test(agent): prompt eval with fake model and real-model comparison harness

Co-Authored-By: Claude Opus 5.5 <[email protected]>

apps/server/test/e2e/harness.ts

@@ -20,6 +20,7 @@ import { type Bootstrapped, bootstrap } from "../../src/context.ts";
 import { createApp } from "../../src/http/app.ts";
 import { EventBus } from "../../src/http/sse.ts";
 import { createLogger } from "../../src/logger.ts";
+import type { AgentModuleOptions } from "../../src/modules/agent/runner.ts";
 import type { BrowserModuleOptions } from "../../src/modules/browser/service.ts";
 import { setDefaultUpstream } from "../../src/modules/pi/store.ts";
 import type { ServerModule } from "../../src/modules/types.ts";
@@ -84,6 +85,8 @@ export interface E2EOptions {
   webSearch?: WebSearchModuleOptions;
   /** 浏览器模块注入(E14) */
   browser?: BrowserModuleOptions;
+  /** 智能体模块的额外参数(E16:替换 / 捕获系统提示) */
+  agent?: Pick<AgentModuleOptions, "systemPrompt">;
 }
 
 export async function waitFor<T>(
@@ -140,6 +143,7 @@ export async function createE2E(name: string, opts: E2EOptions = {}): Promise<E2
     ...(opts.fakeScripts ? { fakeScripts: opts.fakeScripts } : {}),
     ...(opts.webSearch ? { webSearch: opts.webSearch } : {}),
     ...(opts.browser ? { browser: opts.browser } : {}),
+    ...(opts.agent ? { agent: opts.agent } : {}),
   });
   const booted = await bootstrap({
     role: "all",