feat(testing): skill activation integration test infrastructure

package.json

@@ -26,7 +26,8 @@
     "prepublishOnly": "npm run build",
     "version:bump": "npx tsx scripts/bump-version.ts",
     "test": "vitest run",
-    "test:watch": "vitest"
+    "test:watch": "vitest",
+    "test:integration": "vitest run --config vitest.integration.config.ts"
   },
   "keywords": [
     "claude",

tests/integration/ambient-activation.test.ts

@@ -0,0 +1,57 @@
+import { describe, it, expect } from 'vitest';
+import {
+  isClaudeAvailable,
+  runClaude,
+  hasClassification,
+  isQuietResponse,
+  extractIntent,
+  extractDepth,
+} from './helpers.js';
+
+/**
+ * Integration tests for ambient mode skill activation.
+ *
+ * These tests require:
+ * - `claude` CLI installed and authenticated
+ * - Ambient mode enabled (`devflow ambient --enable`)
+ * - DevFlow skills installed (`devflow init`)
+ *
+ * Run manually: npm run test:integration
+ * Not part of `npm test` — each test is an API call.
+ */
+describe.skipIf(!isClaudeAvailable())('ambient classification', () => {
+  // QUICK tier — no skills loaded, no classification output
+  it('classifies "thanks" as QUICK (silent)', () => {
+    const output = runClaude('thanks');
+    expect(isQuietResponse(output)).toBe(true);
+  });
+
+  it('classifies "commit this" as QUICK (git op)', () => {
+    const output = runClaude('commit the current changes');
+    // Git operations should not trigger STANDARD classification
+    expect(isQuietResponse(output) || extractDepth(output) === 'QUICK').toBe(true);
+  });
+
+  // STANDARD tier — skills referenced in output
+  it('classifies "add a login form" as BUILD/STANDARD', () => {
+    const output = runClaude('add a login form with email and password fields');
+    if (hasClassification(output)) {
+      expect(extractIntent(output)).toBe('BUILD');
+      expect(extractDepth(output)).toBe('STANDARD');
+    }
+    // Even without explicit classification, BUILD prompts should reference TDD
+    expect(
+      output.toLowerCase().includes('test') ||
+      output.toLowerCase().includes('tdd') ||
+      hasClassification(output)
+    ).toBe(true);
+  });
+
+  it('classifies "fix the auth error" as DEBUG/STANDARD', () => {
+    const output = runClaude('fix the authentication error in the login handler');
+    if (hasClassification(output)) {
+      expect(extractIntent(output)).toBe('DEBUG');
+      expect(['STANDARD', 'ESCALATE']).toContain(extractDepth(output));
+    }
+  });
+});

tests/integration/helpers.ts

@@ -0,0 +1,64 @@
+import { execSync } from 'child_process';
+
+/**
+ * Check if the `claude` CLI is available on this machine.
+ */
+export function isClaudeAvailable(): boolean {
+  try {
+    execSync('claude --version', { stdio: 'pipe', timeout: 5000 });
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+/**
+ * Run a prompt through claude CLI in non-interactive mode.
+ * Returns the text output.
+ */
+export function runClaude(prompt: string, options?: { timeout?: number }): string {
+  const timeout = options?.timeout ?? 30000;
+
+  const result = execSync(
+    `claude -p --output-format text --model haiku "${prompt.replace(/"/g, '\\"')}"`,
+    {
+      stdio: 'pipe',
+      timeout,
+      encoding: 'utf-8',
+    },
+  );
+
+  return result.trim();
+}
+
+/**
+ * Assert that output contains a classification marker (case-insensitive).
+ * Classification markers look like: "Ambient: BUILD/STANDARD"
+ */
+export function hasClassification(output: string): boolean {
+  return /ambient:\s*(BUILD|DEBUG|REVIEW|PLAN|EXPLORE|CHAT)\s*\/\s*(QUICK|STANDARD|ESCALATE)/i.test(output);
+}
+
+/**
+ * Assert that output does NOT contain a classification marker.
+ * QUICK responses should be silent — no classification output.
+ */
+export function isQuietResponse(output: string): boolean {
+  return !hasClassification(output);
+}
+
+/**
+ * Extract the intent from a classification marker.
+ */
+export function extractIntent(output: string): string | null {
+  const match = output.match(/ambient:\s*(BUILD|DEBUG|REVIEW|PLAN|EXPLORE|CHAT)/i);
+  return match ? match[1].toUpperCase() : null;
+}
+
+/**
+ * Extract the depth from a classification marker.
+ */
+export function extractDepth(output: string): string | null {
+  const match = output.match(/ambient:\s*\w+\s*\/\s*(QUICK|STANDARD|ESCALATE)/i);
+  return match ? match[1].toUpperCase() : null;
+}

vitest.config.ts

@@ -4,6 +4,7 @@ export default defineConfig({
   test: {
     root: '.',
     include: ['tests/**/*.test.ts'],
+    exclude: ['tests/integration/**'],
     globals: false,
     environment: 'node',
     restoreMocks: true,

vitest.integration.config.ts

@@ -0,0 +1,18 @@
+import { defineConfig } from 'vitest/config';
+
+export default defineConfig({
+  test: {
+    root: '.',
+    include: ['tests/integration/**/*.test.ts'],
+    globals: false,
+    environment: 'node',
+    restoreMocks: true,
+    testTimeout: 60000,
+    retry: 2,
+  },
+  resolve: {
+    alias: {
+      '#cli': new URL('./src/cli/', import.meta.url).pathname,
+    },
+  },
+});