testronaut 1.9.3 → 1.10.0

package/tests/toolsTests/chromeBrowserProbe.test.js

@@ -0,0 +1,38 @@
+import { describe, expect, it, vi } from 'vitest';
+import { ChromeBrowser } from '../../tools/chromeBrowser.js';
+
+function makeBrowser({ visible = true, content = 'Mission Tasks' } = {}) {
+  const locator = {
+    first: () => locator,
+    waitFor: vi.fn(() => visible ? Promise.resolve() : Promise.reject(new Error('not visible'))),
+    textContent: vi.fn().mockResolvedValue(content),
+  };
+  const browser = Object.create(ChromeBrowser.prototype);
+  browser.page = {
+    goto: vi.fn().mockResolvedValue(undefined),
+    waitForLoadState: vi.fn().mockResolvedValue(undefined),
+    locator: vi.fn(() => locator),
+    url: vi.fn(() => 'https://example.test/dashboard'),
+  };
+  return { browser, locator };
+}
+
+describe('ChromeBrowser.probe', () => {
+  it('passes after navigating when selector and text match', async () => {
+    const { browser } = makeBrowser();
+    await expect(browser.probe({
+      url: 'https://example.test',
+      selector: '#dashboard',
+      text: 'Mission Tasks',
+    })).resolves.toMatchObject({ passed: true });
+    expect(browser.page.goto).toHaveBeenCalledWith('https://example.test', expect.any(Object));
+  });
+
+  it('fails without invoking an agent when the selector is not visible', async () => {
+    const { browser } = makeBrowser({ visible: false });
+    await expect(browser.probe({ selector: '#dashboard' })).resolves.toMatchObject({
+      passed: false,
+      reason: 'selector not visible: #dashboard',
+    });
+  });
+});

package/tests/toolsTests/contextControl.test.js

@@ -109,6 +109,38 @@ describe('pruneConversationContext', () => {
     expect(nonSystem.map(m => m.content)).toEqual(['u-2', 'a-2']);
   });
 
+  it('keeps the latest real user instruction as an anchor for long tool histories', () => {
+    const messages = [
+      { role: 'system', content: 'system' },
+      { role: 'user', content: 'Complete the mission' },
+    ];
+
+    for (let i = 0; i < 25; i += 1) {
+      messages.push({
+        role: 'assistant',
+        tool_calls: [{
+          id: `call-${i}`,
+          type: 'function',
+          function: { name: 'get_dom', arguments: '{}' },
+        }],
+      });
+      messages.push({
+        role: 'tool',
+        tool_call_id: `call-${i}`,
+        name: 'get_dom',
+        content: `<main>${i}</main>`,
+      });
+    }
+
+    const pruned = pruneConversationContext(messages, { maxNonSystemMessages: 10 });
+
+    expect(pruned[0]).toEqual({ role: 'system', content: 'system' });
+    expect(pruned[1]).toEqual({ role: 'user', content: 'Complete the mission' });
+    expect(pruned.slice(2)).toHaveLength(10);
+    expect(pruned[2].role).toBe('assistant');
+    expect(pruned[3].role).toBe('tool');
+  });
+
   it('drops tool messages whose parent assistant tool_calls were pruned', () => {
     const messages = [
       // Old assistant + tool (should be pruned)
@@ -592,4 +624,4 @@ describe('summarizeGroundControlForPrompt', () => {
     expect(snapshot.navigation.routeRole).toBe('dashboard');
     expect(snapshot.navigation.currentLabel).toBe('Main dashboard');
   });
-});
+});

package/tests/toolsTests/domControl.test.js

@@ -16,6 +16,7 @@ import {
   collapseLongContent,
   chunkBySection,
   reduceRepeatedDivs,
+  reduceTableRows,
   removeObfuscatedClassNames,
   removeAutoGeneratedClasses,
   minifyDomWhitespace,
@@ -232,6 +233,17 @@ describe('domControl helpers', () => {
     });
   });
 
+  describe('reduceTableRows', () => {
+    it('keeps headers, trims body rows, and adds a marker', () => {
+      const $ = cheerio.load('<table><thead><tr><th>Name</th></tr></thead><tbody><tr><td>A</td></tr><tr><td>B</td></tr><tr><td>C</td></tr></tbody></table>');
+      reduceTableRows($, 2);
+      expect($('thead tr')).toHaveLength(1);
+      expect($('tbody > tr')).toHaveLength(3);
+      expect($('tbody > tr').eq(1).text()).toBe('B');
+      expect($('tbody > tr').eq(2).attr('data-collapsed')).toBe('true');
+    });
+  });
+
   describe('removeObfuscatedClassNames', () => {
     it('removes short and hashed-looking class names', () => {
       const html = `

package/tests/toolsTests/toolSchema.test.js

@@ -2,6 +2,14 @@ import { describe, it, expect } from 'vitest';
 import toolsSchema from '../../tools/toolSchema.js';
 
 describe('tools/toolSchema', () => {
+  it('does not advertise the unimplemented get_dom_chunk tool', () => {
+    expect(toolsSchema.map(tool => tool.function.name)).not.toContain('get_dom_chunk');
+  });
+
+  it('allows get_dom calls to be scoped by selector', () => {
+    const entry = toolsSchema.find(tool => tool.function.name === 'get_dom');
+    expect(entry.function.parameters.properties.selector.type).toBe('string');
+  });
   it('includes resource_progress tool with empty params', () => {
     const entry = toolsSchema.find(t => t.function?.name === 'resource_progress');
     expect(entry?.type).toBe('function');

package/tests/toolsTests/turnLoopUtils.test.js

@@ -2,6 +2,22 @@ import { describe, it, expect, vi } from 'vitest';
 import { finalResponseHandler } from '../../tools/turnLoopUtils.js';
 
 describe('tools/turnLoopUtils', () => {
+  it.each([
+    ['SUCCESS: completed', true],
+    ['FAILURE: blocked', false],
+    ['RESULT: SUCCESS - all checkpoints passed', true],
+    ['RESULT: FAILURE - checkpoint failed', false],
+    ['FINAL: success', true],
+    ['**RESULT: SUCCESS**', true],
+  ])('recognizes supported verdict form %s', (content, success) => {
+    const logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
+    try {
+      expect(finalResponseHandler({ content })).toMatchObject({ success });
+    } finally {
+      logSpy.mockRestore();
+    }
+  });
+
   it('redacts verification codes before printing and returning final responses', () => {
     const logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
 

package/tools/chromeBrowser.js

@@ -14,6 +14,7 @@ import * as cheerio from 'cheerio';
 import { 
   chunkBySection, 
   reduceRepeatedDivs, 
+  reduceTableRows,
   removeAutoGeneratedClasses, 
   removeNonContentTags,
   removeObfuscatedClassNames, 
@@ -57,6 +58,7 @@ export class ChromeBrowser {
     this.domListLimit = options.domListLimit;
     this.debugEnabled = !!options.debug;
     this._domLimitLogged = false;
+    this.storageStatePath = options.storageStatePath;
 
     const defaultResourceGuard = {
       enabled: true,
@@ -601,9 +603,12 @@ export class ChromeBrowser {
 
     ensureDir(FILES_DIR); // ⬅️ make sure missions/files exists
 
-    this.context = await this.browser.newContext({
-      acceptDownloads: true // ⬅️ important for Playwright download handling
-    });
+    const contextOptions = { acceptDownloads: true };
+    if (this.storageStatePath && fs.existsSync(this.storageStatePath)) {
+      contextOptions.storageState = this.storageStatePath;
+      this._debugLog('Restoring shared authentication state', { path: this.storageStatePath });
+    }
+    this.context = await this.browser.newContext(contextOptions);
     // this.context = await this.browser.newContext();
     const first = await this.context.newPage();
     const id = this._addAndFocus(first);
@@ -636,6 +641,26 @@ export class ChromeBrowser {
     return `navigated to ${this.page.url()}`;
   }
 
+  async probe({ url, selector, text, timeout = 5000 } = {}) {
+    if (url) {
+      await this.page.goto(url, { timeout: 30_000, waitUntil: 'domcontentloaded' });
+      await this.page.waitForLoadState('domcontentloaded').catch(() => {});
+    }
+
+    let locator = this.page.locator(selector || 'body').first();
+    if (selector) {
+      const visible = await locator.waitFor({ state: 'visible', timeout }).then(() => true).catch(() => false);
+      if (!visible) return { passed: false, reason: `selector not visible: ${selector}`, url: this.page.url() };
+    }
+    if (text) {
+      const content = await locator.textContent({ timeout }).catch(() => '');
+      if (!String(content || '').includes(String(text))) {
+        return { passed: false, reason: `text not found: ${text}`, url: this.page.url() };
+      }
+    }
+    return { passed: true, url: this.page.url() };
+  }
+
   async navigateEmailLink({ url, allowedHosts = [] }) {
     const destination = new URL(url);
     if (destination.protocol !== 'https:' || destination.username || destination.password) {
@@ -859,7 +884,7 @@ export class ChromeBrowser {
 
 
   async get_dom(
-    { limit = 100000, exclude = true, focus = [], loadState = 'networkidle', timeout = 10000 } = {},
+    { limit = 100000, exclude = true, selector, focus = [], loadState = 'networkidle', timeout = 10000 } = {},
     agentMemory = {}
   ) {
     const matchesHrefPattern = (href = '') => {
@@ -985,9 +1010,21 @@ export class ChromeBrowser {
     // --- 4) Optional DOM reduction/cleanup (your existing steps) ---
     let $ = cheerio.load(html);
 
+    if (selector) {
+      let selection;
+      try {
+        selection = $(selector);
+      } catch (error) {
+        throw new Error(`Invalid get_dom selector "${selector}": ${error.message}`);
+      }
+      if (!selection.length) return `[get_dom: no elements matched selector "${selector}"]`;
+      $ = cheerio.load(selection.toString());
+    }
+
     if (exclude) {
       $ = await removeNonContentTags($);
       $ = await reduceRepeatedDivs($, listLimit);
+      reduceTableRows($, listLimit);
       // If you later re-enable:
       // $ = await removeObfuscatedClassNames($);
       // $ = await removeAutoGeneratedClasses($);
@@ -1386,7 +1423,15 @@ export class ChromeBrowser {
 
 
   async close() {
-    await this.browser?.close();
+    try {
+      if (this.storageStatePath && this.context) {
+        ensureDir(path.dirname(this.storageStatePath));
+        await this.context.storageState({ path: this.storageStatePath, indexedDB: true });
+        this._debugLog('Saved shared authentication state', { path: this.storageStatePath });
+      }
+    } finally {
+      await this.browser?.close();
+    }
   }
 }
 

package/tools/contextControl.js

@@ -44,6 +44,7 @@ export function sanitizeHeavyToolHistory(messages, { keepRecentPerTool = 2 } = {
  *
  * Strategy:
  *   - Always keep all system messages.
+ *   - Always keep the latest real user instruction as a protocol anchor.
  *   - Start from the end and keep the last `maxNonSystemMessages` non-system messages.
  *   - Ensure protocol correctness:
  *       * If a tool message is kept, its parent assistant with matching tool_call_id
@@ -63,6 +64,15 @@ export function pruneConversationContext(messages, { maxNonSystemMessages = 40 }
   // Take last N non-system messages (raw window)
   const tail = nonSystem.slice(-maxNonSystemMessages);
 
+  // Gemini function-calling validation locates the current turn by scanning
+  // backward for a normal user message (not a function response). Testronaut's
+  // internal tool results use role `tool`, so retain the newest role `user`
+  // message even when a long agent loop pushes it outside the rolling tail.
+  // This is also useful context for OpenAI and Anthropic: it is the instruction
+  // the retained tool activity is attempting to satisfy.
+  const latestUser = nonSystem.findLast(m => m.role === 'user');
+  const userAnchor = latestUser && !tail.includes(latestUser) ? [latestUser] : [];
+
   // Pass 1: build map of tool_call_id -> index of assistant in tail
   const assistantToolParents = new Map(); // tool_call_id -> index in tail
   tail.forEach((m, idx) => {
@@ -93,7 +103,7 @@ export function pruneConversationContext(messages, { maxNonSystemMessages = 40 }
     cleanedTail.push(m);
   }
 
-  return [...systemMessages, ...cleanedTail];
+  return [...systemMessages, ...userAnchor, ...cleanedTail];
 }
 
 /* ---------------- Ground Control state helpers ----------------

package/tools/domControl.js

@@ -123,7 +123,7 @@ export const chunkBySection = async ($, focus) => {
     footer: $('footer').html() || '',
     aside: $('aside').html() || '',
     body: $('body').html() || ''
-  };
+};
 
   if (Array.isArray(focus) && focus.length > 0) {
     return Object.fromEntries(
@@ -144,6 +144,25 @@ export const chunkBySection = async ($, focus) => {
 
 }
 
+/** Keep table headers and only the first N body rows in each table. */
+export function reduceTableRows($, maxAllowed = 3) {
+  let limit = maxAllowed === Infinity ? Infinity : Number(maxAllowed);
+  if (limit === Infinity) return;
+  limit = Number.isFinite(limit) ? Math.max(0, limit) : 3;
+
+  $('table').each((_, table) => {
+    const $table = $(table);
+    const rows = $table.find('tbody > tr');
+    const candidates = rows.length ? rows : $table.children('tr');
+    if (candidates.length <= limit) return;
+    candidates.slice(limit).remove();
+    const colspan = Math.max(1, $table.find('tr').first().children('th, td').length);
+    const marker = `<tr data-collapsed="true"><td colspan="${colspan}">[...truncated]</td></tr>`;
+    if (rows.length) $table.children('tbody').first().append(marker);
+    else $table.append(marker);
+  });
+}
+
 /**
  * Reduce repeated sibling <div> blocks that share the same class name.
  * - For each parent <body> or <div>, group child divs by `class`.

package/tools/generateHtmlReport.js

@@ -102,7 +102,7 @@ export function generateHtmlReport(report, outputPath) {
             ${humanInput}
             ${planSpan}
             <span class="step-result ${ok ? 'ok' : 'bad'}" ${resultTooltip ? `title="${esc(resultTooltip)}"` : ''}>${esc(resultRaw)}</span>
-            <span class="tokens">tokens: ${esc(step.tokensUsed ?? '—')} / total: ${esc(step.totalTokensUsed ?? '—')}</span>
+            <span class="tokens">tokens: ${esc(step.tokensUsed ?? '—')} (in: ${esc(step.inputTokens ?? '—')}, out: ${esc(step.outputTokens ?? '—')}) / rolling: ${esc(step.totalTokensUsed ?? '—')}</span>
           </summary>
           <pre class="events">${esc(events.join('\n')) || '(no events)'}</pre>
           ${imgTag}

package/tools/tokenControl.js

@@ -85,13 +85,13 @@ function getTokenizer(model) {
   } catch (_) {
     const m = String(model || '').toLowerCase();
 
-    // Treat Gemini 2.5 like modern long-context models
-    const isGemini25 = /^gemini-2\.5/.test(m);
+    // Treat current Gemini generations like modern long-context models.
+    const isGemini = /^gemini-(?:2\.5|3(?:\.|-))/.test(m);
     const isClaude = /^claude-/.test(m);
 
     // OpenAI modern families also map well to o200k_base
     const useO200k =
-      isGemini25 ||
+      isGemini ||
       isClaude ||
       m.startsWith('gpt-5') ||
       m.startsWith('gpt-4o') ||

package/tools/toolSchema.js

@@ -83,6 +83,10 @@ const toolsSchema = [
             default: true,
             description: 'Whether to exclude noisy tags like script/style/etc.',
           },
+          selector: {
+            type: 'string',
+            description: 'Optional CSS selector that scopes the returned DOM to matching elements.',
+          },
         },
       },
     },
@@ -154,21 +158,6 @@ const toolsSchema = [
   //     }
   //   }
   // },
-  {
-    type: 'function',
-    function: {
-      name: 'get_dom_chunk',
-      description: 'Return a specific chunk of the DOM by index',
-      parameters: {
-        type: 'object',
-        properties: {
-          chunkIndex: { type: 'number' },
-          totalChunks: { type: 'number' }
-        },
-        required: ['chunkIndex', 'totalChunks']
-      }
-    }
-  },
   {
     type: 'function',
     function: {

package/tools/turnLoopUtils.js

@@ -3,14 +3,15 @@ import { redactPasswordInText } from '../core/redaction.js';
 export const finalResponseHandler = (msg) => {
   const content = String(msg.content ?? '');
   const safeContent = redactPasswordInText(content);
-  const final = content.trim().toLowerCase();
-  if (final?.startsWith('success')) {
+  const final = content.trim().replace(/^[#*_`\s]+/, '');
+  const verdict = final.match(/^(?:(?:result|final)\s*:\s*)?(success|failure)\b/i)?.[1]?.toLowerCase();
+  if (verdict === 'success') {
     console.log('\n┏━ FINAL AGENT RESPONSE ━━━━━━━━━━━━━━━━━━━');
     console.log(safeContent);
     console.log('┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
     return { finalMessage: safeContent, success: true};
   }
-  if (final?.startsWith('failure')) {
+  if (verdict === 'failure') {
     console.log('\n┏━ FINAL AGENT RESPONSE ━━━━━━━━━━━━━━━━━━━');
     console.log(safeContent);
     console.log('┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');