@tea-agent/loop-agent 0.34.1 → 0.34.2

package/CHANGELOG.md

@@ -2,6 +2,21 @@
 
 ## [Unreleased]
 
+## [0.34.2] - 2026-08-12
+
+### 重点更新
+
+- 修复控制台 DAG 执行在 120 秒后被强制终止导致状态卡死的问题，默认超时现提升至 4 小时
+
+### 改进
+
+- DAG 挂钟时间常量统一收敛至 src/shared 共享模块管理，移除了 workflows 中不必要的内核导入
+
+### 修复
+
+- 修复控制台任务导航执行 DAG 时因 LoopAgentClient 默认 120 秒超时被强制终止的问题，现已将默认超时提升至 4 小时，并支持通过 --default-timeout-ms 参数覆盖
+- 修正自定义模型合并报告中的文档描述错误
+
 ## [0.34.1] - 2026-08-11
 
 ### 重点更新

package/dist/shared/timeout-policy.js

@@ -1,5 +1,10 @@
 import { DEFAULT_TIMEOUT_MS, MAX_ALLOWED_TIMEOUT_MS } from "../executors/pi-executor.js";
 export { MAX_ALLOWED_TIMEOUT_MS };
+/**
+ * DAG runner 绝对最大墙钟（4h）。不可被空心跳续期；
+ * 是 DEFAULT_LIVENESS_POLICY 与 Console LoopAgentClient 默认超时的单一真源。
+ */
+export const DAG_ABSOLUTE_MAX_WALL_CLOCK_MS = 14_400_000;
 /**
  * Clamp timeout values for Pi steps. Undefined or non-positive values fall back to
  * DEFAULT_TIMEOUT_MS so task-level invalid values do not disable Pi execution timeouts.

package/dist/worker/cli.js

@@ -720,6 +720,7 @@ export function buildAgentWorkerProgram() {
         .option("--host <host>", "Bind host (default 127.0.0.1; use 0.0.0.0 for LAN, no network auth)", "127.0.0.1")
         .option("--debug", "Log request source/method/path/status to stderr")
         .option("--app-data <path>", "Override Console application-data root (Draft/Operation/confirmation)")
+        .option("--default-timeout-ms <ms>", "LoopAgentClient hard timeout for Console operations (ms; default 14400000 / 4h; 0 disables)")
         .action(async (options) => {
         await runConsoleServe({
             repoRoot: path.resolve(options.repo),
@@ -729,6 +730,7 @@ export function buildAgentWorkerProgram() {
             appDataRoot: options.appData
                 ? path.resolve(options.appData)
                 : undefined,
+            defaultTimeoutMsRaw: options.defaultTimeoutMs,
         });
     });
     consoleCmd
@@ -738,6 +740,7 @@ export function buildAgentWorkerProgram() {
         .option("--host <host>", "Bind host (default 127.0.0.1; use 0.0.0.0 for LAN, no network auth)", "127.0.0.1")
         .option("--debug", "Log request source/method/path/status to stderr")
         .option("--app-data <path>", "Override Console application-data root (Draft/Operation/confirmation)")
+        .option("--default-timeout-ms <ms>", "LoopAgentClient hard timeout for Console operations (ms; default 14400000 / 4h; 0 disables)")
         .description("Start Loop Operator Console HTTP server (compat entry; equivalent to bare `agent-worker console`)")
         .action(async (options) => {
         await runConsoleServe({
@@ -748,6 +751,7 @@ export function buildAgentWorkerProgram() {
             appDataRoot: options.appData
                 ? path.resolve(options.appData)
                 : undefined,
+            defaultTimeoutMsRaw: options.defaultTimeoutMs,
         });
     });
     consoleCmd
@@ -835,6 +839,17 @@ export async function main(argv = process.argv) {
 async function readTaskYaml(taskSpecPath) {
     return YAML.parse(await readFile(taskSpecPath, "utf-8"));
 }
+/** Parse optional `--default-timeout-ms`; undefined keeps Console default. */
+function parseOptionalTimeoutMs(raw) {
+    if (raw === undefined || raw.trim() === "") {
+        return undefined;
+    }
+    const value = Number(raw);
+    if (!Number.isFinite(value) || value < 0 || !Number.isInteger(value)) {
+        throw new Error(`--default-timeout-ms must be a non-negative integer (ms); got ${JSON.stringify(raw)}`);
+    }
+    return value;
+}
 /** Shared console serve entry used by both bare `console` and `console serve`. */
 async function runConsoleServe(options) {
     const { createConsoleServer } = await import("./console/server.js");
@@ -845,6 +860,7 @@ async function runConsoleServe(options) {
             port: options.port,
             debug: options.debug === true,
             appDataRoot: options.appDataRoot,
+            defaultTimeoutMs: parseOptionalTimeoutMs(options.defaultTimeoutMsRaw),
         });
         process.stdout.write(`${server.url}\n`);
         await new Promise((resolve) => {

package/dist/worker/console/server.js

@@ -4,6 +4,7 @@ import { createServer } from "node:http";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
 import { LoopAgentClient } from "../loop-agent/loop-agent-client.js";
+import { DAG_ABSOLUTE_MAX_WALL_CLOCK_MS } from "../../shared/timeout-policy.js";
 import { openConsoleAppData } from "./app-data.js";
 import { DagConfirmationStore } from "./dag-confirmation.js";
 import { DagExecutionReceiptStore } from "./dag-execution-receipt.js";
@@ -24,6 +25,30 @@ import { resolveSiblingLoopAgentBin } from "./sibling-controller.js";
 import { NightAuxTicker } from "./night-aux-ticker.js";
 import { deriveObserveRouteCapabilities } from "../observe/health.js";
 import { createObserveRouteContext, defaultObserveStaticDir, handleMatchedObserveRoute, matchObserveGetRoute, ROUTES, serveObserveStatic, } from "../observe/routes.js";
+/**
+ * Default hard timeout for the Console-owned LoopAgentClient.
+ *
+ * Console task-nav operations (`dag execute`, `task advance --approve-gate`,
+ * etc.) spawn a long-lived sibling runner via the operation runner. The
+ * worker client hard-kills children after `defaultTimeoutMs` (default 120s),
+ * which is far below multi-node DAG wall clocks and leaves `state.json`
+ * stuck in `running` with no error. Align with shared `DAG_ABSOLUTE_MAX_WALL_CLOCK_MS`
+ * (4h) so the client is a safety net, not a false deadline.
+ */
+export const DEFAULT_CONSOLE_CLIENT_TIMEOUT_MS = DAG_ABSOLUTE_MAX_WALL_CLOCK_MS;
+/**
+ * Resolve the Console LoopAgentClient default timeout.
+ * `0` disables the client hard timeout (DAG kernel owns wall clock).
+ */
+export function resolveConsoleDefaultTimeoutMs(override) {
+    if (override === undefined) {
+        return DEFAULT_CONSOLE_CLIENT_TIMEOUT_MS;
+    }
+    if (!Number.isFinite(override) || override < 0) {
+        throw new Error(`console serve: invalid defaultTimeoutMs ${String(override)} (expected >= 0)`);
+    }
+    return override;
+}
 /** Built static assets live next to the compiled server under dist/worker/console/static. */
 export function defaultConsoleStaticDir(fromFileUrl = import.meta.url) {
     return path.join(path.dirname(fileURLToPath(fromFileUrl)), "static");
@@ -74,6 +99,9 @@ export async function createConsoleServer(options) {
             loopAgentBin: resolveSiblingLoopAgentBin(),
             artifactRoot,
             resolveIdentity: true,
+            // See DEFAULT_CONSOLE_CLIENT_TIMEOUT_MS: dag execute / task advance
+            // must not inherit the 120s worker-command default.
+            defaultTimeoutMs: resolveConsoleDefaultTimeoutMs(options.defaultTimeoutMs),
         });
     if (!options.skipReconcile) {
         await operations.reconcileOnBoot();

package/dist/workflows/dag/liveness-policy.js

@@ -1,4 +1,5 @@
 import { z } from "zod";
+import { DAG_ABSOLUTE_MAX_WALL_CLOCK_MS } from "../../shared/timeout-policy.js";
 /**
  * Adaptive liveness policy for DAG/Pi supervision.
  *
@@ -15,7 +16,7 @@ export const DEFAULT_LIVENESS_POLICY = {
     stallProbeMs: 900_000,
     abortGraceMs: 30_000,
     /** 4h absolute max wall clock — cannot be renewed by empty heartbeats. */
-    absoluteMaxWallClockMs: 14_400_000,
+    absoluteMaxWallClockMs: DAG_ABSOLUTE_MAX_WALL_CLOCK_MS,
 };
 export const dagLivenessPolicySchema = z
     .object({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tea-agent/loop-agent",
-  "version": "0.34.1",
+  "version": "0.34.2",
   "type": "module",
   "bin": {
     "loop-agent": "bin/loop-agent.js",