codex-chief-of-staff/test/app-server-client.test.ts

import assert from "node:assert/strict";
import test from "node:test";

import { AppServerClient } from "../src/app-server-client.js";
import type { ReadMethod, StatusChangedEvent } from "../src/domain.js";
import { FakeAppServer, rawThread, successfulHandshake } from "./fake-app-server.js";

function clientFor(server: FakeAppServer, requestTimeoutMs = 100): AppServerClient {
  return new AppServerClient({
    url: "ws://127.0.0.1:4500",
    requestTimeoutMs,
    connectTimeoutMs: requestTimeoutMs,
    webSocketFactory: server.factory,
  });
}

test("connection completes initialize request and initialized notification", async () => {
  const server = new FakeAppServer(successfulHandshake());
  const client = clientFor(server);
  await client.connect();
  assert.deepEqual(server.frames.map((frame) => frame.method), ["initialize", "initialized"]);
  assert.equal(client.connected, true);
  client.close();
});

test("initialization exposes upstream protocol errors", async () => {
  const server = new FakeAppServer((request, fake) => {
    if (request.method === "initialize") fake.fail(request, -32000, "incompatible client");
  });
  await assert.rejects(clientFor(server).connect(), /incompatible client/);
});

test("initialization times out when the server does not respond", async () => {
  const server = new FakeAppServer(() => undefined);
  await assert.rejects(clientFor(server, 20).connect(), /request timed out: initialize/);
});

test("malformed initialization response is rejected", async () => {
  const server = new FakeAppServer((request, fake) => {
    if (request.method === "initialize") fake.socket?.deliver({ id: request.id });
  });
  await assert.rejects(clientFor(server).connect(), /Malformed response for initialize/);
});

test("runtime allowlist rejects mutating method before sending a frame", async () => {
  const server = new FakeAppServer(successfulHandshake());
  const client = clientFor(server);
  await client.connect();
  const before = server.frames.length;
  const callWithString = client.request.bind(client) as unknown as (
    method: string,
    params: unknown,
  ) => Promise<unknown>;
  assert.throws(() => callWithString("thread/archive", {}), /not allowlisted/);
  assert.equal(server.frames.length, before);
  const compileTimeMethod: ReadMethod = "thread/list";
  assert.equal(compileTimeMethod, "thread/list");
  client.close();
});

test("thread and loaded inventories follow every cursor page", async () => {
  const server = new FakeAppServer(successfulHandshake((request, fake) => {
    const params = request.params as Record<string, unknown>;
    if (request.method === "thread/list") {
      fake.respond(request, params.cursor === null
        ? { data: [rawThread("a")], nextCursor: "next", backwardsCursor: null }
        : { data: [rawThread("b")], nextCursor: null, backwardsCursor: null });
    } else if (request.method === "thread/loaded/list") {
      fake.respond(request, params.cursor === null
        ? { data: ["a"], nextCursor: "loaded-next" }
        : { data: ["b"], nextCursor: null });
    }
  }));
  const client = clientFor(server);
  await client.connect();
  assert.deepEqual((await client.listAllThreads()).map((thread) => thread.id), ["a", "b"]);
  assert.deepEqual([...await client.listLoadedThreadIds()], ["a", "b"]);
  client.close();
});

test("explicit inspection covers known, unknown, oversized, and malformed responses", async () => {
  const server = new FakeAppServer(successfulHandshake((request, fake) => {
    const params = request.params as Record<string, unknown>;
    if (request.method !== "thread/read") return;
    if (params.threadId === "missing") {
      fake.fail(request, -32602, "thread not found");
      return;
    }
    if (params.threadId === "malformed") {
      fake.respond(request, { thread: { id: "malformed" } });
      return;
    }
    fake.respond(request, {
      thread: rawThread("known", {
        turns: [{
          id: "turn",
          items: [
            { id: "user", type: "userMessage", content: [{ type: "text", text: "old" }] },
            { id: "agent", type: "agentMessage", text: "0123456789" },
          ],
        }],
      }),
    });
  }));
  const client = clientFor(server);
  await client.connect();
  const known = await client.inspectThread("known", { maxItems: 1, maxCharacters: 4 });
  assert.deepEqual(known.context.items.map((item) => item.text), ["6789"]);
  assert.equal(known.context.truncated, true);
  await assert.rejects(client.inspectThread("missing", { maxItems: 1, maxCharacters: 4 }), /not found/);
  await assert.rejects(client.inspectThread("malformed", { maxItems: 1, maxCharacters: 4 }), /Malformed/);
  client.close();
});

test("status notifications preserve active, idle, notLoaded, and active flags", async () => {
  const server = new FakeAppServer(successfulHandshake());
  const client = clientFor(server);
  const events: StatusChangedEvent[] = [];
  client.onStatusChanged((event) => events.push(event));
  await client.connect();
  server.notify("thread/status/changed", {
    threadId: "a",
    status: { type: "active", activeFlags: ["waitingOnUserInput"] },
  });
  server.notify("thread/status/changed", { threadId: "a", status: { type: "idle" } });
  server.notify("thread/status/changed", { threadId: "a", status: { type: "notLoaded" } });
  assert.deepEqual(events.map((event) => event.status.type), ["active", "idle", "notLoaded"]);
  assert.deepEqual(events[0]?.status.activeFlags, ["waitingOnUserInput"]);
  client.close();
});