Files
Genarrative/apps/ai-game-creator-shell/tests/agentRuntimeModel.test.ts
T
kdletters c2ef8631f8 明确泥点不足的游戏生成中断原因
统一余额不足分类与稳定原因码
阻止泥点不足进入重试和对账状态
在游戏聊天与阶段记录展示固定安全说明
补齐 API、Runtime 与前端回归测试
同步技术方案与项目决策记录
2026-08-04 18:58:48 +08:00

419 lines
14 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 { describe, expect, test, vi } from 'vitest';
import type {
AgentRuntimeResponseStream,
AgentRuntimeResult,
AgentRuntimeState,
ChatMessage,
} from '../src/app/types';
import {
formatAgentRuntimeEvent,
MUD_POINT_INSUFFICIENT_INTERRUPTION_MESSAGE,
mergeGameChatRuntimeResponseMessagesIntoHistory,
projectRuntimeVisibleCurrentWork,
projectRuntimeVisibleError,
projectSupervisorChatRuntimeStatus,
projectSupervisorResponseStreamIdentity,
projectSupervisorVisibleConversationText,
submitProjectSupervisorRuntimeTask,
} from '../src/features/agent-runtime/model';
describe('Game Chat stream identity and source', () => {
test('response stream identity binds run, slot, and revision', () => {
const base: AgentRuntimeResponseStream = {
schemaVersion: 'game-creator-runtime-response-stream.v1',
agentId: 'project-supervisor',
taskId: 'project-supervisor',
sessionId: 'session-1',
runId: 'run-1',
requestKind: 'final-reply',
requestSlot: 'final-reply-loop-1-revision-0',
appliedSteerCursor: 0,
responseRevision: 0,
sequence: 1,
status: 'ready',
accumulatedText: 'same text',
finishReason: 'stop',
startedAt: 1,
updatedAt: 2,
};
expect(projectSupervisorResponseStreamIdentity(base)).toBe(
'run-1\u001ffinal-reply-loop-1-revision-0\u001f0',
);
expect(
projectSupervisorResponseStreamIdentity({ ...base, runId: 'run-2' }),
).not.toBe(projectSupervisorResponseStreamIdentity(base));
expect(
projectSupervisorResponseStreamIdentity({
...base,
requestSlot: 'final-reply-loop-2-revision-0',
}),
).not.toBe(projectSupervisorResponseStreamIdentity(base));
expect(
projectSupervisorResponseStreamIdentity({ ...base, responseRevision: 1 }),
).not.toBe(projectSupervisorResponseStreamIdentity(base));
});
test('game-chat source is forwarded for start and steer without changing default source behavior', async () => {
const runtimeResult = {
state: {
schemaVersion: 'game-creator-agent-runtime.v1',
agentId: 'project-supervisor',
taskId: 'project-supervisor',
sessionId: 'session-provider-retry',
runId: 'accepted-run',
source: 'project-supervisor-chat',
status: 'running',
phase: 'planning',
currentTask: null,
currentAction: null,
waitingOn: null,
nextStep: null,
plan: [],
planSteps: [],
observations: [],
allowedTools: [],
lastResponse: null,
error: null,
updatedAt: 1,
},
sessionPath: 'session.json',
eventPath: 'events.jsonl',
} satisfies AgentRuntimeResult;
const invoke = vi.fn(
async (_command: string, _args?: Record<string, unknown>) => runtimeResult,
);
await submitProjectSupervisorRuntimeTask({
invoke,
projectPath: '/tmp/game-chat',
sessionId: 'session-provider-retry',
prompt: 'make a game',
runtime: null,
runProfile: 'autonomous-game-build',
source: 'project-supervisor-game-chat',
});
expect(invoke).toHaveBeenCalledWith(
'start_game_creator_supervisor_runtime_task',
expect.objectContaining({ source: 'project-supervisor-game-chat' }),
);
invoke.mockClear();
await submitProjectSupervisorRuntimeTask({
invoke,
projectPath: '/tmp/game-chat',
sessionId: 'session-provider-retry',
prompt: 'chat',
runtime: null,
runProfile: 'standard',
});
expect(invoke.mock.calls[0]?.[1]).not.toHaveProperty('source');
invoke.mockClear();
const steerRuntime = {
...runtimeResult.state,
runId: 'steer-run',
runProfile: 'autonomous-game-build',
status: 'running',
phase: 'planning',
} satisfies AgentRuntimeState;
invoke.mockResolvedValueOnce({
runtime: runtimeResult,
steerId: 'steer-1',
sequence: 1,
status: 'applied',
providerInterrupted: false,
});
await submitProjectSupervisorRuntimeTask({
invoke,
projectPath: '/tmp/game-chat',
sessionId: 'session-provider-retry',
prompt: 'continue this run',
runtime: steerRuntime,
runProfile: 'autonomous-game-build',
source: 'project-supervisor-game-chat',
});
expect(invoke).toHaveBeenCalledWith(
'steer_game_creator_agent_runtime_task',
expect.objectContaining({ source: 'project-supervisor-game-chat' }),
);
invoke.mockClear();
invoke.mockResolvedValueOnce({
runtime: runtimeResult,
steerId: 'steer-2',
sequence: 2,
status: 'applied',
providerInterrupted: false,
});
await submitProjectSupervisorRuntimeTask({
invoke,
projectPath: '/tmp/game-chat',
sessionId: 'session-provider-retry',
prompt: 'continue this run',
runtime: steerRuntime,
runProfile: 'autonomous-game-build',
});
expect(invoke.mock.calls[0]?.[1]).not.toHaveProperty('source');
});
test('game-chat hydration keeps identical text from a different runtime response identity', () => {
const history = [
{
role: 'assistant' as const,
text: 'same text',
messageId: 'persisted-message-1',
updatedAt: 20,
},
];
const pending = [
{
role: 'assistant' as const,
text: 'same text',
messageId: 'runtime-response:run-2\u001fslot-1\u001f0',
updatedAt: 10,
runtimeOwned: true,
},
];
expect(mergeGameChatRuntimeResponseMessagesIntoHistory(history, pending)).toEqual([
pending[0],
history[0],
]);
expect(
mergeGameChatRuntimeResponseMessagesIntoHistory(history, [
{ ...pending[0], messageId: 'persisted-message-1' },
]),
).toEqual(history);
});
test('ready response survives either ordering of runtime commit and conversation hydration', () => {
const history = [
{
role: 'user' as const,
text: 'build a game',
messageId: 'persisted-message-1',
updatedAt: 1,
},
];
const ready = {
role: 'assistant' as const,
text: 'ready response',
messageId: 'runtime-response:run-1\u001fslot-1\u001f0',
updatedAt: 2,
runtimeOwned: true,
};
const commitReady = (current: ChatMessage[]) =>
current.some((message) => message.messageId === ready.messageId)
? current
: [...current, ready];
const hydrateConversation = (current: ChatMessage[]) =>
mergeGameChatRuntimeResponseMessagesIntoHistory(history, current);
expect(hydrateConversation(commitReady([]))).toContainEqual(ready);
expect(commitReady(hydrateConversation([]))).toContainEqual(ready);
});
});
function providerRetryRuntime(): AgentRuntimeState {
return {
schemaVersion: 'game-creator-agent-runtime.v1',
agentId: 'project-supervisor',
taskId: 'task-provider-retry',
sessionId: 'session-provider-retry',
runId: 'run-provider-retry',
source: 'project-supervisor-chat',
status: 'running',
phase: 'waiting-for-provider-retry',
currentTask: '生成项目总控计划',
currentAction: 'Provider 上游返回 HTTP 503,准备自动重试 1/3',
waitingOn: '预计 30 秒后重试',
nextStep: '到期后继续同一 Run',
plan: ['生成项目总控计划'],
planSteps: [
{
step: '这个活跃计划步骤不应覆盖真实 Provider 等待进度',
status: 'in_progress',
index: 0,
},
],
observations: [],
allowedTools: [],
lastResponse: null,
error: null,
updatedAt: 1,
};
}
describe('Agent Runtime Provider 状态投影', () => {
test('等待 503 重试时优先显示 Runtime 的真实动作和等待时间', () => {
const runtime = providerRetryRuntime();
const expected =
'Provider 上游返回 HTTP 503,准备自动重试 1/3;预计 30 秒后重试';
expect(projectSupervisorChatRuntimeStatus(runtime)).toBe(expected);
expect(projectRuntimeVisibleCurrentWork(runtime)).toBe(expected);
});
test('旧等待投影不再显示内部 upstream-5xx 分类', () => {
const legacyRuntime = {
...providerRetryRuntime(),
currentAction: '等待 Provider 瞬态重试 1/3',
waitingOn: 'Provider upstream-5xx 瞬态故障退避到期',
};
const expected = 'Provider 上游服务暂时不可用,准备自动重试 1/3';
expect(projectSupervisorChatRuntimeStatus(legacyRuntime)).toBe(expected);
expect(projectRuntimeVisibleCurrentWork(legacyRuntime)).toBe(expected);
expect(projectSupervisorChatRuntimeStatus(legacyRuntime)).not.toContain(
'upstream-5xx',
);
});
test('严格解析 503 重试耗尽的稳定字段', () => {
const fingerprint = 'a'.repeat(64);
const message =
`agentLlm.project-supervisor 规划调用 LLM 失败:kind=upstream-503 ` +
`httpStatus=503 fingerprint=${fingerprint} chars=248 ` +
'retryAttempt=3 maxRetries=3 retryState=exhausted';
expect(projectRuntimeVisibleError(message, '项目总控 Agent', true)).toBe(
'项目总控 Agent 上游服务返回 HTTP 503;自动重试已耗尽(3/3',
);
const failedRuntime = {
...providerRetryRuntime(),
status: 'failed',
phase: 'failed',
error: message,
};
expect(projectSupervisorChatRuntimeStatus(failedRuntime)).toBe(
'项目总控 Agent 上游服务返回 HTTP 503;自动重试已耗尽(3/3',
);
});
test('明确说明泥点余额不足导致的中断并隐藏附加诊断', () => {
for (const message of [
'agentLlm.project-supervisor 规划调用 LLM 失败:kind=mud-points-insufficient',
'平台图片生成任务失败:泥点余额不足;operationId=private-operation-id',
]) {
expect(projectRuntimeVisibleError(message, '项目总控 Agent', true)).toBe(
MUD_POINT_INSUFFICIENT_INTERRUPTION_MESSAGE,
);
expect(
projectSupervisorVisibleConversationText(`后台任务失败:${message}`),
).toBe(MUD_POINT_INSUFFICIENT_INTERRUPTION_MESSAGE);
}
const failedRuntime = {
...providerRetryRuntime(),
status: 'failed',
phase: 'failed',
error:
'平台图片生成任务失败:泥点余额不足;operationId=private-operation-id',
};
const status = projectSupervisorChatRuntimeStatus(failedRuntime);
expect(status).toBe(MUD_POINT_INSUFFICIENT_INTERRUPTION_MESSAGE);
expect(status).not.toContain('operationId');
expect(status).not.toContain('private-operation-id');
expect(
projectRuntimeVisibleError('上游余额不足', '项目总控 Agent', true),
).toBe('项目总控 Agent 执行失败,请稍后重试');
});
test('即使前缀含恶意上游正文也只展示严格字段派生的摘要', () => {
const fingerprint = 'b'.repeat(64);
const malicious =
'https://provider.example/private?api_key=sk-secret C:\\Users\\victim\\project';
const message =
`${malicious}kind=upstream-503 httpStatus=503 ` +
`fingerprint=${fingerprint} chars=999 retryAttempt=3 ` +
'maxRetries=3 retryState=exhausted';
const visible = projectRuntimeVisibleError(message, '项目总控 Agent', true);
expect(visible).toBe(
'项目总控 Agent 上游服务返回 HTTP 503;自动重试已耗尽(3/3',
);
expect(visible).not.toContain('provider.example');
expect(visible).not.toContain('sk-secret');
expect(visible).not.toContain('victim');
});
test('拒绝字段不一致或带尾随正文的伪造 Provider 摘要', () => {
const fingerprint = 'c'.repeat(64);
const inconsistent =
`kind=upstream-503 httpStatus=500 fingerprint=${fingerprint} chars=1 ` +
'retryAttempt=3 maxRetries=3 retryState=exhausted';
const trailingBody =
`kind=upstream-503 httpStatus=503 fingerprint=${fingerprint} chars=1 ` +
'retryAttempt=3 maxRetries=3 retryState=exhausted provider body';
expect(
projectRuntimeVisibleError(inconsistent, '项目总控 Agent', true),
).toBe('项目总控 Agent 执行失败,请稍后重试');
expect(
projectRuntimeVisibleError(trailingBody, '项目总控 Agent', true),
).toBe('项目总控 Agent 执行失败,请稍后重试');
});
test('隐藏旧失败对话与事件中的内部诊断标记', () => {
const fingerprint = 'd'.repeat(64);
const legacyConversation =
'后台任务失败:<absolute-path> [redacted sensitive context] ' +
`kind=upstream-503 httpStatus=503 fingerprint=${fingerprint} chars=99 ` +
'retryAttempt=3 maxRetries=3 retryState=exhausted';
expect(projectSupervisorVisibleConversationText(legacyConversation)).toBe(
'项目总控 Agent 上游服务返回 HTTP 503;自动重试已耗尽(3/3',
);
expect(
projectSupervisorVisibleConversationText(legacyConversation, 'user'),
).toBe(legacyConversation);
const formattedEvent = formatAgentRuntimeEvent({
schemaVersion: 'game-creator-agent-runtime.v1',
agentId: 'project-supervisor',
taskId: 'task-provider-retry',
sessionId: 'session-provider-retry',
runId: 'run-provider-retry',
source: 'project-supervisor-chat',
eventType: 'turn.failed',
status: 'failed',
phase: 'failed',
summary: 'Agent Runtime 本轮处理失败。',
detail:
`<absolute-path> [redacted sensitive context] ` +
`errorSha256=${fingerprint} · errorChars=99`,
updatedAt: 1,
});
expect(formattedEvent).toBe(
'turn.failed · failed / failed · Agent Runtime 本轮处理失败。',
);
expect(formattedEvent).not.toContain('errorSha256');
expect(formattedEvent).not.toContain(fingerprint);
const emptySummaryEvent = formatAgentRuntimeEvent({
schemaVersion: 'game-creator-agent-runtime.v1',
agentId: 'project-supervisor',
taskId: 'task-provider-retry',
sessionId: 'session-provider-retry',
runId: 'run-provider-retry',
source: 'project-supervisor-chat',
eventType: 'error',
status: 'failed',
phase: 'failed',
summary: '',
detail:
`<absolute-path> [redacted sensitive context] ` +
`fingerprint=${fingerprint} chars=99`,
updatedAt: 1,
});
expect(emptySummaryEvent).toBe(
'error · failed / failed · Agent Runtime 本轮处理失败。',
);
expect(emptySummaryEvent).not.toContain('absolute-path');
expect(emptySummaryEvent).not.toContain(fingerprint);
});
});