Merge pull request #168 from SentienceAPI/debugger

Sentience Debugger for any agent

Author: rcholic

src/agent-runtime.ts

@@ -55,6 +55,7 @@ import { Tracer } from './tracing/tracer';
 import { TraceEventBuilder } from './utils/trace-event-builder';
 import { LLMProvider } from './llm-provider';
 import { FailureArtifactBuffer, FailureArtifactsOptions } from './failure-artifacts';
+import { SentienceBrowser } from './browser';
 import type { ToolRegistry } from './tools/registry';
 import {
   CaptchaContext,
@@ -69,6 +70,13 @@ interface BrowserLike {
   snapshot(page: Page, options?: Record<string, any>): Promise<Snapshot>;
 }
 
+export interface AttachOptions {
+  apiKey?: string;
+  apiUrl?: string;
+  toolRegistry?: ToolRegistry;
+  browser?: BrowserLike;
+}
+
 const DEFAULT_CAPTCHA_OPTIONS: Required<Omit<CaptchaOptions, 'handler' | 'resetSession'>> = {
   policy: 'abort',
   minConfidence: 0.7,
@@ -464,6 +472,30 @@ export class AgentRuntime {
     return new AssertionHandle(this, predicate, label, required);
   }
 
+  /**
+   * Create AgentRuntime from a raw Playwright Page (sidecar mode).
+   */
+  static fromPlaywrightPage(page: Page, tracer: Tracer, options?: AttachOptions): AgentRuntime {
+    const browser =
+      options?.browser ??
+      ((): BrowserLike => {
+        const sentienceBrowser = SentienceBrowser.fromPage(page, options?.apiKey, options?.apiUrl);
+        return {
+          snapshot: async (_page: Page, snapshotOptions?: Record<string, any>) =>
+            sentienceBrowser.snapshot(snapshotOptions),
+        };
+      })();
+
+    return new AgentRuntime(browser, page, tracer, options?.toolRegistry);
+  }
+
+  /**
+   * Sidecar alias for fromPlaywrightPage().
+   */
+  static attach(page: Page, tracer: Tracer, options?: AttachOptions): AgentRuntime {
+    return AgentRuntime.fromPlaywrightPage(page, tracer, options);
+  }
+
   /**
    * Create a new AgentRuntime.
    *

src/debugger.ts

@@ -0,0 +1,50 @@
+import { Page } from 'playwright';
+
+import { AgentRuntime, AttachOptions } from './agent-runtime';
+import { Predicate } from './verification';
+import { Tracer } from './tracing/tracer';
+
+export class SentienceDebugger {
+  readonly runtime: AgentRuntime;
+  private stepOpen: boolean = false;
+
+  constructor(runtime: AgentRuntime) {
+    this.runtime = runtime;
+  }
+
+  static attach(page: Page, tracer: Tracer, options?: AttachOptions): SentienceDebugger {
+    const runtime = AgentRuntime.fromPlaywrightPage(page, tracer, options);
+    return new SentienceDebugger(runtime);
+  }
+
+  beginStep(goal: string, stepIndex?: number): string {
+    this.stepOpen = true;
+    return this.runtime.beginStep(goal, stepIndex);
+  }
+
+  async endStep(opts: Parameters<AgentRuntime['emitStepEnd']>[0] = {}): Promise<any> {
+    this.stepOpen = false;
+    // emitStepEnd is synchronous; wrap to satisfy async/await lint rules.
+    return await Promise.resolve(this.runtime.emitStepEnd(opts));
+  }
+
+  async step(goal: string, fn: () => Promise<void> | void, stepIndex?: number): Promise<void> {
+    this.beginStep(goal, stepIndex);
+    try {
+      await fn();
+    } finally {
+      await this.endStep();
+    }
+  }
+
+  async snapshot(options?: Record<string, any>) {
+    return this.runtime.snapshot(options);
+  }
+
+  check(predicate: Predicate, label: string, required: boolean = false) {
+    if (!this.stepOpen) {
+      this.beginStep(`verify:${label}`);
+    }
+    return this.runtime.check(predicate, label, required);
+  }
+}

src/index.ts

@@ -87,6 +87,7 @@ export {
   isCollapsed,
 } from './verification';
 export { AgentRuntime, AssertionHandle, AssertionRecord, EventuallyOptions } from './agent-runtime';
+export { SentienceDebugger } from './debugger';
 export { RuntimeAgent } from './runtime-agent';
 export type { RuntimeStep, StepVerification } from './runtime-agent';
 export { parseVisionExecutorAction, executeVisionExecutorAction } from './vision-executor';

tests/agent-runtime-attach.test.ts

@@ -0,0 +1,69 @@
+import { AgentRuntime } from '../src/agent-runtime';
+import { SentienceBrowser } from '../src/browser';
+import { TraceSink } from '../src/tracing/sink';
+import { Tracer } from '../src/tracing/tracer';
+import { MockPage } from './mocks/browser-mock';
+
+class MockSink extends TraceSink {
+  emit(): void {
+    // no-op
+  }
+  async close(): Promise<void> {
+    // no-op
+  }
+  getSinkType(): string {
+    return 'MockSink';
+  }
+}
+
+describe('AgentRuntime.fromPlaywrightPage()', () => {
+  it('creates runtime using SentienceBrowser.fromPage()', () => {
+    const sink = new MockSink();
+    const tracer = new Tracer('test-run', sink);
+    const page = new MockPage('https://example.com') as any;
+    const browserLike = {
+      snapshot: async () => ({
+        status: 'success',
+        url: 'https://example.com',
+        elements: [],
+        timestamp: 't1',
+      }),
+    };
+
+    const spy = jest.spyOn(SentienceBrowser, 'fromPage').mockReturnValue(browserLike as any);
+
+    const runtime = AgentRuntime.fromPlaywrightPage(page, tracer);
+
+    expect(spy).toHaveBeenCalledWith(page, undefined, undefined);
+    expect(typeof runtime.browser.snapshot).toBe('function');
+    expect(runtime.page).toBe(page);
+
+    spy.mockRestore();
+  });
+
+  it('passes apiKey and apiUrl to SentienceBrowser.fromPage()', () => {
+    const sink = new MockSink();
+    const tracer = new Tracer('test-run', sink);
+    const page = new MockPage('https://example.com') as any;
+    const browserLike = {
+      snapshot: async () => ({
+        status: 'success',
+        url: 'https://example.com',
+        elements: [],
+        timestamp: 't1',
+      }),
+    };
+
+    const spy = jest.spyOn(SentienceBrowser, 'fromPage').mockReturnValue(browserLike as any);
+
+    const runtime = AgentRuntime.fromPlaywrightPage(page, tracer, {
+      apiKey: 'sk_test',
+      apiUrl: 'https://api.example.com',
+    });
+
+    expect(spy).toHaveBeenCalledWith(page, 'sk_test', 'https://api.example.com');
+    expect(typeof runtime.browser.snapshot).toBe('function');
+
+    spy.mockRestore();
+  });
+});

tests/debugger.test.ts

@@ -0,0 +1,69 @@
+import { AgentRuntime } from '../src/agent-runtime';
+import { SentienceDebugger } from '../src/debugger';
+import { TraceSink } from '../src/tracing/sink';
+import { Tracer } from '../src/tracing/tracer';
+import { MockPage } from './mocks/browser-mock';
+
+class MockSink extends TraceSink {
+  emit(): void {
+    // no-op
+  }
+  async close(): Promise<void> {
+    // no-op
+  }
+  getSinkType(): string {
+    return 'MockSink';
+  }
+}
+
+describe('SentienceDebugger', () => {
+  it('attaches via AgentRuntime.fromPlaywrightPage()', () => {
+    const sink = new MockSink();
+    const tracer = new Tracer('test-run', sink);
+    const page = new MockPage('https://example.com') as any;
+    const runtime = {} as AgentRuntime;
+
+    const spy = jest.spyOn(AgentRuntime, 'fromPlaywrightPage').mockReturnValue(runtime);
+
+    const dbg = SentienceDebugger.attach(page, tracer, { apiKey: 'sk', apiUrl: 'https://api' });
+
+    expect(spy).toHaveBeenCalledWith(page, tracer, { apiKey: 'sk', apiUrl: 'https://api' });
+    expect(dbg.runtime).toBe(runtime);
+
+    spy.mockRestore();
+  });
+
+  it('step() calls beginStep and emitStepEnd', async () => {
+    const runtime = {
+      beginStep: jest.fn().mockReturnValue('step-1'),
+      emitStepEnd: jest.fn().mockResolvedValue({}),
+    } as unknown as AgentRuntime;
+
+    const dbg = new SentienceDebugger(runtime);
+
+    await dbg.step('verify-cart', async () => {
+      // no-op
+    });
+
+    expect(runtime.beginStep).toHaveBeenCalledWith('verify-cart', undefined);
+    expect(runtime.emitStepEnd).toHaveBeenCalled();
+  });
+
+  it('check() auto-opens a step if missing', () => {
+    const runtime = {
+      beginStep: jest.fn().mockReturnValue('step-1'),
+      check: jest.fn().mockReturnValue('handle'),
+    } as unknown as AgentRuntime;
+
+    const dbg = new SentienceDebugger(runtime);
+
+    const handle = dbg.check(
+      (_ctx: any) => ({ passed: true, reason: '', details: {} }),
+      'has_cart'
+    );
+
+    expect(runtime.beginStep).toHaveBeenCalledWith('verify:has_cart', undefined);
+    expect(runtime.check).toHaveBeenCalled();
+    expect(handle).toBe('handle');
+  });
+});