Files
Genarrative/apps/ai-game-creator-shell/tests/appSurface/harness.ts
T
kdletters 485ed50b26
Project CI / AI game creator shell Rust smoke (push) Has been cancelled
Project CI / AI game creator shell Rust crates (push) Has been cancelled
Project CI / Backend tests (push) Has been cancelled
Project CI / Native shell tests (push) Has been cancelled
Project CI / Frontend tests (push) Has been cancelled
Project CI / Repository checks (push) Has been cancelled
Project CI / AI game creator shell web tests (push) Has been cancelled
Project CI / AI game creator shell Rust shard 2/4 (push) Has been cancelled
Project CI / AI game creator shell Rust shard 1/4 (push) Has been cancelled
Project CI / AI game creator shell Rust shard 3/4 (push) Has been cancelled
Project CI / AI game creator shell Rust shard 4/4 (push) Has been cancelled
AGC 七项交付效率优化并升级捆绑 Codex 到 0.155.1 (#439)
## 背景

净月潭案例(2026-09-19 21:53 → 09-20 01:26,3 小时 33 分)的问题不是模型慢,而是环境未就绪、验收靠模型自述、预算只约束单个工具入口、工具调用被 SDK 全局串行闸门卡住。本 PR 落地确认后的七项交付效率合同,并把捆绑 Codex 升到当前 npm latest。

## 结果

- 新建 Web 游戏在正式生成前由宿主自动预检:捆绑 Node/npm、真实 Vite 构建、受限浏览器桌面/移动截图;失败不启动生成或付费素材。
- 首个副作用前冻结交付合同,宿主保存权威证据与预算;证据齐全后先封口、排空并取得执行器完整退出证明,再产出交付报告,模型回复不再当作验收。
- 原生 shell、内置浏览器与托管命令、第三方 MCP 共用同一执行许可和累计执行时间;`validation.maxRuns` 改为执行/返修批次语义,另设 `maxTurnSeconds` 墙钟上限。
- 所有工具可并发:移除 SDK 全局串行的 `apply_patch`/`update_plan` 注册,改由宿主 `agc_apply_patch`(官方 parser + 当前回合写许可 + 受控进程树 + 短项目事务)和 `agc_update_plan` 提供等价能力;真实请求目录里已无串行注册,长 MCP 与补丁/计划实测在同一响应内重叠。
- 付费提交与本地写入绑定原回合原租约:容量与同动作锁等待可取消,封口、取消或预算耗尽后零新增提交;已越过提交边界的请求保留 operation ID 走 GET 对账,不自动重放。
- 新增请求与工具分段计时、有界并行批读和首轮上下文预取,未知耗时不补零。
- 捆绑 Codex 0.147.0 → 0.155.1:固定版本收敛到 `build_support/codex_bundle.rs` 单一声明,vendor 解析源码按 `rust-v0.155.1` 逐字节重取并更新 UPSTREAM 证据,适配 0.155 统一 exec(`exec_command`/`write_stdin`);macOS 侧车最小系统版本仍为 15.0。

## 验证

- 生产 CLI → 捆绑 Codex 0.155.1 → 本地 Responses/MCP 夹具 9/9:补丁、完成收尾、批次、只读/可写 MCP 并发、原生命令并发、原生资源并发、统一 exec 会话、期限终止(真实重叠 775 / 764 / 999 ms)。
- 真实模型目录 2/2;vendor 上游库 111 项;宿主定向与回归 106 项;真实 Windows 进程与取消 21 项;Node 侧门禁 52 项。
- 发行载荷:artifact-only 重建 NSIS,解包验证侧车清单 `codex-cli 0.155.1`、6 个组件哈希、打包后 `bin/codex.exe --version`、Node 运行时 2130 文件与双端 PNG、`--environment-check` ready。
- `check-config`、TypeScript、编码(4991 文件)、文档索引、`git diff --check`、`cargo fmt --check` 全部通过。

## 未覆盖

- 真实陶泥儿登录态 Provider 生成尚未执行(本机 `--llm-status` 返回 `authentication-required`)。
- macOS 侧车只在本机做静态 Mach-O 与清单解析,未在 macOS 上跑 `check-macos-bundle.mjs`。
- 全仓库聚合套件在本机仍因负载敏感的后台 mock-provider 用例而红,与本次改动无关:改动前的旧二进制同样失败,换单线程后相关用例 3/3 通过。

---------

Co-authored-by: kdletters <61648117+kdletters@users.noreply.github.com>
Reviewed-on: http://192.168.35.82/git/GenarrativeAI/Genarrative/pulls/439
2026-09-21 02:29:29 +08:00

1469 lines
46 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { readFileSync } from 'node:fs';
import { resolve } from 'node:path';
import {
act,
cleanup,
fireEvent,
render,
screen,
waitFor,
within,
} from '@testing-library/react';
import React from 'react';
import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
import * as clientApi from '../../src/services/clientApi';
import { resetLlmModelCatalogCacheForTest } from '../../src/services/llmModelCatalog';
import { useLauncherHomeDraftStore } from '../../src/view/home/useHomeDraftStore';
const nativeClipboardMock = vi.hoisted(() => ({
text: '',
}));
const TEST_LOCAL_PROJECT_PATH = '/tmp/genarrative-ai-game-draft';
const originalCreateObjectUrl = Object.getOwnPropertyDescriptor(
URL,
'createObjectURL',
);
const originalRevokeObjectUrl = Object.getOwnPropertyDescriptor(
URL,
'revokeObjectURL',
);
const originalPointerEvent = Object.getOwnPropertyDescriptor(
window,
'PointerEvent',
);
const originalClipboardEvent = Object.getOwnPropertyDescriptor(
window,
'ClipboardEvent',
);
const originalDragEvent = Object.getOwnPropertyDescriptor(window, 'DragEvent');
const originalRangeGetBoundingClientRect = Object.getOwnPropertyDescriptor(
Range.prototype,
'getBoundingClientRect',
);
const originalIntersectionObserver = Object.getOwnPropertyDescriptor(
window,
'IntersectionObserver',
);
const originalResizeObserver = Object.getOwnPropertyDescriptor(
window,
'ResizeObserver',
);
class TestPointerEvent extends MouseEvent {
readonly pointerId: number;
constructor(
type: string,
init: MouseEventInit & { pointerId?: number } = {},
) {
super(type, init);
this.pointerId = init.pointerId ?? 0;
}
}
// jsdom 没有实现 DragEvent / ClipboardEvent,而 Lexical 在粘贴与拖放处理里用
// `instanceof DragEvent` / `instanceof ClipboardEvent` 判定事件来源,缺失会直接抛
// ReferenceError,粘贴通路根本进不去编辑器。这里只补最小的构造器,让事件能带上
// `clipboardData` 并被 Lexical 认出来,不模拟真实浏览器语义。
class TestClipboardEvent extends Event {}
class TestDragEvent extends MouseEvent {}
// jsdom 的 Range 也没有 getBoundingClientRect,Lexical 计算选区矩形时会抛 TypeError。
function emptyRangeRect(): DOMRect {
return {
x: 0,
y: 0,
top: 0,
left: 0,
right: 0,
bottom: 0,
width: 0,
height: 0,
toJSON: () => ({}),
} as DOMRect;
}
function createMemoryStorage(): Storage {
const values = new Map<string, string>();
return {
get length() {
return values.size;
},
clear() {
values.clear();
},
getItem(key) {
return values.get(key) ?? null;
},
key(index) {
return Array.from(values.keys())[index] ?? null;
},
removeItem(key) {
values.delete(key);
},
setItem(key, value) {
values.set(key, String(value));
},
};
}
Object.defineProperty(window, 'localStorage', {
configurable: true,
value: createMemoryStorage(),
});
vi.mock('@tauri-apps/plugin-clipboard-manager', () => ({
readImage: vi.fn(async () => {
throw new Error('no native clipboard image');
}),
readText: vi.fn(async () => nativeClipboardMock.text),
}));
import type { AuthUser } from '../../../../packages/shared/src/contracts/auth';
import {
createGameCreationAppManifest,
createGameCreationAppSeedTasks,
GAME_CREATION_AGENT_RUN_SCHEMA_VERSION,
type GameCreationAgentRunTrace,
} from '../../../../packages/shared/src/contracts/gameCreationApp';
import {
App,
AuthenticatedClient,
deriveAgentStatusCards,
WorkspaceLauncher,
} from '../../src/App';
import { projectNameFromPath } from '../../src/features/agent-runtime/model';
import ProjectDevelopmentView from '../../src/view/project-development';
const testAuthUser: AuthUser = {
id: 'user-test',
publicUserCode: 'tn-test',
displayName: '测试用户',
avatarUrl: null,
phoneNumber: null,
phoneNumberMasked: '138****0000',
loginMethod: 'password',
bindingStatus: 'active',
wechatBound: false,
wechatDisplayName: null,
wechatAccount: null,
};
function renderAppAt(path: string) {
window.history.pushState({}, '', path);
render(React.createElement(App));
const params = new URLSearchParams(window.location.search);
if (params.has('dev') && !params.has('projectPath')) {
const projectPathInput = screen.queryByLabelText('本地项目目录');
if (projectPathInput) {
fireEvent.change(projectPathInput, {
target: { value: TEST_LOCAL_PROJECT_PATH },
});
}
}
}
function renderLauncherAt(
path: string,
initialView: 'home' | 'agent-chat' = 'home',
strictMode = false,
) {
window.history.pushState({}, '', path);
const launcher = React.createElement(WorkspaceLauncher, {
currentUser: testAuthUser,
initialView,
onLogout: vi.fn(),
});
render(
strictMode
? React.createElement(React.StrictMode, null, launcher)
: launcher,
);
}
function renderLauncherAgentChatAt(path: string) {
renderLauncherAt(path, 'agent-chat');
}
function selectDeveloperAgentChatMode(mode: 'run' | 'chat' | 'goal') {
fireEvent.click(
screen.getByRole('button', {
name: mode === 'run' ? '执行' : mode === 'chat' ? '聊天' : '目标',
}),
);
}
function renderLauncherProjectsAt(path: string) {
renderLauncherAt(path);
fireEvent.click(screen.getByRole('button', { name: '项目组' }));
}
function pickProjectFromLauncher(
projectPath: string,
action: '打开项目' | '新建项目' = '打开项目',
) {
const invoke = window.__TAURI__?.core?.invoke;
if (!invoke) {
throw new Error('Tauri invoke is not configured');
}
window.__TAURI__ = {
...window.__TAURI__,
core: {
...window.__TAURI__.core,
invoke: async (command, args) =>
command === 'pick_local_project_directory'
? projectPath
: invoke(command, args),
},
};
fireEvent.click(screen.getByRole('button', { name: action }));
}
function isNativeTextControl(
element: HTMLElement,
): element is HTMLInputElement | HTMLTextAreaElement {
return element.tagName === 'INPUT' || element.tagName === 'TEXTAREA';
}
// Lexical 的编辑器状态提交排在微任务里(外部 value 变化、粘贴、删除都走这条路径),
// 读取或提交前先让 React 追平编辑器内容,否则会读到上一帧的文本/禁用状态。
// 这里让出一个宏任务而不是用 act(async),因为 act 在编辑器持续自更新的用例里会长时间不收敛。
async function settleComposer() {
await new Promise((resolve) => {
setTimeout(resolve, 0);
});
}
// 读取输入区文本:原生 input/textarea 读 value,Lexical contenteditable 读 textContent。
async function composerValue(element: HTMLElement) {
await settleComposer();
return isNativeTextControl(element)
? element.value
: (element.textContent ?? '');
}
type ComposerScope = { getByLabelText: (label: string) => HTMLElement };
async function composerText(label = '创作想法', scope: ComposerScope = screen) {
return composerValue(scope.getByLabelText(label));
}
async function composerDisabled(
label = '创作想法',
scope: ComposerScope = screen,
) {
await settleComposer();
const element = scope.getByLabelText(label);
if (isNativeTextControl(element)) {
return element.disabled;
}
const wrapper = element.closest('.resource-reference-input');
return (
wrapper?.getAttribute('data-disabled') === 'true' ||
element.getAttribute('contenteditable') !== 'true'
);
}
// 把整段内容选进 DOM 选区,并让 Lexical 吸收它:jsdom 不会为程序化选区派发
// selectionchange,编辑器内部 selection 不更新的话,粘贴会插在旧内容旁边而不是覆盖。
function selectComposerContents(element: HTMLElement) {
element.focus();
const range = document.createRange();
range.selectNodeContents(element);
const selection = window.getSelection();
selection?.removeAllRanges();
selection?.addRange(range);
document.dispatchEvent(new Event('selectionchange'));
}
function clearComposerContents(element: HTMLElement) {
fireEvent.keyDown(element, { key: 'Delete', code: 'Delete' });
}
// 写完内容后把光标收回到末尾,模拟真实输入后的选区;否则残留的“全选”会改变
// Shift+Enter 等按键的后续语义。
function collapseComposerSelectionToEnd(element: HTMLElement) {
const range = document.createRange();
range.selectNodeContents(element);
range.collapse(false);
const selection = window.getSelection();
selection?.removeAllRanges();
selection?.addRange(range);
document.dispatchEvent(new Event('selectionchange'));
}
function pasteComposerText(element: HTMLElement, value: string) {
fireEvent.paste(element, {
clipboardData: {
getData: (type: string) => (type === 'text/plain' ? value : ''),
files: [],
items: [],
types: ['text/plain'],
},
});
}
// 向输入区写入文本,保证写入结果等于 value。原生控件用 change 事件;Lexical
// contenteditable 先全选、清空,再走产品真实的 paste 通路写入。
async function setComposerText(element: HTMLElement, value: string) {
if (isNativeTextControl(element)) {
fireEvent.change(element, { target: { value } });
return;
}
selectComposerContents(element);
await settleComposer();
clearComposerContents(element);
await settleComposer();
if (value.length > 0) {
pasteComposerText(element, value);
await settleComposer();
}
collapseComposerSelectionToEnd(element);
await settleComposer();
}
async function submitChat(value: string) {
await setComposerText(screen.getByLabelText('创作想法'), value);
fireEvent.click(screen.getByRole('button', { name: '发送' }));
}
// 同一张资源卡上的按钮变少了:卡片本体只有「选中资源」与媒体播放钮两个(原先右上角
// 那个 @ 引用圆钮已挪进选中工具条)。选中按钮的可访问名模板固定为
// `选中资源:<分类标签> <资源文件名>`;分类标签可能自带空格(例如 `UI 交互`),
// 这里按模板做精确匹配。
function escapeRegExpLiteral(value: string) {
return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
}
function resourceSelectButtonName(label: string) {
return new RegExp(`^选中资源:.+ ${escapeRegExpLiteral(label)}$`);
}
function getResourceSelectButton(label: string) {
return screen.getByRole('button', { name: resourceSelectButtonName(label) });
}
function findResourceSelectButton(
label: string,
options?: { timeout?: number; interval?: number },
) {
return screen.findByRole(
'button',
{ name: resourceSelectButtonName(label) },
options,
);
}
function queryResourceSelectButton(label: string) {
return screen.queryByRole('button', {
name: resourceSelectButtonName(label),
});
}
/**
* 叫出画布唯一的筛选面板(右下角 Dock 的放大镜按钮)并返回关键词输入框。
*
* 面板只在打开时存在于 DOM:用例必须走产品入口打开它,`getByLabelText` 在面板没打开时
* 会直接抛错,所以「输入框在真实 UI 里点不到」那类假绿不会再出现。关键词 / 所在区域 /
* 自定义标签三个字段都在这一次打开的面板里(`Ctrl/Cmd+F` 叫出的是同一个)。
*/
function openResourceFilterPanel() {
fireEvent.click(screen.getByRole('button', { name: '搜索资源' }));
return screen.getByLabelText('查找素材') as HTMLInputElement;
}
function emptyProjectPolicy() {
return {
path: '.agent/policy.json',
policy: {
deniedCommands: [],
confirmCommands: [],
},
};
}
const roleAgentMockReply =
'专业 Agent 已完成本轮判断:先补齐角色规范图、验收口径和后续生成条件。';
function mockRoleAgentReply() {
return roleAgentMockReply;
}
function projectSupervisorResponseStream({
runId,
sequence,
accumulatedText,
status = 'streaming',
appliedSteerCursor = 0,
responseRevision = 0,
loopIteration = 1,
overrides = {},
}: {
runId: string;
sequence: number;
accumulatedText: string;
status?: 'streaming' | 'ready' | 'committed' | 'discarded' | 'failed';
appliedSteerCursor?: number;
responseRevision?: number;
loopIteration?: number;
overrides?: Record<string, unknown>;
}) {
return {
schemaVersion: 'game-creator-runtime-response-stream.v1',
agentId: 'project-supervisor',
taskId: 'project-supervisor',
sessionId: 'supervisor-session-active',
runId,
requestKind: 'final-reply',
requestSlot: `final-reply-loop-${loopIteration}-revision-${responseRevision}`,
appliedSteerCursor,
responseRevision,
sequence,
status,
accumulatedText,
finishReason: status === 'ready' || status === 'committed' ? 'stop' : null,
startedAt: 6000,
updatedAt: 6000 + sequence,
...overrides,
};
}
function agentRuntimeUserInputRequest({
agentId,
sessionId,
runId,
requestId = 'request-user-input-1',
actionId = 'action-user-input-1',
responseId = null,
status = 'pending',
}: {
agentId: string;
sessionId: string;
runId: string;
requestId?: string;
actionId?: string;
responseId?: string | null;
status?: 'pending' | 'answer-prepared';
}) {
return {
schemaVersion: 'game-creator-runtime-user-input.v1',
requestId,
agentId,
taskId: agentId,
sessionId,
runId,
actionId,
status,
questions: [
{
id: 'visual_direction',
header: '美术方向',
question: '首版角色规范图采用哪种美术方向?',
options: [
{
label: '像素风',
description: '优先验证轮廓与动作可读性。',
},
{
label: '手绘风',
description: '优先验证角色气质与细节。',
},
],
},
],
allowFreeform: true,
responseId,
requestedAt: 6000,
updatedAt: 6000,
};
}
/**
* 运行态事件里的条目身份:只有 `item.started` / `item.completed` 带条目。
*
* `item.delta` 只带 itemId 不带条目,属于瞬时事件,不参与 bootstrap 回放。
*/
function directThreadEventItemId(
event: Record<string, unknown>,
): string | null {
if (event.type !== 'item.started' && event.type !== 'item.completed') {
return null;
}
const item = event.item;
if (!item || typeof item !== 'object') return null;
const itemId = (item as Record<string, unknown>).itemId;
return typeof itemId === 'string' && itemId ? itemId : null;
}
function createProjectSupervisorRuntimeHarness({
projectPath = '/tmp/authorized-game',
sessionId = 'supervisor-session-active',
initialSessionExists = true,
projectMessages = [],
supervisorMessages = [],
initialRuntime,
initialResponseStream = null,
initialProjectRevision = 0,
runtimeMapLoader,
expectedRunProfile = 'autonomous-game-build',
designAgentView = null,
designAgentContinueView = null,
designWorkspaceFiles = [],
}: {
projectPath?: string;
sessionId?: string;
initialSessionExists?: boolean;
projectMessages?: Array<Record<string, unknown>>;
supervisorMessages?: Array<Record<string, unknown>>;
initialRuntime?: Record<string, unknown>;
initialResponseStream?: Record<string, unknown> | null;
initialProjectRevision?: number;
runtimeMapLoader?: () => Promise<Array<Record<string, unknown>>>;
expectedRunProfile?: 'standard' | 'autonomous-game-build';
designAgentView?: Record<string, unknown> | null;
designAgentContinueView?: Record<string, unknown> | null;
designWorkspaceFiles?: Array<Record<string, unknown>>;
} = {}) {
const manifest = createGameCreationAppManifest(
'local-project-draft',
'未命名游戏原型',
);
let messageSequence = 0;
let selectedModelId = 'quality';
let selectedModelIsDefault = false;
let reasoningEffort = 'high';
let steerSequence = 0;
let sessionExists = initialSessionExists;
let currentProjectRevision = initialProjectRevision;
const currentProjectMessages = [...projectMessages];
const currentSupervisorMessages = [...supervisorMessages];
const runtimeState = (
overrides: Record<string, unknown> = {},
): Record<string, unknown> => ({
schemaVersion: 'game-creator-agent-runtime.v1',
agentId: 'project-supervisor',
taskId: 'project-supervisor',
sessionId,
runId: 'supervisor-idle',
source: 'project-supervisor',
status: 'idle',
phase: 'idle',
currentTask: '',
currentGoal: '',
currentAction: '',
waitingOn: '',
nextStep: '',
plan: [],
observations: [],
allowedTools: [],
pendingToolAction: null,
lastResponse: null,
error: null,
updatedAt: 1000,
...overrides,
});
let currentRuntime = runtimeState(initialRuntime);
let currentResponseStream = initialResponseStream;
let runtimeReader: (() => Promise<Record<string, unknown>>) | null = null;
let confirmRuntime: Record<string, unknown> | null = null;
let rejectRuntime: Record<string, unknown> | null = null;
let answerRuntime: Record<string, unknown> | null = null;
let answerFailuresRemaining = 0;
let currentDesignAgentView = designAgentView;
const currentDesignContinueView = designAgentContinueView;
const currentDesignWorkspaceFiles = [...designWorkspaceFiles];
let runtimeUpdateHandler:
| ((event: {
payload: {
projectPath: string;
agentId: string;
runId: string;
status: string;
phase: string;
manifestInvalidated: boolean;
runtime: Record<string, unknown>;
};
}) => void)
| null = null;
let manifestInvalidatedHandler:
| ((event: {
payload: {
projectPath: string;
agentId: string;
};
}) => void)
| null = null;
let progressHandler:
| ((event: {
payload: {
projectPath: string;
stage: string;
message: string;
};
}) => void)
| null = null;
let designAgentUpdateHandler:
| ((event: { payload: Record<string, unknown> }) => void)
| null = null;
let directThreadNotifyHandler:
| ((event: { payload: { subscriptionId: string } }) => void)
| null = null;
let directThreadSubscriptionId: string | null = null;
let directThreadSubscriptionSequence = 0;
// 未消费的运行态事件队列:`subscribe` 的 bootstrap 与 `consume` 都从这里取,
// 与 Rust 侧"游标在队尾、事件按序下发"的语义一致。
let pendingDirectThreadEvents: Array<Record<string, unknown>> = [];
let directThreadHistoryItems: Array<Record<string, unknown>> = [];
let directThreadLastCompletedItemId: string | null = null;
// 运行态事件里只有一个"最新回合是否在跑"的布尔:与 Thread Manager 只保留一条生命周期
// 锚点一致,重复 `turn.started` 在生产里不会出现。
let directThreadTurnRunning = false;
// bootstrap 的回放规则与 Rust 侧 `is_bootstrap_event` 对齐:只补"这一刻还没结束的条目"
// 与最新一条生命周期锚点;已完成条目、增量正文这类瞬时事件都不回放。
const directThreadActiveItemIds = new Set<string>();
const directThreadPendingEventSeqs = new Map<
Record<string, unknown>,
number
>();
let directThreadEventSequence = 0;
let directThreadLifecycleAnchor: {
seq: number;
event: Record<string, unknown>;
} | null = null;
const conversationRecord = (
role: 'user' | 'assistant',
content: string,
messageId: string,
) => ({
schemaVersion: 'game-creator-conversation.v1',
role,
content,
agentId: 'project-supervisor',
messageId,
updatedAt: 2000 + ++messageSequence,
});
const runtimeResult = (
state = currentRuntime,
responseStream = state === currentRuntime ? currentResponseStream : null,
) => ({
state,
sessionPath: `${projectPath}/.agent/runtime/agents/${String(state.agentId)}.json`,
eventPath: `${projectPath}/.agent/runtime/events/${String(state.agentId)}.jsonl`,
taskPath: `${projectPath}/.agent/runtime/tasks/${String(state.agentId)}.jsonl`,
taskQueue: {
total: state.status === 'idle' ? 0 : 1,
pending: state.status === 'pending' ? 1 : 0,
running: state.status === 'running' ? 1 : 0,
waitingForConfirmation:
state.status === 'waiting-for-confirmation' ? 1 : 0,
waitingForUserInput: state.status === 'waiting-for-user-input' ? 1 : 0,
completed: state.status === 'completed' ? 1 : 0,
failed: state.status === 'failed' ? 1 : 0,
latestRunId: String(state.runId),
updatedAt: Number(state.updatedAt),
},
recentEvents: state.recentEvents ?? [],
recentTasks: [],
responseStream,
userInputRequest: state.userInputRequest ?? null,
});
const invoke = vi.fn(
async (command: string, args?: Record<string, unknown>) => {
if (command === 'preflight_web_game_creation') return { status: 'ready' };
if (
command === 'read_game_creator_app_config' ||
command === 'select_game_creator_model' ||
command === 'select_game_creator_reasoning_effort'
) {
if (command === 'select_game_creator_model') {
selectedModelId = String(args?.modelId);
selectedModelIsDefault = args?.isDefault === true;
}
if (command === 'select_game_creator_reasoning_effort') {
reasoningEffort = String(args?.effort);
}
return {
path: '/tmp/test-game-creator-config.json',
config: {
schemaVersion: 'game-creator-config.v2',
agentMode: 'codex_app_server',
selectedModelId,
selectedModelIsDefault,
llm: {
apiKey: '',
baseUrl: '',
model: 'quality',
apiKind: 'openai_responses',
reasoningEffort,
stream: true,
webSearchEnabled: true,
contextWindowTokens: 128000,
autoCompactTokenLimit: 64000,
toolOutputTokenLimit: 12000,
requestTimeoutMs: 180000,
maxRetries: 2,
retryBackoffMs: 500,
},
agentLlm: {},
editorApi: { baseUrl: 'https://dev.genarrative.world', apiKey: '' },
},
};
}
if (command === 'append_local_permission_log') {
return {};
}
if (command === 'read_project_permission_policy') {
return emptyProjectPolicy();
}
if (command === 'get_local_game_project_revision') {
return { revision: currentProjectRevision };
}
if (command === 'init_local_game_project') {
const requestedProjectPath = String(args?.projectPath ?? '');
return {
projectPath: requestedProjectPath,
manifestPath: `${requestedProjectPath}/.agent/manifest.json`,
manifest,
};
}
if (command === 'resume_game_creator_agent_runtime_tasks') {
return [runtimeResult()];
}
if (command === 'list_game_creator_agent_sessions') {
return {
path: `${projectPath}/.agent/conversations/agents/project-supervisor/sessions.json`,
agentId: 'project-supervisor',
activeSessionId: sessionExists ? sessionId : null,
sessions: sessionExists
? [
{
sessionId,
title: '项目总控',
createdAt: 1000,
updatedAt: 2000,
archivedAt: null,
messageCount: currentSupervisorMessages.length,
legacy: false,
},
]
: [],
};
}
if (command === 'create_game_creator_agent_session') {
sessionExists = true;
return {
path: `${projectPath}/.agent/conversations/agents/project-supervisor/sessions.json`,
agentId: 'project-supervisor',
activeSessionId: sessionId,
sessions: [
{
sessionId,
title: String(args?.title ?? ''),
createdAt: 1000,
updatedAt: 2000,
archivedAt: null,
messageCount: currentSupervisorMessages.length,
legacy: false,
},
],
};
}
if (command === 'read_local_conversation') {
if (args?.agentId === null) {
return {
path: `${projectPath}/.agent/conversations/project.jsonl`,
agentId: null,
sessionId: null,
messages: [...currentProjectMessages],
};
}
if (args?.agentId === 'project-supervisor') {
if (args?.sessionId !== sessionId) {
throw new Error(`unexpected supervisor session ${args?.sessionId}`);
}
return {
path: `${projectPath}/.agent/conversations/agents/project-supervisor/sessions/${sessionId}.jsonl`,
agentId: 'project-supervisor',
sessionId,
messages: [...currentSupervisorMessages],
};
}
}
if (command === 'append_local_conversation_message') {
if (args?.agentId !== null) {
throw new Error('React must not append Supervisor conversation');
}
const message = args?.message as {
role: 'user' | 'assistant';
content: string;
agentId: null;
updatedAt?: number;
};
currentProjectMessages.push({
schemaVersion: 'game-creator-conversation.v1',
...message,
...(args?.messageId ? { messageId: String(args.messageId) } : {}),
updatedAt:
typeof message.updatedAt === 'number'
? message.updatedAt
: 1500 + ++messageSequence,
});
return {
path: `${projectPath}/.agent/conversations/project.jsonl`,
agentId: null,
sessionId: null,
messages: [...currentProjectMessages],
};
}
if (command === 'read_game_creator_agent_runtime') {
if (
args?.agentId !== 'project-supervisor' ||
args?.sessionId !== sessionId
) {
throw new Error('unexpected Project Supervisor runtime identity');
}
if (runtimeReader) {
return runtimeReader();
}
return runtimeResult();
}
if (command === 'read_game_creator_agent_runtimes') {
const states = runtimeMapLoader ? await runtimeMapLoader() : [];
return states.map((state) => runtimeResult(state));
}
if (command === 'subscribe_direct_project_thread') {
directThreadSubscriptionSequence += 1;
directThreadSubscriptionId = `direct-thread-${directThreadSubscriptionSequence}`;
const pending = pendingDirectThreadEvents;
pendingDirectThreadEvents = [];
// 游标落在队尾:只有"还没结束的条目"与最新生命周期锚点作为 bootstrap 回放,
// 已完成条目与增量正文不再补发(Rust 侧 `is_bootstrap_event` 的同一口径)。
const replay = pending
.filter((event) => {
const itemId = directThreadEventItemId(event);
return Boolean(itemId && directThreadActiveItemIds.has(itemId));
})
.map((event) => ({
seq: directThreadPendingEventSeqs.get(event) ?? 0,
event,
}));
for (const event of pending) {
directThreadPendingEventSeqs.delete(event);
}
if (
directThreadLifecycleAnchor &&
!replay.some(
(entry) => entry.event === directThreadLifecycleAnchor?.event,
)
) {
replay.push(directThreadLifecycleAnchor);
}
replay.sort((left, right) => left.seq - right.seq);
return {
subscriptionId: directThreadSubscriptionId,
lastCompletedItemId: directThreadLastCompletedItemId,
events: replay.map((entry) => entry.event),
};
}
if (command === 'consume_direct_project_thread') {
if (String(args?.subscriptionId ?? '') !== directThreadSubscriptionId) {
throw new Error('SUBSCRIPTION_EXPIRED');
}
const events = pendingDirectThreadEvents;
pendingDirectThreadEvents = [];
return { events };
}
if (command === 'read_direct_project_history_slice') {
// 生产口径:从文件尾反向取一屏可显示条目——`throughItemId` 是窗口新端边界(含该条,
// 首屏用),`beforeItemId` 是旧端边界(不含该条,翻页用);收满一屏之后再见一条才算
// `hasMore`,`firstItemId` 是本屏最老一条的 itemId。
const requestedLimit = Number(args?.limit ?? 20);
const limit = Math.min(
Math.max(Number.isFinite(requestedLimit) ? requestedLimit : 20, 1),
200,
);
const throughItemId =
typeof args?.throughItemId === 'string' && args.throughItemId
? args.throughItemId
: null;
const beforeItemId =
typeof args?.beforeItemId === 'string' && args.beforeItemId
? args.beforeItemId
: null;
let end = directThreadHistoryItems.length;
if (throughItemId) {
const anchorIndex = directThreadHistoryItems.findIndex(
(item) => String(item.itemId ?? '') === throughItemId,
);
if (anchorIndex < 0) {
throw new Error(
`DirectProject 历史中不存在 item:${throughItemId}`,
);
}
end = anchorIndex + 1;
} else if (beforeItemId) {
const anchorIndex = directThreadHistoryItems.findIndex(
(item) => String(item.itemId ?? '') === beforeItemId,
);
end =
anchorIndex >= 0 ? anchorIndex : directThreadHistoryItems.length;
}
const start = Math.max(0, end - limit);
const window = directThreadHistoryItems.slice(start, end);
const oldest = window[0];
return {
items: [...window],
hasMore: start > 0,
firstItemId:
oldest && typeof oldest.itemId === 'string' ? oldest.itemId : null,
};
}
if (command === 'start_game_creator_supervisor_runtime_task') {
if (args?.runProfile !== expectedRunProfile) {
throw new Error('unexpected Project Supervisor run profile');
}
const runId = String(args?.runId ?? '');
currentSupervisorMessages.push(
conversationRecord(
'user',
String(args?.task ?? ''),
`start-user-${runId}`,
),
);
currentRuntime = runtimeState({
runId,
runProfile: args?.runProfile,
source: args?.source ?? 'project-supervisor',
status: 'running',
phase: 'planning',
currentTask: String(args?.task ?? ''),
currentGoal: String(args?.task ?? ''),
updatedAt: 3000,
});
currentResponseStream = null;
return runtimeResult();
}
if (command === 'steer_game_creator_agent_runtime_task') {
steerSequence += 1;
currentResponseStream = null;
currentSupervisorMessages.push(
conversationRecord(
'user',
String(args?.instruction ?? ''),
`steer-user-${steerSequence}`,
),
);
return {
runtime: runtimeResult(),
steerId: String(args?.steerId ?? ''),
sequence: steerSequence,
status: 'queued',
providerInterrupted: false,
};
}
if (command === 'confirm_game_creator_agent_runtime_task') {
currentResponseStream = null;
currentRuntime =
confirmRuntime ??
runtimeState({
runId: String(currentRuntime.runId),
status: 'running',
phase: 'planning',
updatedAt: 4000,
});
return runtimeResult();
}
if (command === 'reject_game_creator_agent_runtime_task') {
currentResponseStream = null;
currentRuntime =
rejectRuntime ??
runtimeState({
runId: String(currentRuntime.runId),
status: 'running',
phase: 'planning',
updatedAt: 5000,
});
return runtimeResult();
}
if (command === 'answer_game_creator_agent_runtime_user_input') {
if (answerFailuresRemaining > 0) {
answerFailuresRemaining -= 1;
throw new Error('模拟回答提交中断');
}
const answers = args?.answers as Record<string, string>;
currentSupervisorMessages.push(
conversationRecord(
'user',
String(answers?.visual_direction ?? ''),
`answer-user-${String(args?.responseId ?? '')}`,
),
);
currentResponseStream = null;
currentRuntime =
answerRuntime ??
runtimeState({
runId: String(currentRuntime.runId),
status: 'running',
phase: 'planning',
currentAction: '根据用户回答继续规划',
userInputRequest: null,
updatedAt: 6100,
});
return runtimeResult();
}
if (command === 'hydrate_design_agent_session') {
return currentDesignAgentView;
}
if (command === 'get_design_agent_runtime_mode') {
return currentDesignAgentView ? { activeRuntime: 'design' } : null;
}
if (command === 'set_design_agent_runtime_mode') {
return { activeRuntime: args?.activeRuntime };
}
if (command === 'continue_design_agent_session') {
const next = currentDesignContinueView ?? currentDesignAgentView;
if (!next) {
throw new Error('DESIGN_AGENT_VIEW_NOT_CONFIGURED');
}
currentDesignAgentView = next;
return next;
}
if (command === 'decide_design_phase') {
if (!currentDesignAgentView) {
throw new Error('DESIGN_AGENT_VIEW_NOT_CONFIGURED');
}
return currentDesignAgentView;
}
if (command === 'list_design_workspace') {
return currentDesignWorkspaceFiles;
}
if (command === 'read_design_workspace_file') {
return '';
}
throw new Error(`unexpected invoke ${command}`);
},
);
const listen = vi.fn(
async (
eventName: string,
handler: (event: {
payload: {
projectPath: string;
agentId: string;
runId: string;
status: string;
phase: string;
runtime: Record<string, unknown>;
};
}) => void,
) => {
if (eventName === 'game-creator-agent-runtime-update') {
runtimeUpdateHandler = handler;
}
if (eventName === 'game-creator-manifest-invalidated') {
manifestInvalidatedHandler =
handler as unknown as typeof manifestInvalidatedHandler;
}
if (eventName === 'game-creator-agent-progress') {
progressHandler = handler as unknown as typeof progressHandler;
}
if (eventName === 'design-agent-update') {
designAgentUpdateHandler =
handler as unknown as typeof designAgentUpdateHandler;
}
if (eventName === 'game-creator-direct-thread-notify') {
directThreadNotifyHandler =
handler as unknown as typeof directThreadNotifyHandler;
}
return () => {
if (runtimeUpdateHandler === handler) {
runtimeUpdateHandler = null;
}
if (manifestInvalidatedHandler === handler) {
manifestInvalidatedHandler = null;
}
if (progressHandler === handler) {
progressHandler = null;
}
if (designAgentUpdateHandler === handler) {
designAgentUpdateHandler = null;
}
if (directThreadNotifyHandler === handler) {
directThreadNotifyHandler = null;
}
};
},
);
/** 追加运行态事件并唤醒订阅者:bootstrap / consume 共用同一份队列。 */
const emitDirectThreadEvents = (
...events: Array<Record<string, unknown>>
) => {
for (const event of events) {
directThreadEventSequence += 1;
directThreadPendingEventSeqs.set(event, directThreadEventSequence);
if (event.type === 'turn.started') directThreadTurnRunning = true;
if (event.type === 'turn.completed') directThreadTurnRunning = false;
if (event.type === 'turn.started' || event.type === 'turn.completed') {
// 生命周期锚点只留最新一条,与 Thread Manager 的 `lifecycle_anchor` 一致。
directThreadLifecycleAnchor = {
seq: directThreadEventSequence,
event,
};
}
const itemId = directThreadEventItemId(event);
if (itemId && event.type === 'item.started') {
directThreadActiveItemIds.add(itemId);
}
if (itemId && event.type === 'item.completed') {
directThreadActiveItemIds.delete(itemId);
}
}
pendingDirectThreadEvents.push(...events);
if (directThreadSubscriptionId) {
directThreadNotifyHandler?.({
payload: { subscriptionId: directThreadSubscriptionId },
});
}
};
return {
invoke,
listen,
projectPath,
sessionId,
runtimeState,
runtimeResult,
setConfirmRuntime(state: Record<string, unknown>) {
confirmRuntime = state;
},
setRejectRuntime(state: Record<string, unknown>) {
rejectRuntime = state;
},
setAnswerRuntime(state: Record<string, unknown>) {
answerRuntime = state;
},
failNextAnswers(count = 1) {
answerFailuresRemaining = count;
},
emitDesignAgentEvent(payload: Record<string, unknown>) {
designAgentUpdateHandler?.({ payload });
},
appendSupervisorMessage(message: Record<string, unknown>) {
currentSupervisorMessages.push(message);
},
setResponseStream(responseStream: Record<string, unknown> | null) {
currentResponseStream = responseStream;
},
setProjectRevision(revision: number) {
currentProjectRevision = revision;
},
setRuntimeReader(reader: (() => Promise<Record<string, unknown>>) | null) {
runtimeReader = reader;
},
emitRuntime(
state: Record<string, unknown>,
responseStream: Record<string, unknown> | null = currentResponseStream,
) {
currentRuntime = state;
currentResponseStream = responseStream;
runtimeUpdateHandler?.({
payload: {
projectPath,
agentId: 'project-supervisor',
runId: String(state.runId),
status: String(state.status),
phase: String(state.phase),
manifestInvalidated: true,
runtime: runtimeResult(currentRuntime, currentResponseStream),
},
});
},
emitAgentRuntime(state: Record<string, unknown>) {
runtimeUpdateHandler?.({
payload: {
projectPath,
agentId: String(state.agentId ?? ''),
runId: String(state.runId ?? ''),
status: String(state.status ?? ''),
phase: String(state.phase ?? ''),
manifestInvalidated: true,
runtime: runtimeResult(state, null),
},
});
},
emitManifestInvalidated(agentId: string, eventProjectPath = projectPath) {
manifestInvalidatedHandler?.({
payload: {
projectPath: eventProjectPath,
agentId,
},
});
},
emitProgress(stage: string, message: string) {
progressHandler?.({
payload: {
projectPath,
stage,
message,
},
});
},
emitDirectThreadEvents,
/**
* 一轮 Direct 回合的标准事件序列:生命周期 → 落盘用户条目 → 助手正文 → 终态。
*
* 与 Rust 侧一致:消息身份是 `direct-codex:{turnId}:{role}`,工具条目另配 itemId。
* 事件级 `at` 与条目 `item.at` 同源:原生在两个阶段事件上都给时间,计时只读事件级那一份。
*/
completeDirectThreadTurn({
turnId,
reply,
prompt = '',
at = 9_000,
status = 'completed',
}: {
turnId: string;
reply: string;
prompt?: string;
at?: number;
status?: string;
}) {
// 生命周期锚点只有一份:回合已经在跑时不再补 `turn.started`。
const events: Array<Record<string, unknown>> = directThreadTurnRunning
? []
: [{ type: 'turn.started', at }];
if (prompt.trim()) {
events.push({
type: 'item.completed',
at,
item: {
itemType: 'message',
itemId: `direct-codex:${turnId}:user`,
role: 'user',
text: prompt,
at,
},
});
}
events.push({
type: 'item.completed',
at: at + 1,
item: {
itemType: 'message',
itemId: `direct-codex:${turnId}:assistant`,
role: 'assistant',
text: reply,
at: at + 1,
},
});
directThreadLastCompletedItemId = `direct-codex:${turnId}:assistant`;
// 终态时间就是调用方给的这一刻:整轮总耗时读它,不读正文条目的展示时间。
events.push({ type: 'turn.completed', status, at });
emitDirectThreadEvents(...events);
},
setDirectThreadHistory(items: Array<Record<string, unknown>>) {
directThreadHistoryItems = [...items];
// 生产口径:`subscribe` 回执里的 `lastCompletedItemId` 是订阅那一刻文件里最后一条可显示
// 条目(Rust 侧由 `read_direct_project_last_item_id_at` 从磁盘回填)。
const newest = directThreadHistoryItems.at(-1);
directThreadLastCompletedItemId =
typeof newest?.itemId === 'string' ? newest.itemId : null;
},
/** 模拟"订阅回执给的就是这一刻的最后一条已完成条目":之后落盘的条目只应从运行态事件来。 */
setDirectThreadLastCompletedItemId(itemId: string | null) {
directThreadLastCompletedItemId = itemId;
},
};
}
async function openMainProject(projectPath: string) {
await submitChat(`/project ${projectPath}`);
fireEvent.click(screen.getByRole('button', { name: '确认' }));
expect(
await screen.findByText(`已打开:${projectNameFromPath(projectPath)}`),
).not.toBeNull();
}
export function installResizeObserverStub() {
let observerCount = 0;
let observerDisconnected = false;
class TestResizeObserver {
constructor(readonly callback: ResizeObserverCallback) {
observerCount += 1;
}
observe() {}
unobserve() {}
disconnect() {
observerDisconnected = true;
}
}
Object.defineProperty(window, 'ResizeObserver', {
configurable: true,
value: TestResizeObserver,
});
return {
observerCount: () => observerCount,
observerDisconnected: () => observerDisconnected,
};
}
beforeEach(() => {
resetLlmModelCatalogCacheForTest();
vi.spyOn(clientApi, 'loadClientLlmModels').mockResolvedValue({
defaultModelId: 'quality',
models: [
{ id: 'quality', displayName: '高质量' },
{ id: 'fast', displayName: '快速' },
],
revision: 1,
});
Object.defineProperty(window, 'PointerEvent', {
configurable: true,
value: TestPointerEvent,
});
Object.defineProperty(window, 'ClipboardEvent', {
configurable: true,
value: TestClipboardEvent,
});
Object.defineProperty(window, 'DragEvent', {
configurable: true,
value: TestDragEvent,
});
Object.defineProperty(Range.prototype, 'getBoundingClientRect', {
configurable: true,
value: emptyRangeRect,
});
Object.defineProperties(URL, {
createObjectURL: {
configurable: true,
value: vi.fn(() => 'blob:mock-attachment-preview'),
},
revokeObjectURL: {
configurable: true,
value: vi.fn(),
},
});
});
afterEach(() => {
cleanup();
// 首页创作类型存在 module 级 zustand store 里,跨用例不会自动回到初始值。
// 建项按钮的文案依赖它(做方案 → 进入立项策划),漏一次就会让下一个用例
// 找不到「开启创作」。
useLauncherHomeDraftStore.getState().reset();
nativeClipboardMock.text = '';
window.history.pushState({}, '', '/');
window.localStorage.clear();
window.sessionStorage.clear();
delete window.__TAURI__;
vi.restoreAllMocks();
if (originalCreateObjectUrl) {
Object.defineProperty(URL, 'createObjectURL', originalCreateObjectUrl);
} else {
Reflect.deleteProperty(URL, 'createObjectURL');
}
if (originalRevokeObjectUrl) {
Object.defineProperty(URL, 'revokeObjectURL', originalRevokeObjectUrl);
} else {
Reflect.deleteProperty(URL, 'revokeObjectURL');
}
if (originalPointerEvent) {
Object.defineProperty(window, 'PointerEvent', originalPointerEvent);
} else {
Reflect.deleteProperty(window, 'PointerEvent');
}
if (originalClipboardEvent) {
Object.defineProperty(window, 'ClipboardEvent', originalClipboardEvent);
} else {
Reflect.deleteProperty(window, 'ClipboardEvent');
}
if (originalDragEvent) {
Object.defineProperty(window, 'DragEvent', originalDragEvent);
} else {
Reflect.deleteProperty(window, 'DragEvent');
}
if (originalRangeGetBoundingClientRect) {
Object.defineProperty(
Range.prototype,
'getBoundingClientRect',
originalRangeGetBoundingClientRect,
);
} else {
Reflect.deleteProperty(Range.prototype, 'getBoundingClientRect');
}
if (originalIntersectionObserver) {
Object.defineProperty(
window,
'IntersectionObserver',
originalIntersectionObserver,
);
} else {
Reflect.deleteProperty(window, 'IntersectionObserver');
}
if (originalResizeObserver) {
Object.defineProperty(window, 'ResizeObserver', originalResizeObserver);
} else {
Reflect.deleteProperty(window, 'ResizeObserver');
}
});
export {
act,
agentRuntimeUserInputRequest,
App,
AuthenticatedClient,
cleanup,
composerDisabled,
composerText,
composerValue,
createGameCreationAppManifest,
createGameCreationAppSeedTasks,
createProjectSupervisorRuntimeHarness,
deriveAgentStatusCards,
describe,
emptyProjectPolicy,
expect,
findResourceSelectButton,
fireEvent,
GAME_CREATION_AGENT_RUN_SCHEMA_VERSION,
getResourceSelectButton,
it,
mockRoleAgentReply,
nativeClipboardMock,
openMainProject,
openResourceFilterPanel,
pickProjectFromLauncher,
ProjectDevelopmentView,
projectSupervisorResponseStream,
queryResourceSelectButton,
React,
readFileSync,
render,
renderAppAt,
renderLauncherAgentChatAt,
renderLauncherAt,
renderLauncherProjectsAt,
resolve,
roleAgentMockReply,
screen,
selectDeveloperAgentChatMode,
setComposerText,
submitChat,
TEST_LOCAL_PROJECT_PATH,
testAuthUser,
vi,
waitFor,
within,
WorkspaceLauncher,
};
export type { GameCreationAgentRunTrace };