krisolo/boss
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: checkpoint Boss app v2.5.11

Browse Source
This commit is contained in:
AI Bot
2026-06-08 12:22:50 +08:00
parent bddbe8b5ba
commit 3b51641d99
78 changed files with 5706 additions and 954 deletions
Download Patch File Download Diff File Expand all files Collapse all files

712
local-agent/server.mjs
View File

@@ -13,6 +13,21 @@ import {
getCodexAppServerRunnerConfig,
shouldUseCodexAppServerTaskRunner,
} from "./codex-app-server-runner.mjs";
import {
shouldSkipCodexAppServerDiscovery,
} from "./codex-app-server-discovery-guard.mjs";
import {
buildLongRunningCodexProgressSnapshot,
normalizeLongRunningProgressIntervalMs,
} from "./master-task-progress-heartbeat.mjs";
import {
resolveHeartbeatProjectsFromSnapshot,
runHeartbeatProjectDiscoveryWithTimeout,
storeHeartbeatProjectsSnapshot,
} from "./heartbeat-project-snapshot.mjs";
import {
recordHeartbeatRunnerError,
} from "./heartbeat-error-state.mjs";
import { appendBossUserMessageToCodexThreadRollout } from "./codex-thread-rollout-writer.mjs";
import {
executeOmxTeamTask,
@@ -65,7 +80,15 @@ import {
buildMasterAgentTaskCompletionRequestBody,
buildRemoteExecutionCompletionPayload,
} from "./master-task-completion.mjs";
import {
postThroughReliableOutbox,
replayReliableOutbox,
} from "./reliable-outbox.mjs";
import {
buildLocalAgentHealthSummary,
} from "./health-summary.mjs";
import { createSerializedRunner } from "./serialized-runner.mjs";
import { fetchWithTimeout } from "./fetch-timeout.mjs";
async function loadConfig(configPath) {
const raw = await readFile(resolve(configPath), "utf8");
@@ -75,6 +98,23 @@ async function loadConfig(configPath) {
async function resolveHeartbeatProjects(config, runtime) {
const staticProjects = Array.isArray(config.projects) ? config.projects : [];
const staticCandidates = Array.isArray(config.projectCandidates) ? config.projectCandidates : [];
const snapshotFallback = runtime.lastHeartbeatProjectsSnapshot && typeof runtime.lastHeartbeatProjectsSnapshot === "object"
? runtime.lastHeartbeatProjectsSnapshot
: {
projects: staticProjects,
projectCandidates: staticCandidates,
guiConnected: runtime.lastCodexGuiConnected === true,
};
const snapshotDecision = resolveHeartbeatProjectsFromSnapshot({ config, runtime });
if (snapshotDecision.shouldUseSnapshot) {
runtime.lastProjectDiscoverySkippedAt = new Date().toISOString();
runtime.lastProjectDiscoverySkipReason = "master_task_running";
return {
projects: snapshotDecision.projects,
projectCandidates: snapshotDecision.projectCandidates,
guiConnected: snapshotDecision.guiConnected,
};
}
if (config.codexSessionDiscoveryEnabled === false) {
return {
projects: staticProjects,
@@ -83,14 +123,38 @@ async function resolveHeartbeatProjects(config, runtime) {
}
try {
const discovered = await discoverCodexProjectCandidatesInWorker({
stateDbPath: config.codexStateDbPath,
logsDbPath: config.codexLogsDbPath,
sessionIndexPath: config.codexSessionIndexPath,
globalStatePath: config.codexGlobalStatePath,
sessionsDir: config.codexSessionsDir,
lookbackHours: config.codexSessionLookbackHours,
const discoveryTimeoutMs = config.codexSessionDiscoveryTimeoutMs ?? 3_500;
const discoveryResult = await runHeartbeatProjectDiscoveryWithTimeout({
timeoutMs: discoveryTimeoutMs,
fallback: snapshotFallback,
discover: () => discoverCodexProjectCandidatesInWorker({
stateDbPath: config.codexStateDbPath,
logsDbPath: config.codexLogsDbPath,
sessionIndexPath: config.codexSessionIndexPath,
globalStatePath: config.codexGlobalStatePath,
sessionsDir: config.codexSessionsDir,
lookbackHours: config.codexSessionLookbackHours,
timeoutMs: discoveryTimeoutMs,
}),
});
if (discoveryResult.error) {
runtime.lastProjectDiscoveryAt = new Date().toISOString();
runtime.lastProjectDiscoveryOk = false;
runtime.lastProjectDiscoverySummary = discoveryResult.error instanceof Error
? discoveryResult.error.message
: String(discoveryResult.error);
runtime.lastCodexGuiConnected = discoveryResult.value.guiConnected === true;
postAppLog(config, runtime, {
level: "warning",
category: "local_agent.codex_discovery_degraded",
message: "Codex 线程扫描超时或失败，已使用缓存项目继续心跳。",
detail: runtime.lastProjectDiscoverySummary,
mirrorToMaster: false,
}).catch(() => null);
return discoveryResult.value;
}
const discovered = discoveryResult.value;
const candidateMap = new Map();
for (const candidate of [...staticCandidates, ...discovered.projectCandidates]) {
candidateMap.set(candidate.codexThreadRef ?? candidate.threadId, candidate);
@@ -101,22 +165,24 @@ async function resolveHeartbeatProjects(config, runtime) {
runtime.lastProjectDiscoveryOk = true;
runtime.lastProjectDiscoverySummary = `${mergedCandidates.length} threads / ${mergedProjects.length} folders`;
runtime.lastCodexGuiConnected = discovered.guiConnected === true;
return {
const heartbeatProjects = {
projects: mergedProjects,
projectCandidates: mergedCandidates,
guiConnected: discovered.guiConnected === true,
};
storeHeartbeatProjectsSnapshot(runtime, heartbeatProjects);
return heartbeatProjects;
} catch (error) {
runtime.lastProjectDiscoveryAt = new Date().toISOString();
runtime.lastProjectDiscoveryOk = false;
runtime.lastProjectDiscoverySummary = error instanceof Error ? error.message : String(error);
await postAppLog(config, runtime, {
postAppLog(config, runtime, {
level: "error",
category: "local_agent.codex_discovery_failed",
message: "Codex 线程扫描失败，已退回静态项目配置。",
detail: runtime.lastProjectDiscoverySummary,
mirrorToMaster: true,
});
}).catch(() => null);
return {
projects: staticProjects,
projectCandidates: staticCandidates,
@@ -252,53 +318,60 @@ async function postHeartbeat(config, runtime, heartbeatProjects) {
...mergedProjectCandidates.map((candidate) => candidate.folderName).filter(Boolean),
]),
];
const response = await fetch(`${config.controlPlaneUrl.replace(/\/$/, "")}/api/device-heartbeat`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
deviceId: config.deviceId,
token: runtime.issuedToken ?? config.token,
pairingCode: runtime.issuedToken ? undefined : config.pairingCode,
name: config.name,
avatar: config.avatar,
account: config.account,
status: config.status,
quota5h: config.quota5h,
quota7d: config.quota7d,
capabilities: {
gui: {
connected: guiConnected,
lastSeenAt: now,
lastActiveProjectId: "",
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/device-heartbeat`,
{
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
deviceId: config.deviceId,
token: runtime.issuedToken ?? config.token,
pairingCode: runtime.issuedToken ? undefined : config.pairingCode,
name: config.name,
avatar: config.avatar,
account: config.account,
status: config.status,
quota5h: config.quota5h,
quota7d: config.quota7d,
capabilities: {
gui: {
connected: guiConnected,
lastSeenAt: now,
lastActiveProjectId: "",
},
cli: {
connected: config.cliConnected !== false,
lastSeenAt: now,
lastActiveProjectId: "",
},
browserAutomation: {
connected: config.browserAutomationConnected !== false || Boolean(browserControlRuntime.enabled && browserControlRuntime.command),
lastSeenAt: now,
lastActiveProjectId: "",
},
computerUse: {
connected: computerUseConnected,
lastSeenAt: now,
lastActiveProjectId: "",
},
codexAppServer: {
connected: codexAppServerConnected,
lastSeenAt: now,
lastActiveProjectId: "",
metadata: codexAppServerMetadata,
},
},
cli: {
connected: config.cliConnected !== false,
lastSeenAt: now,
lastActiveProjectId: "",
},
browserAutomation: {
connected: config.browserAutomationConnected !== false || Boolean(browserControlRuntime.enabled && browserControlRuntime.command),
lastSeenAt: now,
lastActiveProjectId: "",
},
computerUse: {
connected: computerUseConnected,
lastSeenAt: now,
lastActiveProjectId: "",
},
codexAppServer: {
connected: codexAppServerConnected,
lastSeenAt: now,
lastActiveProjectId: "",
metadata: codexAppServerMetadata,
},
},
preferredExecutionMode,
projects: mergedProjects,
projectCandidates: mergedProjectCandidates,
endpoint: config.endpoint,
}),
});
preferredExecutionMode,
projects: mergedProjects,
projectCandidates: mergedProjectCandidates,
endpoint: config.endpoint,
}),
},
{
timeoutMs: config.heartbeatPostTimeoutMs ?? 4_000,
timeoutMessage: "LOCAL_AGENT_HEARTBEAT_POST_TIMEOUT",
},
);
const text = await response.text();
let json = null;
@@ -379,11 +452,48 @@ async function resolveCodexAppServerCapabilityConnected(codexAppServerRuntime) {
return canExecuteCommand(codexAppServerRuntime.command, codexAppServerRuntime.cwd || process.cwd());
}
function refreshCodexAppServerCapabilityMetadataInBackground(config, runtime, codexAppServerRuntime, now) {
if (runtime.codexAppServerCapabilityMetadataRefreshBusy) {
return;
}
runtime.codexAppServerCapabilityMetadataRefreshBusy = true;
runtime.codexAppServerCapabilityMetadataRefreshStartedAt = new Date(now).toISOString();
void (async () => {
try {
const metadata = await discoverCodexAppServerCapabilities(codexAppServerRuntime);
runtime.codexAppServerCapabilityMetadata = metadata;
runtime.codexAppServerCapabilityMetadataAtMs = Date.now();
runtime.codexAppServerCapabilityMetadataError = "";
runtime.codexAppServerCapabilityMetadataRefreshCompletedAt = new Date().toISOString();
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
runtime.codexAppServerCapabilityMetadataError = message;
runtime.codexAppServerCapabilityMetadataRefreshFailedAt = new Date().toISOString();
await postAppLog(config, runtime, {
level: "warn",
category: "local_agent.codex_app_server_capability_discovery_failed",
message: "Codex App Server 能力清单发现失败，设备心跳继续上报连接状态。",
detail: message,
mirrorToMaster: false,
});
} finally {
runtime.codexAppServerCapabilityMetadataRefreshBusy = false;
}
})();
}
async function resolveCodexAppServerCapabilityMetadata(config, runtime, codexAppServerRuntime, connected) {
if (!connected || !codexAppServerRuntime?.enabled || codexAppServerRuntime.discoveryEnabled === false) {
return undefined;
}
const now = Date.now();
const discoveryGuard = shouldSkipCodexAppServerDiscovery({ config, runtime });
if (discoveryGuard.skip) {
runtime.codexAppServerCapabilityMetadataSkippedAt = new Date(now).toISOString();
runtime.codexAppServerCapabilityMetadataSkipReason = discoveryGuard.reason;
runtime.codexAppServerCapabilityMetadataSkipTaskId = discoveryGuard.activeTaskId;
return runtime.codexAppServerCapabilityMetadata;
}
const ttlMs = codexAppServerRuntime.discoveryTtlMs ?? 300_000;
if (
runtime.codexAppServerCapabilityMetadata &&
@@ -393,24 +503,31 @@ async function resolveCodexAppServerCapabilityMetadata(config, runtime, codexApp
return runtime.codexAppServerCapabilityMetadata;
}
try {
const metadata = await discoverCodexAppServerCapabilities(codexAppServerRuntime);
runtime.codexAppServerCapabilityMetadata = metadata;
runtime.codexAppServerCapabilityMetadataAtMs = now;
runtime.codexAppServerCapabilityMetadataError = "";
return metadata;
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
runtime.codexAppServerCapabilityMetadataError = message;
await postAppLog(config, runtime, {
level: "warn",
category: "local_agent.codex_app_server_capability_discovery_failed",
message: "Codex App Server 能力清单发现失败，设备心跳继续上报连接状态。",
detail: message,
mirrorToMaster: false,
});
return runtime.codexAppServerCapabilityMetadata;
if (config.codexAppServerDiscoveryInlineInHeartbeat === true) {
try {
const metadata = await discoverCodexAppServerCapabilities(codexAppServerRuntime);
runtime.codexAppServerCapabilityMetadata = metadata;
runtime.codexAppServerCapabilityMetadataAtMs = now;
runtime.codexAppServerCapabilityMetadataError = "";
return metadata;
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
runtime.codexAppServerCapabilityMetadataError = message;
postAppLog(config, runtime, {
level: "warn",
category: "local_agent.codex_app_server_capability_discovery_failed",
message: "Codex App Server 能力清单发现失败，设备心跳继续上报连接状态。",
detail: message,
mirrorToMaster: false,
}).catch(() => null);
return runtime.codexAppServerCapabilityMetadata;
}
}
refreshCodexAppServerCapabilityMetadataInBackground(config, runtime, codexAppServerRuntime, now);
runtime.codexAppServerCapabilityMetadataSkippedAt = new Date(now).toISOString();
runtime.codexAppServerCapabilityMetadataSkipReason = "background_refresh";
return runtime.codexAppServerCapabilityMetadata;
}
function deviceTokenHeaders(config, runtime) {
@@ -420,7 +537,7 @@ function deviceTokenHeaders(config, runtime) {
async function postThreadContext(config, runtime, snapshot) {
const workerId = snapshot.workerId ?? config.workerId ?? `${config.deviceId}-worker`;
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/workers/${workerId}/thread-context`,
{
method: "POST",
@@ -452,6 +569,10 @@ async function postThreadContext(config, runtime, snapshot) {
capturedAt: new Date().toISOString(),
}),
},
{
timeoutMs: config.threadContextPostTimeoutMs ?? 3_000,
timeoutMessage: "LOCAL_AGENT_THREAD_CONTEXT_POST_TIMEOUT",
},
);
return {
@@ -525,7 +646,7 @@ async function discoverSkills(config) {
}
async function postSkills(config, runtime, skills) {
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/devices/${config.deviceId}/skills`,
{
method: "POST",
@@ -535,6 +656,10 @@ async function postSkills(config, runtime, skills) {
},
body: JSON.stringify({ skills }),
},
{
timeoutMs: config.skillsPostTimeoutMs ?? 3_000,
timeoutMessage: "LOCAL_AGENT_SKILLS_POST_TIMEOUT",
},
);
return {
@@ -547,17 +672,18 @@ async function postSkills(config, runtime, skills) {
async function postAppLog(config, runtime, payload) {
try {
await fetch(`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/app-logs`, {
method: "POST",
await postThroughReliableOutbox(config, {
kind: "app.log",
url: `${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/app-logs`,
headers: {
"Content-Type": "application/json",
...deviceTokenHeaders(config, runtime),
},
body: JSON.stringify({
body: {
deviceId: config.deviceId,
source: "local_agent",
...payload,
}),
},
});
} catch {
// Ignore log transport failures to avoid blocking the agent loop.
@@ -571,7 +697,7 @@ async function claimMasterAgentTask(config, runtime) {
const waitMs = Number.isFinite(configuredWaitMs)
? Math.max(0, Math.min(30_000, Math.floor(configuredWaitMs)))
: 25_000;
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/claim`,
{
method: "POST",
@@ -581,6 +707,10 @@ async function claimMasterAgentTask(config, runtime) {
},
body: JSON.stringify({ deviceId: config.deviceId, waitMs }),
},
{
timeoutMs: waitMs + Number(config.masterAgentClaimTimeoutPaddingMs ?? 5_000),
timeoutMessage: "LOCAL_AGENT_MASTER_TASK_CLAIM_TIMEOUT",
},
);
return {
@@ -591,52 +721,41 @@ async function claimMasterAgentTask(config, runtime) {
}
async function completeMasterAgentTask(config, runtime, payload) {
const response = await fetch(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/${payload.taskId}/complete`,
{
method: "POST",
headers: {
"Content-Type": "application/json",
...deviceTokenHeaders(config, runtime),
},
body: JSON.stringify(buildMasterAgentTaskCompletionRequestBody(config, payload)),
const result = await postThroughReliableOutbox(config, {
kind: "task.complete",
url: `${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/${payload.taskId}/complete`,
headers: {
"Content-Type": "application/json",
...deviceTokenHeaders(config, runtime),
},
);
body: buildMasterAgentTaskCompletionRequestBody(config, payload),
});
return {
ok: response.ok,
status: response.status,
body: await response.text(),
};
return result;
}
async function postMasterAgentTaskProgress(config, runtime, payload) {
const response = await fetch(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/${payload.taskId}/progress`,
{
method: "POST",
headers: {
"Content-Type": "application/json",
...deviceTokenHeaders(config, runtime),
},
body: JSON.stringify({
deviceId: config.deviceId,
status: payload.status || "running",
requestId: payload.requestId,
executionProgress: payload.executionProgress,
}),
const result = await postThroughReliableOutbox(config, {
kind: "task.progress",
url: `${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/${payload.taskId}/progress`,
headers: {
"Content-Type": "application/json",
...deviceTokenHeaders(config, runtime),
},
);
body: {
deviceId: config.deviceId,
status: payload.status || "running",
phase: payload.phase,
requestId: payload.requestId,
executionProgress: payload.executionProgress,
},
});
return {
ok: response.ok,
status: response.status,
body: await response.text(),
};
return result;
}
async function fetchMasterAgentTaskControlState(config, runtime, task) {
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/master-agent/tasks/${task.taskId}/control-state`,
{
method: "GET",
@@ -644,6 +763,10 @@ async function fetchMasterAgentTaskControlState(config, runtime, task) {
...deviceTokenHeaders(config, runtime),
},
},
{
timeoutMs: config.masterAgentControlStateTimeoutMs ?? 3_000,
timeoutMessage: "LOCAL_AGENT_MASTER_TASK_CONTROL_STATE_TIMEOUT",
},
);
if (!response.ok) {
return {
@@ -685,7 +808,7 @@ function buildCodexRemoteControlMaintenanceReply(task, result) {
}
async function claimSkillLifecycleRequest(config, runtime) {
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/devices/${config.deviceId}/skill-requests/claim`,
{
method: "POST",
@@ -695,6 +818,10 @@ async function claimSkillLifecycleRequest(config, runtime) {
},
body: JSON.stringify({ deviceId: config.deviceId }),
},
{
timeoutMs: config.skillLifecycleClaimTimeoutMs ?? 5_000,
timeoutMessage: "LOCAL_AGENT_SKILL_REQUEST_CLAIM_TIMEOUT",
},
);
return {
@@ -705,7 +832,7 @@ async function claimSkillLifecycleRequest(config, runtime) {
}
async function completeSkillLifecycleRequest(config, runtime, request, result) {
const response = await fetch(
const response = await fetchWithTimeout(
`${config.controlPlaneUrl.replace(/\/$/, "")}/api/v1/devices/${config.deviceId}/skill-requests/${request.requestId}/complete`,
{
method: "POST",
@@ -719,6 +846,10 @@ async function completeSkillLifecycleRequest(config, runtime, request, result) {
error: result.error,
}),
},
{
timeoutMs: config.skillLifecycleCompleteTimeoutMs ?? 5_000,
timeoutMessage: "LOCAL_AGENT_SKILL_REQUEST_COMPLETE_TIMEOUT",
},
);
return {
@@ -897,9 +1028,60 @@ async function runMasterAgentTask(config, runtime, task) {
status: "running",
startedAt: new Date().toISOString(),
};
const emitTaskPhase = async (phase, executionProgress) => {
try {
const result = await postMasterAgentTaskProgress(config, runtime, {
taskId: task.taskId,
status: "running",
phase,
executionProgress: {
...(executionProgress || {}),
phase,
},
});
return result;
} catch (error) {
return {
ok: false,
status: 0,
body: error instanceof Error ? error.message : String(error),
};
}
};
const createLongRunningProgressHeartbeat = ({ phase = "awaiting_reply", getProgress } = {}) => {
const intervalMs = normalizeLongRunningProgressIntervalMs(
config.masterAgentLongTaskProgressIntervalMs ?? config.masterAgentProgressHeartbeatIntervalMs,
);
if (intervalMs <= 0) {
return () => {};
}
const startedAtMs = Date.now();
let heartbeatCount = 0;
const sendHeartbeat = async () => {
heartbeatCount += 1;
await emitTaskPhase(
phase,
buildLongRunningCodexProgressSnapshot({
task,
phase,
startedAtMs,
nowMs: Date.now(),
baseProgress: typeof getProgress === "function" ? getProgress() : undefined,
heartbeatCount,
}),
);
};
const timer = setInterval(() => {
void sendHeartbeat();
}, intervalMs);
return () => {
clearInterval(timer);
};
};
try {
let activeChild = null;
await emitTaskPhase("executor_starting");
const executionResult = await (async () => {
if (canHandleCodexRemoteControlMaintenanceTask(task)) {
const daemonResult = await runCodexRemoteControlDaemonAction(
@@ -966,43 +1148,55 @@ async function runMasterAgentTask(config, runtime, task) {
const codexAppServerRunner = getCodexAppServerRunnerConfig(process.env, config);
if (shouldUseCodexAppServerTaskRunner(codexAppServerRunner, task)) {
const appServerResult = await executeCodexAppServerTask(
{
...codexAppServerRunner,
interruptPollIntervalMs: normalizeInterruptPollIntervalMs(config),
shouldInterruptActiveTurn: async () => {
const controlState = await fetchMasterAgentTaskControlState(config, runtime, task);
if (!controlState.ok) {
let latestCodexExecutionProgress;
const stopLongRunningProgressHeartbeat = createLongRunningProgressHeartbeat({
phase: "awaiting_reply",
getProgress: () => latestCodexExecutionProgress,
});
let appServerResult;
try {
appServerResult = await executeCodexAppServerTask(
{
...codexAppServerRunner,
interruptPollIntervalMs: normalizeInterruptPollIntervalMs(config),
shouldInterruptActiveTurn: async () => {
const controlState = await fetchMasterAgentTaskControlState(config, runtime, task);
if (!controlState.ok) {
return false;
}
if (controlState.body?.canceled === true || controlState.body?.status === "canceled") {
return {
interrupt: true,
reason: controlState.body?.cancelReason || "USER_CANCELED_TASK",
};
}
return false;
}
if (controlState.body?.canceled === true || controlState.body?.status === "canceled") {
return {
interrupt: true,
reason: controlState.body?.cancelReason || "USER_CANCELED_TASK",
};
}
return false;
},
onProgress: async (executionProgress) => {
const progressResult = await postMasterAgentTaskProgress(config, runtime, {
taskId: task.taskId,
status: "running",
executionProgress,
});
if (!progressResult.ok) {
await postAppLog(config, runtime, {
projectId: task.projectId,
level: "warn",
category: "local_agent.codex_app_server_progress_failed",
message: "Codex App Server 进度实时回写失败，完成回写仍会携带最终进度。",
detail: progressResult.body,
mirrorToMaster: false,
},
onProgress: async (executionProgress) => {
latestCodexExecutionProgress = executionProgress;
const progressResult = await postMasterAgentTaskProgress(config, runtime, {
taskId: task.taskId,
status: "running",
phase: "awaiting_reply",
executionProgress,
});
}
if (!progressResult.ok) {
await postAppLog(config, runtime, {
projectId: task.projectId,
level: "warn",
category: "local_agent.codex_app_server_progress_failed",
message: "Codex App Server 进度实时回写失败，完成回写仍会携带最终进度。",
detail: progressResult.body,
mirrorToMaster: false,
});
}
},
},
},
task,
);
task,
);
} finally {
stopLongRunningProgressHeartbeat();
}
if (appServerResult.status === "interrupted") {
return {
interruptedCompletion: {
@@ -1114,6 +1308,7 @@ async function runMasterAgentTask(config, runtime, task) {
},
async () =>
await new Promise((resolveTask, rejectTask) => {
void emitTaskPhase("turn_started");
const child = spawn("codex", codexExecution.args, {
cwd: codexExecution.cwd,
env: process.env,
@@ -1193,6 +1388,7 @@ async function runMasterAgentTask(config, runtime, task) {
return;
}
const { replyBody, dispatchExecutionCompletion, executionProgress } = executionResult;
await emitTaskPhase("completing", executionProgress);
const completion = await completeMasterAgentTask(
config,
@@ -1210,6 +1406,24 @@ async function runMasterAgentTask(config, runtime, task) {
executionProgress,
}),
);
if (!completion.ok) {
await emitTaskPhase("completing", {
...(executionProgress && typeof executionProgress === "object" ? executionProgress : {}),
title: "结果已生成，正在同步",
warnings: [
...(
Array.isArray(executionProgress?.warnings)
? executionProgress.warnings.filter(Boolean).slice(0, 6)
: []
),
{
id: "task-complete-sync-retrying",
severity: "warning",
message: "本机已生成任务结果，正在重试同步到 Boss 对话窗口。",
},
],
});
}
runtime.activeMasterTask = {
taskId: task.taskId,
status: completion.ok ? "completed" : "complete_failed",
@@ -1218,10 +1432,14 @@ async function runMasterAgentTask(config, runtime, task) {
};
await postAppLog(config, runtime, {
projectId: "master-agent",
level: "info",
category: "local_agent.master_agent_task_completed",
message: `Master Codex Node 已完成主 Agent 任务 ${task.taskId}。`,
detail: replyBody.slice(0, 280),
level: completion.ok ? "info" : "warn",
category: completion.ok
? "local_agent.master_agent_task_completed"
: "local_agent.master_agent_task_completion_sync_retrying",
message: completion.ok
? `Master Codex Node 已完成主 Agent 任务 ${task.taskId}。`
: `Master Codex Node 已生成结果，正在重试同步主 Agent 任务 ${task.taskId}。`,
detail: completion.ok ? replyBody.slice(0, 280) : completion.body,
mirrorToMaster: false,
});
} catch (error) {
@@ -1440,10 +1658,124 @@ const runtime = {
lastProjectDiscoveryAt: null,
lastProjectDiscoveryOk: false,
lastProjectDiscoverySummary: null,
lastReliableOutboxReplay: null,
};
function replayReliableOutboxInBackground(config, runtime) {
if (runtime.reliableOutboxReplayBusy) {
return;
}
runtime.reliableOutboxReplayBusy = true;
runtime.lastReliableOutboxReplayStartedAt = new Date().toISOString();
void replayReliableOutbox(config, {
limit: config.heartbeatOutboxReplayLimit ?? 5,
requestTimeoutMs: config.heartbeatOutboxRequestTimeoutMs ?? 1_000,
maxDurationMs: config.heartbeatOutboxReplayBudgetMs ?? 2_500,
})
.then((result) => {
runtime.lastReliableOutboxReplay = result;
runtime.lastReliableOutboxReplayAt = new Date().toISOString();
})
.catch((error) => {
runtime.lastReliableOutboxReplay = {
attempted: 0,
sent: 0,
retained: 0,
stoppedByBudget: false,
error: error instanceof Error ? error.message : String(error),
};
runtime.lastReliableOutboxReplayAt = new Date().toISOString();
})
.finally(() => {
runtime.reliableOutboxReplayBusy = false;
});
}
function syncThreadContextsInBackground(config, runtime, snapshots) {
if (runtime.threadContextSyncBusy || !Array.isArray(snapshots) || snapshots.length === 0) {
return;
}
runtime.threadContextSyncBusy = true;
runtime.lastThreadContextSyncStartedAt = new Date().toISOString();
void (async () => {
const results = [];
for (const snapshot of snapshots) {
let threadResult;
try {
threadResult = await postThreadContext(config, runtime, snapshot);
} catch (error) {
threadResult = {
ok: false,
status: 0,
body: error instanceof Error ? error.message : String(error),
workerId: snapshot.workerId ?? config.workerId ?? `${config.deviceId}-worker`,
threadId: snapshot.threadId,
};
}
results.push(threadResult);
if (!threadResult.ok) {
postAppLog(config, runtime, {
projectId: snapshot.projectId,
level: "error",
category: "local_agent.thread_context_failed",
message: `线程预算上报失败：${snapshot.threadId}`,
detail: threadResult.body,
mirrorToMaster: true,
}).catch(() => null);
}
}
runtime.lastThreadContextResults = results;
runtime.lastThreadContextSyncAt = new Date().toISOString();
})()
.catch((error) => {
runtime.lastThreadContextResults = [{
ok: false,
status: 0,
body: error instanceof Error ? error.message : String(error),
}];
runtime.lastThreadContextSyncAt = new Date().toISOString();
})
.finally(() => {
runtime.threadContextSyncBusy = false;
});
}
function syncSkillsInBackground(config, runtime) {
if (runtime.skillSyncBusy) {
return;
}
runtime.skillSyncBusy = true;
runtime.lastSkillSyncStartedAt = new Date().toISOString();
void (async () => {
const skills = await discoverSkills(config);
runtime.lastSkills = skills;
const skillSyncResult = await postSkills(config, runtime, skills);
runtime.lastSkillSyncAt = new Date().toISOString();
runtime.lastSkillSyncOk = skillSyncResult.ok;
runtime.lastSkillSyncStatus = skillSyncResult.status;
runtime.lastSkillSyncBody = skillSyncResult.body;
})()
.catch((error) => {
runtime.lastSkillSyncAt = new Date().toISOString();
runtime.lastSkillSyncOk = false;
runtime.lastSkillSyncStatus = 0;
runtime.lastSkillSyncBody = error instanceof Error ? error.message : String(error);
postAppLog(config, runtime, {
level: "error",
category: "local_agent.skills_sync_failed",
message: "Skill 扫描或同步失败。",
detail: runtime.lastSkillSyncBody,
mirrorToMaster: true,
}).catch(() => null);
})
.finally(() => {
runtime.skillSyncBusy = false;
});
}
async function performHeartbeat() {
try {
replayReliableOutboxInBackground(config, runtime);
const heartbeatProjects = await resolveHeartbeatProjects(config, runtime);
const result = await postHeartbeat(config, runtime, heartbeatProjects);
runtime.lastHeartbeatAt = new Date().toISOString();
@@ -1464,43 +1796,8 @@ async function performHeartbeat() {
}
const snapshots = Array.isArray(config.threadContexts) ? config.threadContexts : [];
runtime.lastThreadContextResults = [];
for (const snapshot of snapshots) {
const threadResult = await postThreadContext(config, runtime, snapshot);
runtime.lastThreadContextResults.push(threadResult);
if (!threadResult.ok) {
await postAppLog(config, runtime, {
projectId: snapshot.projectId,
level: "error",
category: "local_agent.thread_context_failed",
message: `线程预算上报失败：${snapshot.threadId}`,
detail: threadResult.body,
mirrorToMaster: true,
});
}
}
try {
const skills = await discoverSkills(config);
runtime.lastSkills = skills;
const skillSyncResult = await postSkills(config, runtime, skills);
runtime.lastSkillSyncAt = new Date().toISOString();
runtime.lastSkillSyncOk = skillSyncResult.ok;
runtime.lastSkillSyncStatus = skillSyncResult.status;
runtime.lastSkillSyncBody = skillSyncResult.body;
} catch (error) {
runtime.lastSkillSyncAt = new Date().toISOString();
runtime.lastSkillSyncOk = false;
runtime.lastSkillSyncStatus = 0;
runtime.lastSkillSyncBody = error instanceof Error ? error.message : String(error);
await postAppLog(config, runtime, {
level: "error",
category: "local_agent.skills_sync_failed",
message: "Skill 扫描或同步失败。",
detail: runtime.lastSkillSyncBody,
mirrorToMaster: true,
});
}
syncThreadContextsInBackground(config, runtime, snapshots);
syncSkillsInBackground(config, runtime);
} catch (error) {
runtime.lastHeartbeatAt = new Date().toISOString();
runtime.lastHeartbeatOk = false;
@@ -1516,7 +1813,10 @@ async function performHeartbeat() {
}
}
const heartbeat = createSerializedRunner(performHeartbeat);
const heartbeat = createSerializedRunner(performHeartbeat, {
timeoutMs: config.heartbeatTimeoutMs ?? 12_000,
timeoutErrorMessage: "LOCAL_AGENT_HEARTBEAT_TIMEOUT",
});
const masterTaskPoll = createSerializedRunner(async () => {
await pollMasterAgentTasks(config, runtime);
});
@@ -1646,13 +1946,19 @@ const server = createServer(async (request, response) => {
if (requestUrl.pathname === "/health") {
response.writeHead(200, { "Content-Type": "application/json" });
response.end(
JSON.stringify({
if (requestUrl.searchParams.get("verbose") === "1") {
response.end(
JSON.stringify({
ok: true,
service: "boss-local-agent",
runtime,
}),
);
return;
}
response.end(
JSON.stringify(buildLocalAgentHealthSummary(config, runtime)),
);
return;
}
@@ -1681,9 +1987,11 @@ const server = createServer(async (request, response) => {
}
if (requestUrl.pathname === "/api/v1/heartbeat" && request.method === "POST") {
await heartbeat();
await heartbeat().catch((error) => {
recordHeartbeatRunnerError(runtime, error);
});
response.writeHead(200, { "Content-Type": "application/json" });
response.end(JSON.stringify({ ok: runtime.lastHeartbeatOk, runtime }));
response.end(JSON.stringify(buildLocalAgentHealthSummary(config, runtime)));
return;
}
@@ -1704,14 +2012,18 @@ server.listen(config.port, config.bindHost, () => {
});
void (async () => {
await heartbeat();
await heartbeat().catch((error) => {
recordHeartbeatRunnerError(runtime, error);
});
await masterTaskPoll();
await skillLifecyclePoll();
await bossAgentOtaPoll();
})();
setInterval(() => {
void heartbeat();
void heartbeat().catch((error) => {
recordHeartbeatRunnerError(runtime, error);
});
}, config.heartbeatIntervalMs ?? 15000);
setInterval(() => {