stopped due to failed goal.\n');
+        return missionResults;
       }
+
+      // if (!result.success) {
+      //   console.log('🛑 Agent stopped due to failed goal.\n');
+      //   const endTime = Date.now();
+      //   return {
+      //     missionName: missionName,
+      //     status: 'failed',
+      //     steps: result.steps,
+      //     startTime: startTime,
+      //     endTime: endTime,
+      //   };
+      // }
     }
 
     console.log('✅ All goals completed successfully.\n');
-    const endTime = Date.now();
-    return {
-      missionName: missionName,
-      status: 'passed',
-      steps: [ /* tool steps or key moments */ ],
-      startTime: startTime,
-      endTime: endTime,
-    }
-    ;
+    // const endTime = Date.now();
+    return missionResults;
+    // return {
+    //   missionName: missionName,
+    //   status: 'passed',
+    //   steps: result.steps,
+    //   startTime: startTime,
+    //   endTime: endTime,
+    // };
 
   } finally {
     await browser.close();

package/core/turnLoop.js

@@ -19,6 +19,7 @@ const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
 let totalTokensUsed = 0;
 let turnTimestamps = [];
 let shouldBackoff;
+let steps = [];
 
 const pushDOMAssistant = async (browser, messages, agentMemory, { skipIfLastTool } = {}) => {
   if (skipIfLastTool && skipIfLastTool.includes(messages.at(-1)?.name)) {
@@ -55,7 +56,16 @@ const pushDOMAssistant = async (browser, messages, agentMemory, { skipIfLastTool
   });
 }
 
-export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, retryCount = 0) => {
+export const turnLoop = async (
+  browser, 
+  messages, 
+  maxTurns, 
+  currentTurn = 0, 
+  retryCount = 0, 
+  currentStep = {},
+  ctx = {} // 👈 { steps, missionName }
+) => {
+  const { steps = [], missionName } = ctx;
   let agentMemory = {
     lastMenuExpanded: false
   };
@@ -63,6 +73,8 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
   for (let turn = currentTurn; turn < maxTurns; turn++) {
     console.log(`\n🔄 Turn ${turn + 1}/${maxTurns}`);
     let response;
+    // const currentStep = { turn, events: [], result: '🟡 In Progress' };
+    const currentStep = { turn, events: [], result: '🟡 In Progress', missionName }; // 👈 tag it
     try {
 
       // Recalculate the rolling window before checking cooldown
@@ -70,12 +82,15 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
 
       // Cool off check
       ({ totalTokensUsed, turnTimestamps, shouldBackoff } = await tokenUseCoolOff(totalTokensUsed, turnTimestamps));
-      if(shouldBackoff) return await turnLoop(browser, messages, maxTurns, turn);
+      if(shouldBackoff) return await turnLoop(browser, messages, maxTurns, turn, currentStep);
 
       //Tool Validator
       if (!validateAndInsertMissingToolResponses(messages, { insertPlaceholders: true })) {
         console.error('❌ Tool call structure invalid and no placeholders inserted.');
-        return false;
+        currentStep.events.push('❌ Tool call structure invalid and no placeholders inserted.')
+        currentStep.result = '❌ Failure';
+        steps.push(currentStep);
+        return steps;
       }
 
       const unrespondedCalls = messages.filter(
@@ -92,7 +107,10 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
       if (unrespondedCalls.length) {
         console.error('🛑 Detected assistant tool calls without matching tool responses:');
         console.dir(unrespondedCalls, { depth: 5 });
-        return false;
+        currentStep.events.push('🛑 Detected assistant tool calls without matching tool responses')
+        currentStep.result = '❌ Failure';
+        steps.push(currentStep);
+        return steps;
       }
       
 
@@ -109,15 +127,21 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
 
         if (retryCount >= 5) {
           console.error('❌ Too many retries. Exiting.');
-          return false;
+          currentStep.events.push('❌ Too many retries. Exiting.')
+          currentStep.result = '❌ Failure';
+          steps.push(currentStep);
+          return steps;
         }
 
-        return await turnLoop(browser, messages, maxTurns, turn, retryCount + 1);
+        return await turnLoop(browser, messages, maxTurns, turn, retryCount + 1, currentStep);
       } else if (err.status === 400) {
         console.error('❌ Bad request:', err.message);
         // console.log("current tools", toolsSchema);
         // console.log("messages: ", messages)
-        return false;
+        currentStep.events.push(`❌ Bad request: ${err.message}`)
+        currentStep.result = '❌ Failure';
+        steps.push(currentStep);
+        return steps;
       } else {
         throw err;
       }
@@ -127,10 +151,12 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
     if (usage) {
       const tokensUsed = usage.total_tokens || 0;
       console.log(`📊 Token Usage This Turn → Total: ${tokensUsed}`);
+      currentStep.tokensUsed = tokensUsed;
       recordTokenUsage(turnTimestamps, tokensUsed);
       ({ turnTimestamps, totalTokensUsed } = pruneOldTokenUsage(turnTimestamps));
 
       console.log(`📈 Running Total Tokens Used (Rolling 60s): ${totalTokensUsed}`);
+      currentStep.totalTokensUsed = totalTokensUsed;
     }
     
     const msg = response.choices[0].message;
@@ -145,6 +171,7 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
         lastToolName = fnName;
         const args = JSON.parse(call.function.arguments || '{}');
         console.log(`[model] → ${fnName}`, args);
+        currentStep.events.push(`[model] → ${fnName} ${args}`);
         // console.log('Calling tool:', fnName);
         let result;
         let errorMessage = null;
@@ -161,9 +188,19 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
         }
 
         console.log(`[tool ] ← ${fnName} result:`, errorMessage ? '❌ Failed' : '✅ Success');
+        currentStep.events.push(`[tool ] ← ${fnName} result: ${errorMessage ? '❌ Failed' : '✅ Success'}`)
         const truncated = result.length > 100 ? result.slice(0, 100) + '…' : result;
         console.log(`[tool ] ← ${truncated}`);
 
+        if (fnName === 'screenshot') {
+          const match = result.match(/screenshot.*?saved at: (.+\.png)/i);
+          if (match && match[1]) {
+            currentStep.screenshotPath = match[1];
+            currentStep.events.push(`🖼️ Screenshot captured: ${match[1]}`);
+          }
+        }
+        
+        currentStep.events.push(`[tool ] ← ${truncated}`)
         toolResponses.push({
           role: 'tool',
           tool_call_id: call.id,
@@ -178,31 +215,41 @@ export const turnLoop = async (browser, messages, maxTurns, currentTurn = 0, ret
         
         if (fnName === 'click_text' || fnName === 'click' || fnName === 'expand_menu') {
           console.log(`[auto] → Injecting DOM after menu expand...`);
+          currentStep.events.push(`[auto] → Injecting DOM after menu expand...`);
           const domHtml = await CHROME_TOOL_MAP.get_dom(browser, {
             limit: 100000,
             exclude: true,
             focus: [],
           }, agentMemory);
           await tokenEstimate('gpt-4o', domHtml);
-          // fs.writeFileSync(`missions/mission_reports/debug-expanded-${Date.now()}.html`, domHtml);
           await pushDOMAssistant(browser, messages, agentMemory, {
             skipIfLastTool: ['get_dom', 'check_text']
           });
           console.log(`[auto] → DOM size after ${fnName}: ${domHtml.length} chars`);
+          currentStep.events.push(`[auto] → DOM size after ${fnName}: ${domHtml.length} chars`)
         }
         
       }
       messages.push(msg, ...toolResponses);
+      currentStep.result = '✅ Passed';
+      steps.push(currentStep);
       continue;
     }
 
-    const isAgentDone = finalResponseHandler(msg)
-    if (isAgentDone !== null) return isAgentDone;
+    const finalResponse = finalResponseHandler(msg);
+    if (finalResponse !== null) {
+      currentStep.events.push(finalResponse.finalMessage);
+      currentStep.result = finalResponse.success ? '✅ Mission Success': '❌ Mission Failure';
+      steps.push(currentStep);
+      return { success: finalResponse.finalMessage, steps };
+    }
+
 
     await pushDOMAssistant(browser, messages, agentMemory,{
       skipIfLastTool: ['get_dom', 'check_text']
     });
 
     console.log(`[auto] → Injected DOM for next reasoning step`);
+    currentStep.events.push(`[auto] → Injected DOM for next reasoning step`)
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "testronaut",
-  "version": "1.0.9",
+  "version": "1.0.10",
   "main": "./runner/testronaut.js",
   "bin": {
     "testronaut": "./bin/cli.js"

package/runner/testronaut.js

@@ -47,6 +47,15 @@ export async function runMissions({ preMission, mission, postMission }, missionN
     console.log(`❌ Aborting after failed goal.`);
     return;
   }
+  // console.log("success[0]: ", success[0]);
+  // console.log("success[0].steps.length: ", success[0].steps.length);
+  // console.log("success[0].steps[success[0].steps.length - 1].result: ", success[0].steps[success[0].steps.length - 1].result)
+  const missionStatus = success[0].steps[success[0].steps.length - 1].result;
+  // console.log("missionStatus: ", missionStatus)
+  if (missionStatus.toLowerCase().includes('failure')) {
+    // console.log("we detected the failure!")
+    success[0].status = 'failed';
+  }
   console.log('\n✅ Mission flow complete.');
   return success;
 }

package/tools/chromeBrowser.js

@@ -90,9 +90,7 @@ export class ChromeBrowser {
       if (innerText.includes(text)) {
         await el.scrollIntoViewIfNeeded().catch(() => {});
         await el.click();
-        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` });
         return `clicked text "${text}"`;
         
       }
@@ -119,7 +117,6 @@ 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: `missions/mission_reports/screenshot_after_click_${Date.now()}.png` });
         return `clicked visible text "${text}"`;
       }
     }
@@ -147,11 +144,18 @@ export class ChromeBrowser {
     await this.page.waitForSelector(selector, { timeout: 30_000 });
     await this.page.click(selector);
     await this.page.waitForTimeout(delayMs);
-    await this.page.screenshot({ path: `missions/mission_reports/screenshot_expand_menu_${Date.now()}.png` });
-
     return `expanded menu using ${selector}`;
   }
 
+  async screenshot({ name = 'screenshot' }) {
+    const timestamp = new Date().toISOString().replace(/[:.]/g, '-');
+    const screenshotPath = `missions/mission_reports/screenshots/${name}_${timestamp}.png`;
+  
+    await this.page.screenshot({ path: screenshotPath });
+  
+    // ✅ Return relative path for HTML report
+    return `Screenshot saved at: ./screenshots/${name}_${timestamp}.png`;
+  }
   
   
   async close() {
@@ -173,7 +177,6 @@ export const CHROME_TOOL_MAP = {
       return await browser.click_text(args);
     }
   },
-  
   expand_menu: (b, args) => b.expand_menu(args),
-  // click_text: (b, args) => b.click_text(args),
+  screenshot: (b, args) => b.screenshot(args),
 };

package/tools/generateHtmlReport.js

@@ -1,52 +1,122 @@
 import fs from 'fs';
+import path from 'path';
 
 export function generateHtmlReport(report, outputPath) {
-  const { runId, startTime, endTime, summary, missions } = report;
+  const { runId, startTime, endTime, missions = [], summary = {} } = report;
+  const durationSec = ((new Date(endTime) - new Date(startTime)) / 1000).toFixed(2);
+
+  // tiny helpers
+  const esc = (s) => String(s ?? '')
+    .replace(/&/g, '&amp;').replace(/</g, '&lt;')
+    .replace(/>/g, '&gt;');
+  const badge = (status) =>
+    status === 'passed' ? '✅ Passed' :
+    status === 'failed' ? '❌ Failed' : (status || '—');
+
+  const missionBlock = (m, idx) => {
+    const mDurationSec = m.endTime && m.startTime
+      ? ((m.endTime - m.startTime) / 1000).toFixed(2)
+      : '—';
 
-  const missionHtml = missions.map(m => {
-    const status = m.status?.toUpperCase?.() || 'UNKNOWN';
-    const missionName = m.missionName || 'Unnamed';
-    const file = m.file || 'N/A';
     const steps = Array.isArray(m.steps) ? m.steps : [];
+    const stepItems = steps.map(step => {
+      const events = Array.isArray(step.events) ? step.events : [];
+      const imgTag = step.screenshotPath
+        ? `<img src="${esc(step.screenshotPath)}" alt="screenshot turn ${esc(step.turn)}">`
+        : '';
+      return `
+        <details class="step" ${/Mission Success|Passed/.test(step.result||'') ? '' : 'open'}>
+          <summary>
+            <span class="turn">Turn ${esc((step.turn ?? 0) + 1)}</span>
+            <span class="step-result ${/✅/.test(step.result||'') ? 'ok' : 'bad'}">${esc(step.result || '')}</span>
+            <span class="tokens">tokens: ${esc(step.tokensUsed ?? '—')} / total: ${esc(step.totalTokensUsed ?? '—')}</span>
+          </summary>
+          <pre class="events">${esc(events.join('\n')) || '(no events)'}</pre>
+          ${imgTag}
+        </details>
+      `;
+    }).join('');
 
     return `
-      <div class="mission ${status.toLowerCase()}">
-        <h3>${file} - ${missionName} - <span>${status}</span></h3>
-        <p><em>${m.startTime || 'N/A'} → ${m.endTime || 'N/A'}</em></p>
-        <ul>
-          ${steps.map(step => `
-            <li>
-              <strong>${step.tool}</strong> → ${step.output}
-            </li>
-          `).join('')}
-        </ul>
-      </div>
+      <details class="mission">
+        <summary>
+          <span class="name">${esc(m.missionName || path.basename(m.file || `mission_${idx+1}`))}</span>
+          <span class="status ${m.status === 'passed' ? 'ok' : (m.status === 'failed' ? 'bad' : '')}">${badge(m.status)}</span>
+          <span class="meta">steps: ${steps.length} • duration: ${mDurationSec}s • file: ${esc(m.file || '')}</span>
+        </summary>
+        <div class="steps">
+          ${stepItems || '<div class="empty">No steps recorded.</div>'}
+        </div>
+      </details>
     `;
-  }).join('');
+  };
 
   const html = `
-    <html>
-    <head>
-      <style>
-        body { font-family: sans-serif; padding: 20px; }
-        .passed { background-color: #e6ffed; }
-        .failed { background-color: #ffe6e6; }
-        .unknown { background-color: #f9f9f9; }
-        h3 span { font-weight: bold; }
-        ul { margin-top: 0; }
-      </style>
-    </head>
-    <body>
-      <h1>🧑‍🚀 Testronaut Report</h1>
-      <p><strong>Run ID:</strong> ${runId}</p>
-      <p><strong>Start:</strong> ${startTime}</p>
-      <p><strong>End:</strong> ${endTime}</p>
-      <p><strong>Summary:</strong> ${summary.passed}/${summary.totalMissions} passed</p>
-      ${missionHtml}
-    </body>
-    </html>
+<!DOCTYPE html>
+<html>
+<head>
+  <meta charset="UTF-8" />
+  <title>Testronaut Report – ${esc(runId)}</title>
+  <style>
+    :root {
+      --ok: #16a34a;
+      --bad: #dc2626;
+      --muted: #6b7280;
+      --bg: #f8fafc;
+      --card: #ffffff;
+      --border: #e5e7eb;
+      --mono: ui-monospace,SFMono-Regular,Menlo,Monaco,Consolas,"Liberation Mono","Courier New",monospace;
+    }
+    body { margin:0; padding:24px; font-family: ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,Ubuntu; background: var(--bg); color:#111827; }
+    h1 { margin: 0 0 6px 0; font-size: 20px; }
+    .run-meta { color: var(--muted); margin-bottom: 16px; }
+    .summary { display:flex; gap:12px; margin: 12px 0 20px; }
+    .pill { border:1px solid var(--border); background: var(--card); padding:6px 10px; border-radius:999px; font-size:12px; }
+    .pill.ok { border-color: #bbf7d0; background:#ecfdf5; color: var(--ok); }
+    .pill.bad { border-color: #fecaca; background:#fef2f2; color: var(--bad); }
+
+    details { background: var(--card); border:1px solid var(--border); border-radius:10px; margin:10px 0; }
+    summary { cursor:pointer; padding:12px 14px; display:flex; align-items:center; gap:12px; }
+    .mission > summary { font-weight:600; }
+    .name { flex:1; }
+    .status.ok { color: var(--ok); }
+    .status.bad { color: var(--bad); }
+    .meta { color: var(--muted); font-size:12px; }
+
+    .steps { padding: 0 14px 12px; }
+    .step summary { background: #fafafa; border-bottom:1px solid var(--border); border-radius: 10px 10px 0 0; }
+    .turn { font-weight:600; }
+    .step-result.ok { color: var(--ok); }
+    .step-result.bad { color: var(--bad); }
+    .tokens { color: var(--muted); font-size:12px; margin-left:auto; }
+    .events { background:#0b1020; color:#e5e7eb; padding:12px; border-radius:8px; margin:12px 0 0; font: 12px/1.4 var(--mono); max-height: 280px; overflow:auto; }
+    img { display:block; max-width:100%; border:1px solid var(--border); border-radius:8px; margin-top:10px; }
+    .empty { color: var(--muted); padding: 12px; }
+  </style>
+</head>
+<body>
+  <h1>🧑‍🚀 Testronaut Report</h1>
+  <div class="run-meta">
+    <div><strong>Run ID:</strong> ${esc(runId)}</div>
+    <div><strong>Start:</strong> ${esc(startTime)}</div>
+    <div><strong>End:</strong> ${esc(endTime)} • <strong>Duration:</strong> ${durationSec}s</div>
+  </div>
+
+  <div class="summary">
+    <div class="pill">Missions: ${esc(summary.totalMissions ?? missions.length)}</div>
+    <div class="pill ok">Passed: ${esc(summary.passed ?? missions.filter(m=>m.status==='passed').length)}</div>
+    <div class="pill bad">Failed: ${esc(summary.failed ?? missions.filter(m=>m.status==='failed').length)}</div>
+  </div>
+
+  ${missions.map(missionBlock).join('')}
+
+</body>
+</html>
   `;
 
-  fs.writeFileSync(outputPath, html);
-  console.log(`📝 Report generated at: ${outputPath}`);
+  const out = outputPath ?? path.resolve('missions/mission_reports', `${runId}.html`);
+  fs.mkdirSync(path.dirname(out), { recursive: true });
+  fs.writeFileSync(out, html);
+  console.log(`📝 Report generated at: ${out}`);
+  return out;
 }

package/tools/toolSchema.js

@@ -155,7 +155,25 @@ const toolsSchema = [
         required: ['chunkIndex', 'totalChunks']
       }
     }
+  },
+  {
+    type: 'function',
+    function: {
+      name: 'screenshot',
+      description: 'Takes a screenshot of the current page',
+      parameters: {
+        type: 'object',
+        properties: {
+          label: {
+            type: 'string',
+            description: 'Optional label for the screenshot filename',
+          },
+        },
+        required: [],
+      },
+    },
   }
+  
     
 ];
 

package/tools/turnLoopUtils.js

@@ -4,13 +4,13 @@ export const finalResponseHandler = (msg) => {
     console.log('\n┏━ FINAL AGENT RESPONSE ━━━━━━━━━━━━━━━━━━━');
     console.log(msg.content);
     console.log('┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
-    return true;
+    return { finalMessage: msg.content, success: true};
   }
   if (final?.startsWith('failure')) {
     console.log('\n┏━ FINAL AGENT RESPONSE ━━━━━━━━━━━━━━━━━━━');
     console.log(msg.content);
     console.log('┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
-    return false;
+    return { finalMessage: msg.content, success: false};
   }
   return null;
 }