281 lines
8.6 KiB
JavaScript
281 lines
8.6 KiB
JavaScript
import test from "node:test";
|
|
import assert from "node:assert/strict";
|
|
import { mkdtemp, writeFile, chmod, mkdir, symlink } from "node:fs/promises";
|
|
import os from "node:os";
|
|
import path from "node:path";
|
|
import { fileURLToPath } from "node:url";
|
|
import {
|
|
buildCuaLaunchArgs,
|
|
detectCuaTargetApp,
|
|
isSubmitLikeObjective,
|
|
runCuaDriverComputerUseTask,
|
|
} from "../scripts/cua-driver-computer-use-runtime.mjs";
|
|
|
|
const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
|
|
|
|
async function createFakeCuaDriver() {
|
|
const dir = await mkdtemp(path.join(os.tmpdir(), "boss-cua-driver-"));
|
|
const command = path.join(dir, "fake-cua-driver.mjs");
|
|
const logPath = path.join(dir, "calls.jsonl");
|
|
await writeFile(
|
|
command,
|
|
`#!/usr/bin/env node
|
|
import { appendFileSync } from "node:fs";
|
|
|
|
const logPath = process.env.FAKE_CUA_LOG_PATH;
|
|
const args = process.argv.slice(2);
|
|
const callIndex = args.indexOf("call");
|
|
const tool = callIndex >= 0 ? args[callIndex + 1] : args[0];
|
|
const rawJson = callIndex >= 0 ? args[callIndex + 2] : args[1];
|
|
let payload = {};
|
|
try { payload = rawJson ? JSON.parse(rawJson) : {}; } catch {}
|
|
appendFileSync(logPath, JSON.stringify({ tool, payload }) + "\\n");
|
|
|
|
if (tool === "launch_app") {
|
|
if (process.env.FAKE_CUA_FAIL_LAUNCH === "true") {
|
|
process.stdout.write(JSON.stringify({
|
|
content: [{ type: "text", text: "Launch failed: app not found" }],
|
|
isError: true,
|
|
}));
|
|
process.exit(1);
|
|
}
|
|
process.stdout.write(JSON.stringify({
|
|
structuredContent: {
|
|
pid: 2468,
|
|
name: payload.name || "Google Chrome",
|
|
bundle_id: payload.bundle_id || "com.google.Chrome",
|
|
windows: [
|
|
{ window_id: 1357, title: "Boss Cua Test", is_on_screen: true, on_current_space: true },
|
|
],
|
|
},
|
|
content: [{ type: "text", text: "✅ Launched app" }],
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
if (tool === "list_apps") {
|
|
process.stdout.write(JSON.stringify({
|
|
structuredContent: {
|
|
apps: [
|
|
{ pid: 2468, name: "Google Chrome", bundle_id: "com.google.Chrome", running: true },
|
|
],
|
|
},
|
|
content: [{ type: "text", text: "✅ Found app" }],
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
if (tool === "list_windows") {
|
|
process.stdout.write(JSON.stringify({
|
|
structuredContent: {
|
|
windows: [
|
|
{ window_id: 1357, title: "Boss Cua Test", is_on_screen: true, on_current_space: true },
|
|
],
|
|
},
|
|
content: [{ type: "text", text: "✅ Found window" }],
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
if (tool === "get_window_state") {
|
|
process.stdout.write(JSON.stringify({
|
|
content: [{ type: "text", text: "✅ Google Chrome — 3 elements, turn 1 + screenshot\\n- [0] AXWindow Boss Cua Test" }],
|
|
structuredContent: { element_count: 3, has_screenshot: true },
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
if (tool === "type_text") {
|
|
process.stdout.write(JSON.stringify({
|
|
content: [{ type: "text", text: "✅ Inserted text" }],
|
|
structuredContent: { ok: true },
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
if (tool === "press_key") {
|
|
process.stdout.write(JSON.stringify({
|
|
content: [{ type: "text", text: "✅ Pressed key" }],
|
|
structuredContent: { ok: true },
|
|
}));
|
|
process.exit(0);
|
|
}
|
|
|
|
process.stderr.write("unknown tool " + tool);
|
|
process.exit(64);
|
|
`,
|
|
"utf8",
|
|
);
|
|
await chmod(command, 0o755);
|
|
return { command, logPath, cwd: dir };
|
|
}
|
|
|
|
test("cua runtime detects common macOS targets", () => {
|
|
assert.equal(detectCuaTargetApp("打开 Chrome 浏览器搜索测试")?.name, "Google Chrome");
|
|
assert.equal(detectCuaTargetApp("打开系统设置看屏幕录制")?.bundleId, "com.apple.systempreferences");
|
|
assert.equal(detectCuaTargetApp("打开 QQ 群 AI开发")?.name, "QQ");
|
|
});
|
|
|
|
test("cua runtime adds about:blank for browser launch when objective has no URL", () => {
|
|
const target = detectCuaTargetApp("打开 Safari");
|
|
const launchArgs = buildCuaLaunchArgs(target, "打开 Safari");
|
|
|
|
assert.equal(launchArgs.bundle_id, "com.apple.Safari");
|
|
assert.deepEqual(launchArgs.urls, ["about:blank"]);
|
|
});
|
|
|
|
test("cua runtime uses the current macOS system settings bundle id", () => {
|
|
const target = detectCuaTargetApp("打开系统设置");
|
|
const launchArgs = buildCuaLaunchArgs(target, "打开系统设置");
|
|
|
|
assert.equal(launchArgs.bundle_id, "com.apple.systempreferences");
|
|
});
|
|
|
|
test("cua runtime launches target app and observes window state through cua-driver", async () => {
|
|
const fake = await createFakeCuaDriver();
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestKind: "desktop_control",
|
|
requestId: "cua-task-1",
|
|
objective: "打开 Chrome 浏览器",
|
|
platform: "macos",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {
|
|
BOSS_CUA_DRIVER_COMMAND: fake.command,
|
|
BOSS_CUA_DRIVER_TIMEOUT_MS: "4000",
|
|
FAKE_CUA_LOG_PATH: fake.logPath,
|
|
},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "completed");
|
|
assert.equal(result.requestId, "cua-task-1");
|
|
assert.equal(result.targetApp, "Google Chrome");
|
|
assert.match(result.replyBody, /已通过 Cua Driver/);
|
|
assert.match(result.executionSummary, /launch_app -> get_window_state/);
|
|
});
|
|
|
|
test("cua runtime falls back to running app windows when launch_app fails", async () => {
|
|
const fake = await createFakeCuaDriver();
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestKind: "desktop_control",
|
|
requestId: "cua-task-running-app",
|
|
objective: "打开 Chrome 浏览器",
|
|
platform: "macos",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {
|
|
BOSS_CUA_DRIVER_COMMAND: fake.command,
|
|
BOSS_CUA_DRIVER_TIMEOUT_MS: "4000",
|
|
FAKE_CUA_LOG_PATH: fake.logPath,
|
|
FAKE_CUA_FAIL_LAUNCH: "true",
|
|
},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "completed");
|
|
assert.match(result.executionSummary, /launch_app_failed -> list_apps -> list_windows -> get_window_state/);
|
|
});
|
|
|
|
test("cua runtime can discover cua-driver from HOME local bin when PATH is empty", async () => {
|
|
const fake = await createFakeCuaDriver();
|
|
const localBin = path.join(fake.cwd, ".local", "bin");
|
|
await mkdir(localBin, { recursive: true });
|
|
await symlink(fake.command, path.join(localBin, "cua-driver"));
|
|
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestKind: "desktop_control",
|
|
requestId: "cua-task-path",
|
|
objective: "打开 Chrome 浏览器",
|
|
platform: "macos",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {
|
|
HOME: fake.cwd,
|
|
PATH: path.dirname(process.execPath),
|
|
BOSS_CUA_DRIVER_COMMAND: "cua-driver",
|
|
BOSS_CUA_DRIVER_TIMEOUT_MS: "4000",
|
|
FAKE_CUA_LOG_PATH: fake.logPath,
|
|
},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "completed");
|
|
});
|
|
|
|
test("cua runtime types quoted text but does not submit without explicit submit allowance", async () => {
|
|
const fake = await createFakeCuaDriver();
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestKind: "desktop_control",
|
|
requestId: "cua-task-typing",
|
|
objective: "打开 QQ 输入“我是测试”",
|
|
platform: "macos",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {
|
|
BOSS_CUA_DRIVER_COMMAND: fake.command,
|
|
BOSS_CUA_DRIVER_TIMEOUT_MS: "4000",
|
|
FAKE_CUA_LOG_PATH: fake.logPath,
|
|
},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "completed");
|
|
assert.match(result.executionSummary, /type_text/);
|
|
});
|
|
|
|
test("cua runtime asks for confirmation before send-like objectives", async () => {
|
|
assert.equal(isSubmitLikeObjective("在 QQ 里输入“你好”并发送"), true);
|
|
const fake = await createFakeCuaDriver();
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestKind: "desktop_control",
|
|
requestId: "cua-task-send",
|
|
objective: "在 QQ 里输入“你好”并发送",
|
|
platform: "macos",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {
|
|
BOSS_CUA_DRIVER_COMMAND: fake.command,
|
|
FAKE_CUA_LOG_PATH: fake.logPath,
|
|
},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "needs_user_action");
|
|
assert.equal(result.kind, "desktop_submit_confirmation_required");
|
|
assert.deepEqual(result.availableActions, ["allow_once", "deny"]);
|
|
});
|
|
|
|
test("cua runtime rejects non-mac platforms", async () => {
|
|
const result = await runCuaDriverComputerUseTask(
|
|
{
|
|
requestId: "cua-task-windows",
|
|
objective: "打开系统设置",
|
|
platform: "windows",
|
|
provider: "cua-driver-computer-use",
|
|
},
|
|
{
|
|
env: {},
|
|
cwd: repoRoot,
|
|
},
|
|
);
|
|
|
|
assert.equal(result.status, "failed");
|
|
assert.equal(result.error, "UNSUPPORTED_CONTROL_PLATFORM");
|
|
});
|