惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Vercel News
Vercel News
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
雷峰网
雷峰网
有赞技术团队
有赞技术团队
罗磊的独立博客
博客园 - 叶小钗
Jina AI
Jina AI
博客园 - 司徒正美
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
T
Tailwind CSS Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
人人都是产品经理
人人都是产品经理
Apple Machine Learning Research
Apple Machine Learning Research
阮一峰的网络日志
阮一峰的网络日志
Microsoft Security Blog
Microsoft Security Blog
大猫的无限游戏
大猫的无限游戏
量子位
MyScale Blog
MyScale Blog
V
Visual Studio Blog
博客园 - 聂微东
The Cloudflare Blog
Engineering at Meta
Engineering at Meta
小众软件
小众软件
宝玉的分享
宝玉的分享

Recent Commits to openclaw:main

test: merge chat side-result checks · openclaw/openclaw@ddd2c2a test: merge cron history checks · openclaw/openclaw@f7eb746 test: merge responsive navigation shell checks · openclaw/openclaw@c2e4b47 docs(changelog): add codex oauth fixes · openclaw/openclaw@628e6cd test: merge navigation routing cases · openclaw/openclaw@5d8cecb Tests: mock channel registry bundled fallback · openclaw/openclaw@2b08233 Secrets: avoid broad web search discovery for single plugin config · openclaw/openclaw@a464f59 test: merge config view browser checks · openclaw/openclaw@20cf511 fix(status): align oauth health with runtime · openclaw/openclaw@eed7116 feat: add macOS screen snapshots for monitor preview (#67954) thanks … · openclaw/openclaw@f377db1 fix: report shared auth scopes in hello-ok (#67810) thanks @BunsDev · openclaw/openclaw@0b6c39b Auto-reply: avoid eager bundled route fallback · openclaw/openclaw@3ea1bf4 Tests: narrow session binding contract setup · openclaw/openclaw@54e4e16 fix(macOS): enable undo/redo in webchat composer text input (#34962) · openclaw/openclaw@00951dc Tests: speed up channel setup promotion · openclaw/openclaw@82b529a Docs: refresh agent instructions · openclaw/openclaw@5775fe2 fix(auth): serialize OAuth refresh across agents to fix #26322 (#67876) · openclaw/openclaw@8e79080 test: allow ollama public surface boundary test · openclaw/openclaw@7d4f1a6 Docs: add test performance guardrails · openclaw/openclaw@89706d3 Tests: restore context-engine usage proof · openclaw/openclaw@e4c4f95 Tests: slim context engine runtime coverage · openclaw/openclaw@74c198f ci: retry failed custom checkouts · openclaw/openclaw@0ee5baf test: trim duplicate provider auth onboarding cases · openclaw/openclaw@1ffc02e matrix: fix sessions_spawn --thread subagent session spawning (#67643) · openclaw/openclaw@1ce2596 test: reduce auth choice fixture churn · openclaw/openclaw@857b9cd test: mock health status config boundaries · openclaw/openclaw@9d5ab4a test: mock onboard config io boundary · openclaw/openclaw@299694d test: mock legacy state plugin boundaries · openclaw/openclaw@2713089 test: mock channel install boundaries · openclaw/openclaw@b945248 test: mock doctor preview channel boundaries · openclaw/openclaw@b1a3ad4
test: add live subagent steering proof · openclaw/opencla...
steipete · 2026-05-13 · via Recent Commits to openclaw:main

@@ -0,0 +1,317 @@

1+

import { randomBytes, randomUUID } from "node:crypto";

2+

import path from "node:path";

3+

import { afterEach, describe, expect, it } from "vitest";

4+

import { clearRuntimeConfigSnapshot, type OpenClawConfig } from "../config/config.js";

5+

import { callGateway as realCallGateway } from "../gateway/call.js";

6+

import { GatewayClient } from "../gateway/client.js";

7+

import { dispatchGatewayMethodInProcess as realDispatchGatewayMethodInProcess } from "../gateway/server-plugins.js";

8+

import { startGatewayServer, type GatewayServer } from "../gateway/server.js";

9+

import { extractPayloadText } from "../gateway/test-helpers.agent-results.js";

10+

import { isTruthyEnvValue } from "../infra/env.js";

11+

import { clearCurrentPluginMetadataSnapshot } from "../plugins/current-plugin-metadata-snapshot.js";

12+

import {

13+

createOpenClawTestState,

14+

type OpenClawTestState,

15+

} from "../test-utils/openclaw-test-state.js";

16+

import { GATEWAY_CLIENT_MODES, GATEWAY_CLIENT_NAMES } from "../utils/message-channel.js";

17+

import { isLiveTestEnabled } from "./live-test-helpers.js";

18+

import { __testing as subagentAnnounceDeliveryTesting } from "./subagent-announce-delivery.js";

19+

import { __testing as subagentAnnounceTesting } from "./subagent-announce.js";

20+

import { listSubagentRunsForRequester } from "./subagent-registry.js";

21+22+

const LIVE = isLiveTestEnabled() && isTruthyEnvValue(process.env.OPENCLAW_LIVE_SUBAGENT_E2E);

23+

const describeLive = LIVE ? describe : describe.skip;

24+25+

type AgentPayload = {

26+

status?: string;

27+

result?: unknown;

28+

};

29+30+

type InProcessAgentDispatch =

31+

| { phase: "started"; resultText?: undefined }

32+

| { phase: "completed"; resultText: string };

33+34+

const REQUEST_TIMEOUT_MS = 4 * 60_000;

35+

const WAIT_TIMEOUT_MS = 5 * 60_000;

36+37+

function sleep(ms: number): Promise<void> {

38+

return new Promise((resolve) => setTimeout(resolve, ms));

39+

}

40+41+

function openAiConfig(

42+

modelKey: string,

43+

workspace: string,

44+

port: number,

45+

token: string,

46+

): OpenClawConfig {

47+

return {

48+

gateway: {

49+

mode: "local",

50+

port,

51+

auth: { mode: "token", token },

52+

controlUi: { enabled: false },

53+

},

54+

plugins: { enabled: false },

55+

tools: {

56+

allow: ["sessions_spawn", "sessions_yield", "subagents"],

57+

},

58+

models: {

59+

providers: {

60+

openai: {

61+

api: "openai-responses",

62+

agentRuntime: { id: "pi" },

63+

apiKey: { source: "env", provider: "default", id: "OPENAI_API_KEY" },

64+

baseUrl: "https://api.openai.com/v1",

65+

timeoutSeconds: 180,

66+

models: [

67+

{

68+

id: modelKey.replace(/^openai\//u, ""),

69+

name: modelKey.replace(/^openai\//u, ""),

70+

api: "openai-responses",

71+

agentRuntime: { id: "pi" },

72+

input: ["text"],

73+

reasoning: true,

74+

contextWindow: 1_047_576,

75+

maxTokens: 8_192,

76+

cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },

77+

},

78+

],

79+

},

80+

},

81+

},

82+

agents: {

83+

defaults: {

84+

workspace,

85+

model: { primary: modelKey },

86+

models: { [modelKey]: { agentRuntime: { id: "pi" }, params: { maxTokens: 1024 } } },

87+

sandbox: { mode: "off" },

88+

subagents: {

89+

allowAgents: ["*"],

90+

runTimeoutSeconds: 120,

91+

announceTimeoutMs: 120_000,

92+

archiveAfterMinutes: 60,

93+

},

94+

},

95+

},

96+

};

97+

}

98+99+

async function waitFor<T>(

100+

label: string,

101+

fn: () => T | undefined | Promise<T | undefined>,

102+

): Promise<T> {

103+

const started = Date.now();

104+

let lastValue: T | undefined;

105+

while (Date.now() - started < WAIT_TIMEOUT_MS) {

106+

lastValue = await fn();

107+

if (lastValue !== undefined) {

108+

return lastValue;

109+

}

110+

await sleep(1_000);

111+

}

112+

throw new Error(`timed out waiting for ${label}`);

113+

}

114+115+

function createGatewayClient(params: {

116+

port: number;

117+

token: string;

118+

onEvent?: ConstructorParameters<typeof GatewayClient>[0]["onEvent"];

119+

}): Promise<GatewayClient> {

120+

return new Promise((resolve, reject) => {

121+

const client = new GatewayClient({

122+

url: `ws://127.0.0.1:${params.port}`,

123+

token: params.token,

124+

deviceIdentity: null,

125+

clientName: GATEWAY_CLIENT_NAMES.GATEWAY_CLIENT,

126+

mode: GATEWAY_CLIENT_MODES.BACKEND,

127+

scopes: ["operator.admin"],

128+

requestTimeoutMs: REQUEST_TIMEOUT_MS,

129+

onEvent: params.onEvent,

130+

onHelloOk: () => resolve(client),

131+

onConnectError: reject,

132+

});

133+

client.start();

134+

});

135+

}

136+137+

describeLive("subagent announce live", () => {

138+

let state: OpenClawTestState | undefined;

139+

let server: GatewayServer | undefined;

140+

let client: GatewayClient | undefined;

141+142+

afterEach(async () => {

143+

subagentAnnounceTesting.setDepsForTest();

144+

subagentAnnounceDeliveryTesting.setDepsForTest();

145+

await client?.stopAndWait().catch(() => undefined);

146+

await server?.close({ reason: "subagent announce live test done" }).catch(() => undefined);

147+

await state?.cleanup().catch(() => undefined);

148+

clearRuntimeConfigSnapshot();

149+

clearCurrentPluginMetadataSnapshot();

150+

client = undefined;

151+

server = undefined;

152+

state = undefined;

153+

});

154+155+

it(

156+

"lets a parent steer a subagent and receives completion through in-process agent dispatch",

157+

async () => {

158+

expect(process.env.OPENAI_API_KEY?.trim(), "OPENAI_API_KEY").toBeTruthy();

159+160+

const token = `subagent-live-${randomUUID()}`;

161+

const port = 30_000 + Math.floor(Math.random() * 10_000);

162+

const modelKey = process.env.OPENCLAW_LIVE_SUBAGENT_E2E_MODEL?.trim() || "openai/gpt-5.5";

163+

const nonce = randomBytes(3).toString("hex").toUpperCase();

164+

const childToken = `CHILD_STEERED_${nonce}`;

165+

const parentToken = `PARENT_SAW_${childToken}`;

166+

const steerToken = `STEER_${nonce}`;

167+

const childTask = [

168+

`Immediately call sessions_yield with message="waiting for ${steerToken}".`,

169+

`After a steering message containing ${steerToken} arrives, reply exactly ${childToken}.`,

170+

`Do not reply with ${childToken} before receiving ${steerToken}.`,

171+

].join(" ");

172+

const sessionKey = `agent:main:live-subagent-${nonce.toLowerCase()}`;

173+

const inProcessAgentDispatches: InProcessAgentDispatch[] = [];

174+175+

const forbiddenAgentRpc: typeof realCallGateway = async (request) => {

176+

if (request.method === "agent") {

177+

throw new Error("subagent announce live test forbids gateway RPC method=agent");

178+

}

179+

return await realCallGateway(request);

180+

};

181+

const instrumentedDispatch: typeof realDispatchGatewayMethodInProcess = async <T>(

182+

method: string,

183+

params: Record<string, unknown>,

184+

options?: Parameters<typeof realDispatchGatewayMethodInProcess>[2],

185+

): Promise<T> => {

186+

if (method === "agent") {

187+

inProcessAgentDispatches.push({ phase: "started" });

188+

}

189+

const result = await realDispatchGatewayMethodInProcess<T>(method, params, options);

190+

if (method === "agent") {

191+

inProcessAgentDispatches.push({

192+

phase: "completed",

193+

resultText: extractPayloadText((result as AgentPayload).result),

194+

});

195+

}

196+

return result;

197+

};

198+199+

subagentAnnounceTesting.setDepsForTest({

200+

callGateway: forbiddenAgentRpc,

201+

dispatchGatewayMethodInProcess: instrumentedDispatch,

202+

});

203+

subagentAnnounceDeliveryTesting.setDepsForTest({

204+

callGateway: forbiddenAgentRpc,

205+

dispatchGatewayMethodInProcess: instrumentedDispatch,

206+

getRequesterSessionActivity: () => ({

207+

sessionId: "requester-session-local",

208+

isActive: false,

209+

}),

210+

});

211+212+

state = await createOpenClawTestState({

213+

label: "subagent-announce-live",

214+

layout: "split",

215+

env: {

216+

OPENCLAW_SKIP_CHANNELS: "1",

217+

OPENCLAW_SKIP_CRON: "1",

218+

OPENCLAW_SKIP_BROWSER_CONTROL_SERVER: "1",

219+

OPENCLAW_SKIP_CANVAS_HOST: "1",

220+

OPENCLAW_TEST_MINIMAL_GATEWAY: "1",

221+

OPENCLAW_DISABLE_BUNDLED_PLUGINS: undefined,

222+

OPENCLAW_DISABLE_PERSISTED_PLUGIN_REGISTRY: "1",

223+

OPENCLAW_BUNDLED_PLUGINS_DIR: path.resolve("extensions"),

224+

OPENCLAW_TEST_TRUST_BUNDLED_PLUGINS_DIR: "1",

225+

OPENCLAW_PLUGIN_CATALOG_PATHS: undefined,

226+

OPENCLAW_PLUGINS_PATHS: undefined,

227+

},

228+

});

229+

await state.writeConfig(openAiConfig(modelKey, state.workspaceDir, port, token));

230+

clearRuntimeConfigSnapshot();

231+

clearCurrentPluginMetadataSnapshot();

232+233+

server = await startGatewayServer(port, {

234+

bind: "loopback",

235+

auth: { mode: "token", token },

236+

controlUiEnabled: false,

237+

});

238+

client = await createGatewayClient({ port, token });

239+240+

let initialError: unknown;

241+

const initialRequest = client.request<AgentPayload>(

242+

"agent",

243+

{

244+

sessionKey,

245+

idempotencyKey: `live-subagent-${randomUUID()}`,

246+

deliver: false,

247+

timeout: 180,

248+

message: [

249+

"Run this exact OpenClaw subagent steering scenario. Use tool calls, not prose.",

250+

`Use nonce ${nonce}.`,

251+

`Step 1: call sessions_spawn with exactly this JSON input: ${JSON.stringify({

252+

task: childTask,

253+

taskName: "steered_child",

254+

cleanup: "keep",

255+

context: "isolated",

256+

runTimeoutSeconds: 120,

257+

})}.`,

258+

`Step 2: after spawn returns status="accepted", call subagents with exactly this JSON input: ${JSON.stringify(

259+

{

260+

action: "steer",

261+

target: "steered_child",

262+

message: steerToken,

263+

},

264+

)}.`,

265+

`Step 3: call sessions_yield with message="waiting for ${childToken}" and wait for the child completion event.`,

266+

`Step 4: after the completion event arrives, reply exactly ${parentToken}.`,

267+

"Do not reply with the parent token until the child completion event is visible.",

268+

].join("\n"),

269+

},

270+

{ expectFinal: true, timeoutMs: REQUEST_TIMEOUT_MS },

271+

);

272+

initialRequest.catch((error: unknown) => {

273+

initialError = error;

274+

});

275+276+

const steeredRun = await waitFor("steered child completion", () => {

277+

if (initialError) {

278+

throw initialError;

279+

}

280+

return listSubagentRunsForRequester(sessionKey).find(

281+

(run) =>

282+

run.taskName === "steered_child" &&

283+

run.frozenResultText?.includes(childToken) === true &&

284+

run.outcome?.status === "ok",

285+

);

286+

});

287+

expect(steeredRun.endedReason).toBe("subagent-complete");

288+

expect(steeredRun.lastAnnounceDeliveryError).toBeUndefined();

289+290+

await waitFor("in-process subagent completion agent dispatch start", () => {

291+

if (initialError) {

292+

throw initialError;

293+

}

294+

return inProcessAgentDispatches.some((entry) => entry.phase === "started")

295+

? true

296+

: undefined;

297+

});

298+299+

const completedDispatch = inProcessAgentDispatches.find(

300+

(entry) => entry.phase === "completed",

301+

);

302+

if (completedDispatch) {

303+

expect(completedDispatch.resultText).toContain(childToken);

304+

}

305+

expect(

306+

inProcessAgentDispatches.some((entry) => {

307+

if (initialError) {

308+

throw initialError;

309+

}

310+

return entry.phase === "started";

311+

}),

312+

).toBe(true);

313+

expect(inProcessAgentDispatches.length).toBeGreaterThanOrEqual(1);

314+

},

315+

6 * 60_000,

316+

);

317+

});