@nowcrew/daemon 0.6.73 → 0.6.75

package/dist/runtimes/claude-models.js

@@ -0,0 +1,123 @@
+/**
+ * claude 模型清单探测:上报 CLI 族别名(fable/opus/sonnet/haiku,恒指本机已装版本的最新
+ * 对应模型),并把每个别名解析成具体模型 id 放进展示名——用户反馈"只写 latest 根本不知道
+ * 实际是什么模型"。
+ *
+ * 解析方式:claude 没有模型枚举命令,但 `--print` 会话启动时的 `system/init` 事件明文携带
+ * 解析后的 `model`(2.1.258 实测 opus→claude-opus-5、fable→claude-fable-5-1)。探测把
+ * ANTHROPIC_BASE_URL 指向本机不可达端口(零 API 调用、零 token),`--setting-sources=`
+ * 关掉用户 hooks/项目配置,读到 init 行即 SIGKILL。结果按 CLI 版本缓存:只有全部别名都
+ * 解析成功才缓存,失败的下次请求重试。
+ */
+import { execFile, spawn } from "node:child_process";
+import { tmpdir } from "node:os";
+import { createInterface } from "node:readline";
+import { promisify } from "node:util";
+import { isWin } from "../platform.js";
+const execFileP = promisify(execFile);
+const PROBE_TIMEOUT_MS = 8_000;
+// 本机丢弃端口:连不上也无所谓,init 事件在任何网络请求之前就已输出。
+const UNROUTABLE_API_BASE = "http://127.0.0.1:9";
+const CLAUDE_ALIASES = [
+    ["fable", "Fable"],
+    ["sonnet", "Sonnet"],
+    ["opus", "Opus"],
+    ["haiku", "Haiku"],
+];
+const DEFAULT_ALIAS = "sonnet";
+const cacheByVersion = new Map();
+/** "2.1.258 (Claude Code)" → "2.1.258";空/异常输出 → ""。 */
+export function normalizeClaudeVersion(raw) {
+    return raw.replace(/\s*\(Claude Code\)\s*$/, "").trim();
+}
+export async function listClaudeModels(deps = defaultDeps) {
+    const version = normalizeClaudeVersion(await deps.probeVersion());
+    const cached = version ? cacheByVersion.get(version) : undefined;
+    if (cached)
+        return cached;
+    const resolved = await Promise.all(CLAUDE_ALIASES.map(async ([alias]) => {
+        try {
+            return await deps.resolveAlias(alias);
+        }
+        catch {
+            return null;
+        }
+    }));
+    const suffix = version ? ` · Claude Code ${version}` : "";
+    const models = CLAUDE_ALIASES.map(([alias, name], index) => {
+        const concrete = resolved[index];
+        return {
+            id: alias,
+            label: `${name} (latest${concrete ? `: ${concrete}` : ""})${suffix}`,
+            ...(alias === DEFAULT_ALIAS ? { default: true } : {}),
+        };
+    });
+    if (version && resolved.every((model) => model !== null))
+        cacheByVersion.set(version, models);
+    return models;
+}
+/** 仅测试用:清空版本缓存。 */
+export function resetClaudeModelCache() {
+    cacheByVersion.clear();
+}
+async function probeVersion() {
+    const { stdout } = await execFileP("claude", ["--version"], {
+        shell: isWin(),
+        timeout: PROBE_TIMEOUT_MS,
+        killSignal: "SIGKILL",
+    });
+    return stdout;
+}
+function resolveAlias(alias) {
+    return new Promise((resolve) => {
+        const child = spawn("claude", [
+            "--model", alias,
+            "--print", "--output-format", "stream-json", "--verbose",
+            "--setting-sources=",
+            "ping",
+        ], {
+            cwd: tmpdir(),
+            env: {
+                ...process.env,
+                ANTHROPIC_BASE_URL: UNROUTABLE_API_BASE,
+                CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC: "1",
+            },
+            shell: isWin(),
+            stdio: ["ignore", "pipe", "ignore"],
+        });
+        let settled = false;
+        const finish = (value) => {
+            if (settled)
+                return;
+            settled = true;
+            clearTimeout(timer);
+            try {
+                child.kill("SIGKILL");
+            }
+            catch { /* 已退出 */ }
+            resolve(value);
+        };
+        const timer = setTimeout(() => finish(null), PROBE_TIMEOUT_MS);
+        child.on("error", () => finish(null));
+        child.on("exit", () => finish(null));
+        const lines = createInterface({ input: child.stdout });
+        lines.on("line", (line) => {
+            const model = initModelOf(line);
+            if (model !== null)
+                finish(model);
+        });
+    });
+}
+/** stream-json 一行 → system/init 事件里的 model;其他行返回 null。 */
+export function initModelOf(line) {
+    try {
+        const event = JSON.parse(line);
+        if (event.type !== "system" || event.subtype !== "init")
+            return null;
+        return typeof event.model === "string" && event.model.trim() ? event.model.trim() : null;
+    }
+    catch {
+        return null;
+    }
+}
+const defaultDeps = { probeVersion, resolveAlias };

package/dist/runtimes/codex-app-server-runner.js

@@ -31,6 +31,7 @@ const ERROR_MESSAGE_CAP = 2_000;
 const STDERR_TAIL_CAP = 1_200;
 const STDERR_LINE_CAPTURE_CAP = 1_200;
 const STDERR_LINE_OMITTED = "[stderr line omitted: exceeded capture limit]\n";
+const API_RETRY_DETAIL_CAP = 240;
 const MAX_INITIALIZE_ATTEMPTS = 2;
 // 模型网关瞬态故障(过载/限流)导致 turn 失败时,整轮重试(15s→45s 递进退避):codex 自身
 // 的重试窗口只有 ~10-30s,网关过载往往持续数分钟,这里再兜一层,否则 agent 直接失败
@@ -188,6 +189,21 @@ function logStage(stage, status, attempt, startedAt, detail) {
     process.stderr.write(`[codex-app-server] stage=${stage} status=${status} attempt=${attempt}`
         + ` elapsed_ms=${Date.now() - startedAt}${suffix}\n`);
 }
+function logApiRetry(turnAttempt, apiRetryAttempt, detail, secrets) {
+    const safeDetail = detail === undefined
+        ? undefined
+        : safeErrorMessage(detail, secrets).slice(0, API_RETRY_DETAIL_CAP);
+    process.stderr.write(`[codex-app-server] stage=api_retry status=retry attempt=${apiRetryAttempt}`
+        + ` turn_attempt=${turnAttempt}`
+        + (safeDetail === undefined ? "" : ` error=${JSON.stringify(safeDetail)}`)
+        + "\n");
+}
+function logApiRetrySummary(turnAttempt, apiRetryCount) {
+    if (apiRetryCount === 0)
+        return;
+    process.stderr.write(`[codex-app-server] stage=api_retry status=summary turn_attempt=${turnAttempt}`
+        + ` api_retry_count=${apiRetryCount}\n`);
+}
 function jsonLine(event) {
     if (process.stdout.write(`${JSON.stringify(event)}\n`))
         return Promise.resolve();
@@ -445,6 +461,7 @@ async function runCodexAppServerAttempt(bin, input, attempt, initializeTimeoutMs
     let threadId = null;
     let announcedThreadId = null;
     let turnId = null;
+    let apiRetryCount = 0;
     let turnHasUnsafeSideEffect = false;
     const startedItemIds = new Set();
     let lastUsage;
@@ -489,10 +506,17 @@ async function runCodexAppServerAttempt(bin, input, attempt, initializeTimeoutMs
         }
         if (method === "turn/completed")
             completionResolve(params);
-        if (method === "error" && params.willRetry === false) {
-            const detail = params.error?.message;
-            if (detail)
-                process.stderr.write(sanitizeCodexDiagnostic(`Codex turn error: ${detail.slice(0, ERROR_MESSAGE_CAP)}\n`, secrets));
+        if (method === "error") {
+            const notification = (params !== null && typeof params === "object" ? params : {});
+            const detail = notification.error?.message;
+            if (notification.willRetry === true) {
+                apiRetryCount += 1;
+                logApiRetry(attempt, apiRetryCount, detail, secrets);
+            }
+            else if (notification.willRetry === false && detail) {
+                process.stderr.write(sanitizeCodexDiagnostic(`Codex turn error: ${detail.slice(0, ERROR_MESSAGE_CAP)}`
+                    + ` turn_attempt=${attempt} api_retry_count=${apiRetryCount}\n`, secrets));
+            }
         }
     }, completionReject, secrets);
     let treeStopPromise = null;
@@ -608,10 +632,12 @@ async function runCodexAppServerAttempt(bin, input, attempt, initializeTimeoutMs
                 },
             }),
         });
+        logApiRetrySummary(attempt, apiRetryCount);
         if (completed.turn?.status === "completed")
             return { code: 0, initializeTimedOut: false, unsafeToRetry: false };
         const detail = completed.turn?.error?.message ?? `turn status ${completed.turn?.status ?? "unknown"}`;
-        process.stderr.write(sanitizeCodexDiagnostic(`Codex turn failed: ${detail.slice(0, ERROR_MESSAGE_CAP)}\n`, secrets));
+        process.stderr.write(sanitizeCodexDiagnostic(`Codex turn failed: ${detail.slice(0, ERROR_MESSAGE_CAP)}`
+            + ` turn_attempt=${attempt} api_retry_count=${apiRetryCount}\n`, secrets));
         const status = completed.turn?.status;
         const retryReason = cancelling ? undefined
             : status === "interrupted" ? "collab_interrupted"
@@ -633,7 +659,8 @@ async function runCodexAppServerAttempt(bin, input, attempt, initializeTimeoutMs
         else {
             logStage(activeStage, "error", attempt, stageStartedAt);
         }
-        process.stderr.write(`Codex app-server execution failed: ${safeErrorMessage(error, secrets)}\n`);
+        process.stderr.write(`Codex app-server execution failed: ${safeErrorMessage(error, secrets)}`
+            + ` turn_attempt=${attempt} api_retry_count=${apiRetryCount}\n`);
         return { code: 1, initializeTimedOut, unsafeToRetry: false };
     }
     finally {

package/dist/serve.js

@@ -5,10 +5,11 @@
 import { WebSocket } from "ws";
 import { join } from "node:path";
 import { randomUUID } from "node:crypto";
-import { initSlog, dslog, setSlogDefaults, drainSpool, flushSlog } from "./slog.js";
+import { dslog, setSlogDefaults, drainSpool, flushSlog } from "./slog.js";
 import { mergeRunAgentResults, reportScheduledStartFailure, runAgent } from "./runner.js";
 import { buildOriginDecisionRetryPrompt, buildScheduledPrompt } from "./prompt.js";
-import { collectMachineHello, cliVersion, daemonCapabilities, daemonVersion, detectExecutionRuntimesWithSignal, } from "./machine-info.js";
+import { collectMachineHello, daemonCapabilities, detectExecutionRuntimesWithSignal, } from "./machine-info.js";
+import { initReleaseSlog, reportDaemonConnectionClose, reportReleaseError } from "./release-channel-guard.js";
 import { conservativeExecutionRuntimes, createRuntimeProbeCoordinator, } from "./runtime-probe.js";
 import { listWorkspace, readWorkspaceFile } from "./workspace-fs.js";
 import { listSkills } from "./skills.js";
@@ -311,7 +312,7 @@ export function serve(config, opts = {}) {
             reservation.release();
     };
     let connectedAt = 0; // 本次 WS 连接建立时刻(断开日志算在线时长用)
-    initSlog(config.serverUrl, config.machineToken, { daemonVersion: daemonVersion(), cliVersion: cliVersion(), ...(opts.profileName === undefined ? {} : { profileName: opts.profileName }), agentsRoot: config.agentsRoot });
+    initReleaseSlog({ serverUrl: config.serverUrl, machineToken: config.machineToken, ...(opts.profileName === undefined ? {} : { profileName: opts.profileName }), agentsRoot: config.agentsRoot });
     dslog("daemon.start", "daemon 常驻模式启动", { server_url: config.serverUrl, runtime: config.runtimeBin });
     // 并行调度:同一 agent 可并行处理多个【不同任务】(线程/频道),每任务隔离 cwd+work-log。
     // scheduled 重复 run 仍由 running 去重;普通同线程 wake 用 legacyTaskTails 串成 FIFO。
@@ -854,6 +855,8 @@ export function serve(config, opts = {}) {
                     dslog("daemon.ws_auth_rejected", "控制面拒绝鉴权:机器凭证无效或已吊销", { level: "ERROR" });
                     backoff = maxBackoff; // 退避到最大,停止每秒重连刷屏
                 }
+                if (reportReleaseError(msg.code, log))
+                    backoff = maxBackoff;
                 return;
             }
             if (msg.type === "project:add"
@@ -1309,11 +1312,8 @@ export function serve(config, opts = {}) {
             completionRetransmitter.pause();
             // 4001 = 控制面应用级「鉴权失败」关闭码(见 server control-plane.ts)。即使上面的
             // error 帧因 close 抢先而丢失,也能据关闭码识别这是凭证失效——退避拉满,不再每秒热循环。
-            if (code === 4001) {
+            if (reportDaemonConnectionClose(code, log))
                 backoff = maxBackoff;
-                log(`🛑 控制面以鉴权失败关闭连接 (code 4001):机器凭证无效或已吊销。`);
-                log(`   请在 NowWork 重新 "Add Computer" 获取新连接命令再重跑(当前 --api-key 已失效)。`);
-            }
             log(`🔁 控制面断开,${Math.round(backoff / 1000)}s 后重连`);
             // 此刻 server 大概率不可达 → 这条会落 spool,重连 drainSpool 时补传;
             // ts 是现在(断开时刻),排查「daemon 为什么断」以它对齐 server 侧 machine_disconnected。

package/dist/slog.js

@@ -59,6 +59,10 @@ export function initSlog(serverUrl, machineToken, identity = {}) {
         pid: process.pid,
         daemon_version: identity.daemonVersion,
         cli_version: identity.cliVersion,
+        release_channel: identity.releaseChannel,
+        release_tag: identity.releaseTag,
+        daemon_git_sha: identity.gitSha,
+        daemon_artifact_digest: identity.artifactDigest,
         daemon_profile: identity.profileName,
         server_host: serverHost,
         env: deploymentEnvironment(serverHost),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nowcrew/daemon",
-  "version": "0.6.73",
+  "version": "0.6.75",
   "type": "module",
   "description": "crew daemon — 运行在用户机器:拉起/管理 agent 进程,注入 crew CLI,归一化 runtime 事件",
   "license": "Apache-2.0",
@@ -18,8 +18,9 @@
   },
   "dependencies": {
     "@agentclientprotocol/sdk": "1.2.1",
-    "@nowcrew/cli": "^0.4.13",
+    "@nowcrew/cli": "0.4.13",
     "cross-spawn": "^7.0.6",
+    "semver": "7.8.5",
     "ws": "^8",
     "yaml": "^2.8.1",
     "zod": "^3.23.0"
@@ -30,6 +31,7 @@
   "devDependencies": {
     "@types/cross-spawn": "^6.0.6",
     "@types/node": "^22.0.0",
+    "@types/semver": "7.7.1",
     "@types/ws": "^8",
     "tsx": "^4.19.0",
     "typescript": "^5.6.0",