diff --git a/v2/tests/e2e/fixtures.ts b/v2/tests/e2e/fixtures.ts
new file mode 100644
index 0000000..11ff9c0
--- /dev/null
+++ b/v2/tests/e2e/fixtures.ts
@@ -0,0 +1,142 @@
+// Test fixture generator — creates isolated test environments
+// Used by E2E tests to set up temp data/config dirs with valid groups.json
+
+import { mkdirSync, writeFileSync, rmSync, existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { execSync } from 'node:child_process';
+import { tmpdir } from 'node:os';
+
+export interface TestFixture {
+  /** Root temp directory for this test run */
+  rootDir: string;
+  /** BTERMINAL_TEST_DATA_DIR — isolated data dir */
+  dataDir: string;
+  /** BTERMINAL_TEST_CONFIG_DIR — isolated config dir */
+  configDir: string;
+  /** Path to a minimal git repo for agent testing */
+  projectDir: string;
+  /** Environment variables to pass to the app */
+  env: Record<string, string>;
+}
+
+/**
+ * Create an isolated test fixture with:
+ * - Temp data dir (sessions.db, btmsg.db created at runtime)
+ * - Temp config dir with a minimal groups.json
+ * - A simple git repo with one file for agent testing
+ */
+export function createTestFixture(name = 'bterminal-e2e'): TestFixture {
+  const rootDir = join(tmpdir(), `${name}-${Date.now()}`);
+  const dataDir = join(rootDir, 'data');
+  const configDir = join(rootDir, 'config');
+  const projectDir = join(rootDir, 'test-project');
+
+  // Create directory structure
+  mkdirSync(dataDir, { recursive: true });
+  mkdirSync(configDir, { recursive: true });
+  mkdirSync(projectDir, { recursive: true });
+
+  // Create a minimal git repo for agent testing
+  execSync('git init', { cwd: projectDir, stdio: 'ignore' });
+  execSync('git config user.email "test@bterminal.dev"', { cwd: projectDir, stdio: 'ignore' });
+  execSync('git config user.name "BTerminal Test"', { cwd: projectDir, stdio: 'ignore' });
+  writeFileSync(join(projectDir, 'README.md'), '# Test Project\n\nA simple test project for BTerminal E2E tests.\n');
+  writeFileSync(join(projectDir, 'hello.py'), 'def greet(name: str) -> str:\n    return f"Hello, {name}!"\n');
+  execSync('git add -A && git commit -m "initial commit"', { cwd: projectDir, stdio: 'ignore' });
+
+  // Write groups.json with one group containing the test project
+  const groupsJson = {
+    version: 1,
+    groups: [
+      {
+        id: 'test-group',
+        name: 'Test Group',
+        projects: [
+          {
+            id: 'test-project',
+            name: 'Test Project',
+            identifier: 'test-project',
+            description: 'E2E test project',
+            icon: '\uf120',
+            cwd: projectDir,
+            profile: 'default',
+            enabled: true,
+          },
+        ],
+        agents: [],
+      },
+    ],
+    activeGroupId: 'test-group',
+  };
+
+  writeFileSync(
+    join(configDir, 'groups.json'),
+    JSON.stringify(groupsJson, null, 2),
+  );
+
+  const env: Record<string, string> = {
+    BTERMINAL_TEST: '1',
+    BTERMINAL_TEST_DATA_DIR: dataDir,
+    BTERMINAL_TEST_CONFIG_DIR: configDir,
+  };
+
+  return { rootDir, dataDir, configDir, projectDir, env };
+}
+
+/**
+ * Clean up a test fixture's temporary directories.
+ */
+export function destroyTestFixture(fixture: TestFixture): void {
+  if (existsSync(fixture.rootDir)) {
+    rmSync(fixture.rootDir, { recursive: true, force: true });
+  }
+}
+
+/**
+ * Create a groups.json with multiple projects for multi-project testing.
+ */
+export function createMultiProjectFixture(projectCount = 3): TestFixture {
+  const fixture = createTestFixture('bterminal-multi');
+
+  const projects = [];
+  for (let i = 0; i < projectCount; i++) {
+    const projDir = join(fixture.rootDir, `project-${i}`);
+    mkdirSync(projDir, { recursive: true });
+    execSync('git init', { cwd: projDir, stdio: 'ignore' });
+    execSync('git config user.email "test@bterminal.dev"', { cwd: projDir, stdio: 'ignore' });
+    execSync('git config user.name "BTerminal Test"', { cwd: projDir, stdio: 'ignore' });
+    writeFileSync(join(projDir, 'README.md'), `# Project ${i}\n`);
+    execSync('git add -A && git commit -m "init"', { cwd: projDir, stdio: 'ignore' });
+
+    projects.push({
+      id: `project-${i}`,
+      name: `Project ${i}`,
+      identifier: `project-${i}`,
+      description: `Test project ${i}`,
+      icon: '\uf120',
+      cwd: projDir,
+      profile: 'default',
+      enabled: true,
+    });
+  }
+
+  const groupsJson = {
+    version: 1,
+    groups: [
+      {
+        id: 'multi-group',
+        name: 'Multi Project Group',
+        projects,
+        agents: [],
+      },
+    ],
+    activeGroupId: 'multi-group',
+  };
+
+  writeFileSync(
+    join(fixture.configDir, 'groups.json'),
+    JSON.stringify(groupsJson, null, 2),
+  );
+
+  return fixture;
+}
diff --git a/v2/tests/e2e/results-db.ts b/v2/tests/e2e/results-db.ts
new file mode 100644
index 0000000..513088c
--- /dev/null
+++ b/v2/tests/e2e/results-db.ts
@@ -0,0 +1,113 @@
+// Test results store — persists test run outcomes as JSON for analysis
+// No native deps needed — reads/writes a JSON file
+
+import { resolve, dirname } from 'node:path';
+import { mkdirSync, readFileSync, writeFileSync, existsSync } from 'node:fs';
+import { fileURLToPath } from 'node:url';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const DEFAULT_PATH = resolve(__dirname, '../../test-results/results.json');
+
+export interface TestRunRow {
+  run_id: string;
+  started_at: string;
+  finished_at: string | null;
+  status: 'running' | 'passed' | 'failed' | 'error';
+  total_tests: number;
+  passed_tests: number;
+  failed_tests: number;
+  duration_ms: number | null;
+  git_branch: string | null;
+  git_sha: string | null;
+}
+
+export interface TestStepRow {
+  run_id: string;
+  scenario_name: string;
+  step_name: string;
+  status: 'passed' | 'failed' | 'skipped' | 'error';
+  duration_ms: number | null;
+  error_message: string | null;
+  screenshot_path: string | null;
+  agent_cost_usd: number | null;
+  created_at: string;
+}
+
+interface ResultsStore {
+  runs: TestRunRow[];
+  steps: TestStepRow[];
+}
+
+export class ResultsDb {
+  private filePath: string;
+  private store: ResultsStore;
+
+  constructor(filePath = DEFAULT_PATH) {
+    this.filePath = filePath;
+    mkdirSync(dirname(filePath), { recursive: true });
+    this.store = this.load();
+  }
+
+  private load(): ResultsStore {
+    if (existsSync(this.filePath)) {
+      try {
+        return JSON.parse(readFileSync(this.filePath, 'utf-8'));
+      } catch {
+        return { runs: [], steps: [] };
+      }
+    }
+    return { runs: [], steps: [] };
+  }
+
+  private save(): void {
+    writeFileSync(this.filePath, JSON.stringify(this.store, null, 2));
+  }
+
+  startRun(runId: string, gitBranch?: string, gitSha?: string): void {
+    this.store.runs.push({
+      run_id: runId,
+      started_at: new Date().toISOString(),
+      finished_at: null,
+      status: 'running',
+      total_tests: 0,
+      passed_tests: 0,
+      failed_tests: 0,
+      duration_ms: null,
+      git_branch: gitBranch ?? null,
+      git_sha: gitSha ?? null,
+    });
+    this.save();
+  }
+
+  finishRun(runId: string, status: 'passed' | 'failed' | 'error', durationMs: number): void {
+    const run = this.store.runs.find(r => r.run_id === runId);
+    if (!run) return;
+
+    const steps = this.store.steps.filter(s => s.run_id === runId);
+    run.finished_at = new Date().toISOString();
+    run.status = status;
+    run.duration_ms = durationMs;
+    run.total_tests = steps.length;
+    run.passed_tests = steps.filter(s => s.status === 'passed').length;
+    run.failed_tests = steps.filter(s => s.status === 'failed' || s.status === 'error').length;
+    this.save();
+  }
+
+  recordStep(step: Omit<TestStepRow, 'created_at'>): void {
+    this.store.steps.push({
+      ...step,
+      created_at: new Date().toISOString(),
+    });
+    this.save();
+  }
+
+  getRecentRuns(limit = 20): TestRunRow[] {
+    return this.store.runs
+      .sort((a, b) => b.started_at.localeCompare(a.started_at))
+      .slice(0, limit);
+  }
+
+  getStepsForRun(runId: string): TestStepRow[] {
+    return this.store.steps.filter(s => s.run_id === runId);
+  }
+}
diff --git a/v2/tests/e2e/specs/agent-scenarios.test.ts b/v2/tests/e2e/specs/agent-scenarios.test.ts
new file mode 100644
index 0000000..243cd77
--- /dev/null
+++ b/v2/tests/e2e/specs/agent-scenarios.test.ts
@@ -0,0 +1,411 @@
+import { browser, expect } from '@wdio/globals';
+
+// Phase A: Human-authored E2E scenarios with deterministic assertions.
+// These test the agent UI flow end-to-end using stable data-testid selectors.
+// Agent-interaction tests require a real Claude CLI install + API key.
+
+// ─── Helpers ──────────────────────────────────────────────────────────
+
+/** Wait for agent status to reach a target value within timeout. */
+async function waitForAgentStatus(
+  status: string,
+  timeout = 30_000,
+): Promise<void> {
+  await browser.waitUntil(
+    async () => {
+      const attr = await browser.execute(() => {
+        const el = document.querySelector('[data-testid="agent-pane"]');
+        return el?.getAttribute('data-agent-status') ?? 'idle';
+      });
+      return attr === status;
+    },
+    { timeout, timeoutMsg: `Agent did not reach status "${status}" within ${timeout}ms` },
+  );
+}
+
+/** Check if an agent pane exists and is visible. */
+async function agentPaneExists(): Promise<boolean> {
+  const el = await browser.$('[data-testid="agent-pane"]');
+  return el.isExisting();
+}
+
+/** Type a prompt into the agent textarea and submit. */
+async function sendAgentPrompt(text: string): Promise<void> {
+  const textarea = await browser.$('[data-testid="agent-prompt"]');
+  await textarea.waitForDisplayed({ timeout: 5000 });
+  await textarea.setValue(text);
+  // Small delay for Svelte reactivity
+  await browser.pause(200);
+  const submitBtn = await browser.$('[data-testid="agent-submit"]');
+  await browser.execute((el) => (el as HTMLElement).click(), submitBtn);
+}
+
+// ─── Scenario 1: App renders with project grid and data-testid anchors ───
+
+describe('Scenario 1 — App Structural Integrity', () => {
+  it('should render the status bar with data-testid', async () => {
+    const bar = await browser.$('[data-testid="status-bar"]');
+    await expect(bar).toBeDisplayed();
+  });
+
+  it('should render the sidebar rail with data-testid', async () => {
+    const rail = await browser.$('[data-testid="sidebar-rail"]');
+    await expect(rail).toBeDisplayed();
+  });
+
+  it('should render at least one project box with data-testid', async () => {
+    const boxes = await browser.$$('[data-testid="project-box"]');
+    expect(boxes.length).toBeGreaterThanOrEqual(1);
+  });
+
+  it('should have data-project-id on project boxes', async () => {
+    const projectId = await browser.execute(() => {
+      const box = document.querySelector('[data-testid="project-box"]');
+      return box?.getAttribute('data-project-id') ?? null;
+    });
+    expect(projectId).not.toBeNull();
+    expect((projectId as string).length).toBeGreaterThan(0);
+  });
+
+  it('should render project tabs with data-testid', async () => {
+    const tabs = await browser.$('[data-testid="project-tabs"]');
+    await expect(tabs).toBeDisplayed();
+  });
+
+  it('should render agent session component', async () => {
+    const session = await browser.$('[data-testid="agent-session"]');
+    await expect(session).toBeDisplayed();
+  });
+});
+
+// ─── Scenario 2: Settings panel via data-testid ──────────────────────
+
+describe('Scenario 2 — Settings Panel (data-testid)', () => {
+  it('should open settings via data-testid button', async () => {
+    const btn = await browser.$('[data-testid="settings-btn"]');
+    await btn.click();
+
+    const panel = await browser.$('.sidebar-panel');
+    await panel.waitForDisplayed({ timeout: 5000 });
+    await expect(panel).toBeDisplayed();
+  });
+
+  it('should close settings with Escape', async () => {
+    await browser.keys('Escape');
+    const panel = await browser.$('.sidebar-panel');
+    await panel.waitForDisplayed({ timeout: 3000, reverse: true });
+  });
+});
+
+// ─── Scenario 3: Agent pane initial state ────────────────────────────
+
+describe('Scenario 3 — Agent Pane Initial State', () => {
+  it('should display agent pane in idle status', async () => {
+    const exists = await agentPaneExists();
+    if (!exists) {
+      // Agent pane might not be visible until Model tab is active
+      await browser.execute(() => {
+        const tab = document.querySelector('[data-testid="project-tabs"] .ptab');
+        if (tab) (tab as HTMLElement).click();
+      });
+      await browser.pause(300);
+    }
+
+    const pane = await browser.$('[data-testid="agent-pane"]');
+    await expect(pane).toBeExisting();
+
+    const status = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="agent-pane"]');
+      return el?.getAttribute('data-agent-status') ?? 'unknown';
+    });
+    expect(status).toBe('idle');
+  });
+
+  it('should show prompt textarea', async () => {
+    const textarea = await browser.$('[data-testid="agent-prompt"]');
+    await expect(textarea).toBeDisplayed();
+  });
+
+  it('should show submit button', async () => {
+    const btn = await browser.$('[data-testid="agent-submit"]');
+    await expect(btn).toBeExisting();
+  });
+
+  it('should have empty messages area initially', async () => {
+    const msgArea = await browser.$('[data-testid="agent-messages"]');
+    await expect(msgArea).toBeExisting();
+
+    // No message bubbles should exist in a fresh session
+    const msgCount = await browser.execute(() => {
+      const area = document.querySelector('[data-testid="agent-messages"]');
+      if (!area) return 0;
+      return area.querySelectorAll('.message').length;
+    });
+    expect(msgCount).toBe(0);
+  });
+});
+
+// ─── Scenario 4: Terminal tab management ─────────────────────────────
+
+describe('Scenario 4 — Terminal Tab Management (data-testid)', () => {
+  before(async () => {
+    // Ensure Model tab is active and terminal section visible
+    await browser.execute(() => {
+      const tab = document.querySelector('[data-testid="project-tabs"] .ptab');
+      if (tab) (tab as HTMLElement).click();
+    });
+    await browser.pause(300);
+
+    // Expand terminal section
+    await browser.execute(() => {
+      const toggle = document.querySelector('[data-testid="terminal-toggle"]');
+      if (toggle) (toggle as HTMLElement).click();
+    });
+    await browser.pause(500);
+  });
+
+  it('should display terminal tabs container', async () => {
+    const tabs = await browser.$('[data-testid="terminal-tabs"]');
+    await expect(tabs).toBeDisplayed();
+  });
+
+  it('should add a shell tab via data-testid button', async () => {
+    await browser.execute(() => {
+      const btn = document.querySelector('[data-testid="tab-add"]');
+      if (btn) (btn as HTMLElement).click();
+    });
+    await browser.pause(500);
+
+    const tabTitle = await browser.execute(() => {
+      const el = document.querySelector('.tab-bar .tab-title');
+      return el?.textContent ?? '';
+    });
+    expect(tabTitle.toLowerCase()).toContain('shell');
+  });
+
+  it('should show active tab styling', async () => {
+    const activeTab = await browser.$('.tab.active');
+    await expect(activeTab).toBeExisting();
+  });
+
+  it('should close tab and show empty state', async () => {
+    // Close all tabs
+    await browser.execute(() => {
+      const closeBtns = document.querySelectorAll('.tab-close');
+      closeBtns.forEach(btn => (btn as HTMLElement).click());
+    });
+    await browser.pause(500);
+
+    // Should show empty terminal area with "Open terminal" button
+    const emptyBtn = await browser.$('.add-first');
+    await expect(emptyBtn).toBeDisplayed();
+  });
+
+  after(async () => {
+    // Collapse terminal section
+    await browser.execute(() => {
+      const toggle = document.querySelector('[data-testid="terminal-toggle"]');
+      const chevron = toggle?.querySelector('.toggle-chevron.expanded');
+      if (chevron) (toggle as HTMLElement).click();
+    });
+    await browser.pause(300);
+  });
+});
+
+// ─── Scenario 5: Command palette with data-testid ───────────────────
+
+describe('Scenario 5 — Command Palette (data-testid)', () => {
+  it('should open palette and show data-testid input', async () => {
+    await browser.execute(() => document.body.focus());
+    await browser.pause(200);
+    await browser.keys(['Control', 'k']);
+
+    const palette = await browser.$('[data-testid="command-palette"]');
+    await palette.waitForDisplayed({ timeout: 3000 });
+
+    const input = await browser.$('[data-testid="palette-input"]');
+    await expect(input).toBeDisplayed();
+  });
+
+  it('should have focused input', async () => {
+    const isFocused = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="palette-input"]');
+      return el === document.activeElement;
+    });
+    expect(isFocused).toBe(true);
+  });
+
+  it('should show at least one group item', async () => {
+    const items = await browser.$$('.palette-item');
+    expect(items.length).toBeGreaterThanOrEqual(1);
+  });
+
+  it('should filter and show no-results for nonsense query', async () => {
+    const input = await browser.$('[data-testid="palette-input"]');
+    await input.setValue('zzz_no_match_xyz');
+    await browser.pause(300);
+
+    const noResults = await browser.$('.no-results');
+    await expect(noResults).toBeDisplayed();
+  });
+
+  it('should close on Escape', async () => {
+    await browser.keys('Escape');
+    const palette = await browser.$('[data-testid="command-palette"]');
+    await browser.waitUntil(
+      async () => !(await palette.isDisplayed()),
+      { timeout: 3000 },
+    );
+  });
+});
+
+// ─── Scenario 6: Project focus and tab switching ─────────────────────
+
+describe('Scenario 6 — Project Focus & Tab Switching', () => {
+  it('should focus project on header click', async () => {
+    await browser.execute(() => {
+      const header = document.querySelector('.project-header');
+      if (header) (header as HTMLElement).click();
+    });
+    await browser.pause(300);
+
+    const activeBox = await browser.$('.project-box.active');
+    await expect(activeBox).toBeDisplayed();
+  });
+
+  it('should switch to Files tab and back without losing agent session', async () => {
+    // Get current agent session element reference
+    const sessionBefore = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="agent-session"]');
+      return el !== null;
+    });
+    expect(sessionBefore).toBe(true);
+
+    // Switch to Files tab (second tab)
+    await browser.execute(() => {
+      const tabs = document.querySelectorAll('[data-testid="project-tabs"] .ptab');
+      if (tabs.length >= 2) (tabs[1] as HTMLElement).click();
+    });
+    await browser.pause(500);
+
+    // AgentSession should still exist in DOM (display:none, not unmounted)
+    const sessionDuring = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="agent-session"]');
+      return el !== null;
+    });
+    expect(sessionDuring).toBe(true);
+
+    // Switch back to Model tab
+    await browser.execute(() => {
+      const tab = document.querySelector('[data-testid="project-tabs"] .ptab');
+      if (tab) (tab as HTMLElement).click();
+    });
+    await browser.pause(300);
+
+    // Agent session should be visible again
+    const session = await browser.$('[data-testid="agent-session"]');
+    await expect(session).toBeDisplayed();
+  });
+
+  it('should preserve agent status across tab switches', async () => {
+    const statusBefore = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="agent-pane"]');
+      return el?.getAttribute('data-agent-status') ?? 'unknown';
+    });
+
+    // Switch to Context tab (third tab) and back
+    await browser.execute(() => {
+      const tabs = document.querySelectorAll('[data-testid="project-tabs"] .ptab');
+      if (tabs.length >= 3) (tabs[2] as HTMLElement).click();
+    });
+    await browser.pause(300);
+    await browser.execute(() => {
+      const tab = document.querySelector('[data-testid="project-tabs"] .ptab');
+      if (tab) (tab as HTMLElement).click();
+    });
+    await browser.pause(300);
+
+    const statusAfter = await browser.execute(() => {
+      const el = document.querySelector('[data-testid="agent-pane"]');
+      return el?.getAttribute('data-agent-status') ?? 'unknown';
+    });
+    expect(statusAfter).toBe(statusBefore);
+  });
+});
+
+// ─── Scenario 7: Agent prompt interaction (requires Claude CLI) ──────
+
+describe('Scenario 7 — Agent Prompt Submission', () => {
+  // This scenario requires a real Claude CLI + API key.
+  // Skip gracefully if agent doesn't transition to "running" within timeout.
+
+  it('should accept text in prompt textarea', async () => {
+    const textarea = await browser.$('[data-testid="agent-prompt"]');
+    await textarea.waitForDisplayed({ timeout: 5000 });
+    await textarea.setValue('Say hello');
+    await browser.pause(200);
+
+    const value = await textarea.getValue();
+    expect(value).toBe('Say hello');
+
+    // Clear without submitting
+    await textarea.clearValue();
+  });
+
+  it('should enable submit button when prompt has text', async () => {
+    const textarea = await browser.$('[data-testid="agent-prompt"]');
+    await textarea.setValue('Test prompt');
+    await browser.pause(200);
+
+    // Submit button should be interactable (not disabled)
+    const isDisabled = await browser.execute(() => {
+      const btn = document.querySelector('[data-testid="agent-submit"]');
+      if (!btn) return true;
+      return (btn as HTMLButtonElement).disabled;
+    });
+    expect(isDisabled).toBe(false);
+
+    await textarea.clearValue();
+  });
+
+  it('should show stop button during agent execution (if Claude available)', async function () {
+    // Check if Claude CLI is likely available by looking for env hint
+    const hasClaude = await browser.execute(() => {
+      // If test mode is active and no ANTHROPIC_API_KEY, skip
+      return true; // Optimistic — let the timeout catch failures
+    });
+
+    if (!hasClaude) {
+      this.skip();
+      return;
+    }
+
+    // Send a minimal prompt
+    await sendAgentPrompt('Reply with exactly: BTERMINAL_TEST_OK');
+
+    // Wait for running status (generous timeout for sidecar spin-up)
+    try {
+      await waitForAgentStatus('running', 15_000);
+    } catch {
+      // Claude CLI not available — skip remaining assertions
+      console.log('Agent did not start — Claude CLI may not be available. Skipping.');
+      this.skip();
+      return;
+    }
+
+    // Stop button should appear while running
+    const stopBtn = await browser.$('[data-testid="agent-stop"]');
+    await expect(stopBtn).toBeDisplayed();
+
+    // Wait for completion
+    await waitForAgentStatus('idle', 60_000);
+
+    // Messages area should now have content
+    const msgCount = await browser.execute(() => {
+      const area = document.querySelector('[data-testid="agent-messages"]');
+      if (!area) return 0;
+      return area.children.length;
+    });
+    expect(msgCount).toBeGreaterThan(0);
+  });
+});
diff --git a/v2/tests/e2e/tsconfig.json b/v2/tests/e2e/tsconfig.json
index f7e974d..a1c3b70 100644
--- a/v2/tests/e2e/tsconfig.json
+++ b/v2/tests/e2e/tsconfig.json
@@ -7,5 +7,5 @@
     "esModuleInterop": true,
     "skipLibCheck": true
   },
-  "include": ["specs/**/*.ts"]
+  "include": ["specs/**/*.ts", "*.ts"]
 }
diff --git a/v2/tests/e2e/wdio.conf.js b/v2/tests/e2e/wdio.conf.js
index 35a71f5..37dbec5 100644
--- a/v2/tests/e2e/wdio.conf.js
+++ b/v2/tests/e2e/wdio.conf.js
@@ -1,4 +1,5 @@
 import { spawn } from 'node:child_process';
+import { createConnection } from 'node:net';
 import { resolve, dirname } from 'node:path';
 import { fileURLToPath } from 'node:url';
 
@@ -24,7 +25,10 @@ export const config = {
   // ── Specs ──
   // Single spec file — Tauri launches one app instance per session,
   // and tauri-driver can't re-create sessions between spec files.
-  specs: [resolve(__dirname, 'specs/bterminal.test.ts')],
+  specs: [
+    resolve(__dirname, 'specs/bterminal.test.ts'),
+    resolve(__dirname, 'specs/agent-scenarios.test.ts'),
+  ],
 
   // ── Capabilities ──
   capabilities: [{
@@ -32,6 +36,12 @@ export const config = {
     'wdio:enforceWebDriverClassic': true,
     'tauri:options': {
       application: tauriBinary,
+      // Test isolation: separate data/config dirs, disable watchers/telemetry
+      env: {
+        BTERMINAL_TEST: '1',
+        ...(process.env.BTERMINAL_TEST_DATA_DIR && { BTERMINAL_TEST_DATA_DIR: process.env.BTERMINAL_TEST_DATA_DIR }),
+        ...(process.env.BTERMINAL_TEST_CONFIG_DIR && { BTERMINAL_TEST_CONFIG_DIR: process.env.BTERMINAL_TEST_CONFIG_DIR }),
+      },
     },
   }],
 
@@ -85,9 +95,10 @@ export const config = {
   /**
    * Spawn tauri-driver before the session.
    * tauri-driver bridges WebDriver protocol to WebKit2GTK's inspector.
+   * Uses TCP probe to confirm port 4444 is accepting connections.
    */
   beforeSession() {
-    return new Promise((resolve, reject) => {
+    return new Promise((res, reject) => {
       tauriDriver = spawn('tauri-driver', [], {
         stdio: ['ignore', 'pipe', 'pipe'],
       });
@@ -99,14 +110,28 @@ export const config = {
         ));
       });
 
-      // Wait for tauri-driver to be ready (listens on port 4444)
-      const timeout = setTimeout(() => resolve(), 2000);
-      tauriDriver.stdout.on('data', (data) => {
-        if (data.toString().includes('4444')) {
-          clearTimeout(timeout);
-          resolve();
+      // TCP readiness probe — poll port 4444 until it accepts a connection
+      const maxWaitMs = 10_000;
+      const intervalMs = 200;
+      const deadline = Date.now() + maxWaitMs;
+
+      function probe() {
+        if (Date.now() > deadline) {
+          reject(new Error('tauri-driver did not become ready within 10s'));
+          return;
         }
-      });
+        const sock = createConnection({ port: 4444, host: 'localhost' }, () => {
+          sock.destroy();
+          res();
+        });
+        sock.on('error', () => {
+          sock.destroy();
+          setTimeout(probe, intervalMs);
+        });
+      }
+
+      // Give it a moment before first probe
+      setTimeout(probe, 300);
     });
   },