@co0ontty/wand 4.85.1 → 4.87.0-beta.gd2f0f8f

package/dist/server-session-routes.js

@@ -4,7 +4,7 @@ import { getDefaultModelForProvider, isExecutionMode } from "./config.js";
 import { defaultRoleForCli } from "./default-employee.js";
 import { defaultModelGroupSelector } from "./model-groups.js";
 import { inspectPiExecution } from "./pi-execution.js";
-import { alignedBlockStart, blockWindowMessagesForTransport, compactToolMessagesForTransport, sliceTurnBlocksForTransport, truncateMessagesForTransport, visibleBlockCount, windowMessagesForTransport } from "./message-truncator.js";
+import { alignedBlockStart, blockWindowMessagesForTransport, compactToolMessagesForTransport, messageWindowByteBudget, sliceTurnBlocksForTransport, truncateMessagesForTransport, visibleBlockCount, windowMessagesForTransport } from "./message-truncator.js";
 import { toSessionDetailDTO, toSessionListItemDTO } from "./session-transport.js";
 import { checkSessionWorktreeMergeabilityAsync, cleanupSessionWorktreeAsync, getWorktreeMergeErrorCode, mergeSessionWorktreeAsync, WorktreeMergeError, } from "./git-worktree.js";
 import { resolveSessionCwd } from "./session-cwd.js";
@@ -173,6 +173,7 @@ function sessionListRevision(entries, directoryNames = []) {
         entry.session.completionRevision ?? 0,
         entry.session.viewedCompletionRevision ?? 0,
         entry.session.ptyBusy === true,
+        entry.session.ptyCommandRunning === true,
         entry.session.structuredState?.inFlight === true,
     ]);
     const revisionState = directoryNames.length === 0
@@ -489,6 +490,21 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
             const provider = requestedProvider
                 ?? employeeAgent?.provider
                 ?? (isSessionProvider(body.provider) ? body.provider : config.defaultProvider ?? "claude");
+            // 引擎是第二个维度：Pi CLI（cli）与 Wand Agent（sdk）都是 provider=pi，但执行方式不同。
+            // 员工候选自己带引擎；只有直选 CLI（含员工面板里的「指定 CLI」）才接受请求里的 engine。
+            const rawEngine = typeof body.engine === "string" ? body.engine.trim() : "";
+            if (rawEngine && rawEngine !== "cli" && rawEngine !== "sdk") {
+                res.status(400).json({ error: "请选择有效的执行引擎。" });
+                return;
+            }
+            const requestedEngine = (employee && !requestedProvider) || !rawEngine
+                ? undefined
+                : rawEngine;
+            if (requestedEngine === "sdk" && provider !== "pi") {
+                res.status(400).json({ error: "Wand Agent 只支持 Pi 结构化会话。" });
+                return;
+            }
+            const engineDecision = requestedEngine ? structured.resolveNewSessionPiEngine(requestedEngine) : null;
             const rawModel = typeof body.model === "string" ? body.model.trim() : "";
             const origin = parseSessionCreationOrigin(body);
             // Explicit employees/custom role rules and internal automation retain their own role.
@@ -523,6 +539,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
                 employeeAvatar: role?.avatar,
                 employeeCandidates: employee?.agents,
                 employeeCandidateIndex: requestedProvider ? undefined : selectedEmployeeCandidate?.index,
+                ...(engineDecision ? { engine: engineDecision.engine } : {}),
                 ...origin,
             });
             onSessionCreated?.(snapshot.cwd);
@@ -582,8 +599,17 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
     });
     app.post("/api/sessions/:id/provider", (req, res) => {
         const provider = req.body?.provider;
+        const engine = req.body?.engine;
+        if (engine !== undefined && engine !== "cli" && engine !== "sdk") {
+            res.status(400).json({ error: "执行引擎必须是 cli 或 sdk。" });
+            return;
+        }
+        if (engine === "sdk" && provider !== "pi") {
+            res.status(400).json({ error: "Wand Agent 只支持 Pi 结构化会话。" });
+            return;
+        }
         if (!isSessionProvider(provider)) {
-            res.status(400).json({ error: "请选择有效的 CLI 工具。" });
+            res.status(400).json({ error: "请选择有效的执行工具。" });
             return;
         }
         const id = req.params.id;
@@ -597,7 +623,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
             return;
         }
         try {
-            res.json(sessionResponseDTO(structured.setSessionProvider(id, provider), req));
+            res.json(sessionResponseDTO(structured.setSessionProvider(id, provider, engine), req));
         }
         catch (error) {
             sendRouteError(res, error, "切换工具失败。");
@@ -614,7 +640,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
         const localDecisionAvailable = config.localDecision?.enabled === true;
         const inventory = await discoverPiResources(config, snapshot.cwd);
         const resourceCatalog = { ...inventory.catalog, supported: modifiable && !sdk && inventory.catalog.supported,
-            reason: archived ? "请先恢复已归档的会话。" : sdk ? "当前 SDK 候选尚不支持 Skills / MCP 会话选择。" : inventory.catalog.reason };
+            reason: archived ? "请先恢复已归档的会话。" : sdk ? "Wand Agent 尚不支持 Skills / MCP 会话选择。" : inventory.catalog.reason };
         const controls = {
             // 基础工具与 CodeMode 在两种引擎下都能表达（CLI 通过 --tools 白名单）。
             tools: modifiable,
@@ -1346,7 +1372,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
         }
         const transcriptOutput = (snapshot.sessionKind ?? "pty") === "pty"
             ? processes.getPtyTranscript(snapshot.id) ?? snapshot.output
-            : snapshot.output;
+            : req.query.format === "chat" && req.query.output === "omit" ? "" : snapshot.output;
         const enriched = enrichStructuredMessages(snapshot.messages ?? [], snapshot.id);
         const messages = wantsCompactTools(req) ? compactToolMessagesForTransport(enriched) : enriched;
         if (req.query.format === "chat") {
@@ -1355,7 +1381,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
             const rawBudget = req.query.blockBudget;
             if (typeof rawBudget === "string" && /^\d+$/.test(rawBudget) && Number(rawBudget) > 0) {
                 const blockBudget = parseBoundedInteger(rawBudget, 1, 1, 2_000);
-                const windowed = blockWindowMessagesForTransport(messages, config.cardDefaults ?? {}, blockBudget);
+                const windowed = blockWindowMessagesForTransport(messages, config.cardDefaults ?? {}, blockBudget, messageWindowByteBudget(req.query.byteBudget), typeof req.query.byteBudget === "string");
                 res.json(toSessionDetailDTO(snapshot, {
                     output: transcriptOutput,
                     messages: windowed.messages,
@@ -1398,7 +1424,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
             && typeof req.query.blockBudget === "string" && /^\d+$/.test(req.query.blockBudget)) {
             const before = parseBoundedInteger(req.query.before, total, 0, total);
             const budget = parseBoundedInteger(req.query.blockBudget, 60, 1, 200);
-            const windowed = blockWindowMessagesForTransport(all.slice(0, before), config.cardDefaults ?? {}, budget);
+            const windowed = blockWindowMessagesForTransport(all.slice(0, before), config.cardDefaults ?? {}, budget, messageWindowByteBudget(req.query.byteBudget), typeof req.query.byteBudget === "string");
             res.json({ wandProtocolVersion: WAND_PROTOCOL_VERSION, ...windowed,
                 offset: windowed.messageOffset, total });
             return;
@@ -1422,7 +1448,7 @@ export function registerSessionRoutes(app, processes, structured, storage, defau
             const blockEnd = Math.min(Math.max(Number.isFinite(rawBlockOffset) ? rawBlockOffset : blockTotal, 0), blockTotal);
             const blockStart = Math.max(0, blockEnd - blockLimit);
             const cardDefaults = config.cardDefaults ?? {};
-            const startOffset = alignedBlockStart(turn.content, cardDefaults, blockStart, blockEnd);
+            const startOffset = alignedBlockStart(turn.content, cardDefaults, blockStart, blockEnd, Math.min(128 * 1024, messageWindowByteBudget(req.query.byteBudget)));
             const blocks = enrichStructuredMessages([{
                     ...turn,
                     content: sliceTurnBlocksForTransport(turn, startOffset, blockEnd, cardDefaults),

package/dist/server-settings-routes.js

@@ -127,6 +127,8 @@ export function registerSettingsRoutes(app, deps) {
         };
         let touchedDeployField = false;
         try {
+            if (Object.hasOwn(body, "localDecision"))
+                throw new Error("请使用本地模型设置管理 LAYA；通用配置不接受运行时路径或命令。");
             if (body.modelGroups !== undefined && body.expectedModelGroups !== undefined
                 && JSON.stringify(normalizeModelGroups(body.expectedModelGroups)) !== JSON.stringify(config.modelGroups ?? [])) {
                 res.status(409).json({ error: "模型分组已在其他设备修改，请先刷新；当前草稿已保留。" });

package/dist/server-speech-routes.d.ts

@@ -0,0 +1,8 @@
+import { type Express, type RequestHandler } from "express";
+import { type SpeechService } from "./speech-service.js";
+/** Mounted behind /api authentication. Clients may transcribe, only admins configure/download. */
+export declare function registerSpeechRoutes(app: Express, deps: {
+    speech: SpeechService;
+    requireSessions: RequestHandler;
+    requireAdmin: RequestHandler;
+}): void;

package/dist/server-speech-routes.js

@@ -0,0 +1,87 @@
+import express from "express";
+import { SpeechError } from "./speech-service.js";
+import { SPEECH_MAX_WAV_BYTES } from "./speech-types.js";
+/** Mounted behind /api authentication. Clients may transcribe, only admins configure/download. */
+export function registerSpeechRoutes(app, deps) {
+    const noStore = (_req, res, next) => { res.set("Cache-Control", "no-store"); next(); };
+    app.use("/api/speech", noStore, deps.requireSessions);
+    app.get("/api/speech/status", async (_req, res, next) => {
+        try {
+            res.json(await deps.speech.status());
+        }
+        catch (error) {
+            next(error);
+        }
+    });
+    app.patch("/api/speech/settings", deps.requireAdmin, async (req, res, next) => {
+        try {
+            deps.speech.configure(req.body);
+            res.json(await deps.speech.status());
+        }
+        catch (error) {
+            next(error);
+        }
+    });
+    app.post("/api/speech/models/:id/download", deps.requireAdmin, async (req, res, next) => {
+        try {
+            await deps.speech.startDownload(req.params.id);
+            res.status(202).json(await deps.speech.status());
+        }
+        catch (error) {
+            next(error);
+        }
+    });
+    // Bound raw uploads as well as inference; don't buffer unbounded concurrent audio requests.
+    let uploads = 0;
+    const admission = (req, res, next) => {
+        if (!req.is("audio/wav")) {
+            res.status(415).json({ error: "请上传 audio/wav 音频。", code: "INVALID_AUDIO_TYPE" });
+            return;
+        }
+        if (uploads >= 2) {
+            res.status(429).json({ error: "语音服务繁忙，请稍后重试。", code: "BUSY" });
+            return;
+        }
+        uploads += 1;
+        let released = false;
+        const release = () => { if (!released) {
+            released = true;
+            uploads -= 1;
+        } };
+        res.once("finish", release);
+        res.once("close", release);
+        next();
+    };
+    app.post("/api/speech/transcribe", admission, express.raw({ type: "audio/wav", limit: SPEECH_MAX_WAV_BYTES }), async (req, res, next) => {
+        const abort = new AbortController();
+        const disconnected = () => { if (!res.writableEnded)
+            abort.abort(); };
+        res.once("close", disconnected);
+        try {
+            if (!Buffer.isBuffer(req.body))
+                throw new SpeechError("INVALID_AUDIO", "未收到 WAV 音频。");
+            const result = await deps.speech.transcribe(req.body, abort.signal);
+            if (!res.destroyed)
+                res.json(result);
+        }
+        catch (error) {
+            if (!res.destroyed)
+                next(error);
+        }
+        finally {
+            res.off("close", disconnected);
+        }
+    });
+    const errors = (error, _req, res, next) => {
+        if (error instanceof SpeechError) {
+            res.status(error.status).json({ error: error.message, code: error.code });
+            return;
+        }
+        if (error?.type === "entity.too.large") {
+            res.status(413).json({ error: "语音最长 60 秒。", code: "AUDIO_TOO_LARGE" });
+            return;
+        }
+        next(error);
+    };
+    app.use("/api/speech", errors);
+}

package/dist/server-task-routes.js

@@ -193,7 +193,7 @@ export function parseTaskAgent(value, fallbackMode = DEFAULT_WAND_TASK_AGENT_MOD
     if (!isWandTaskAgentEngine(rawEngine))
         throw new Error("执行引擎无效。");
     if (rawEngine === "sdk" && (provider !== "pi" || rawKind !== "structured")) {
-        throw new Error("SDK 执行引擎仅支持 Pi 结构化会话。");
+        throw new Error("Wand Agent 仅支持结构化会话。");
     }
     const engine = rawEngine === "sdk" ? rawEngine : undefined;
     return { provider, model, thinkingEffort, mode, kind: rawKind, ...(engine ? { engine } : {}) };
@@ -232,6 +232,8 @@ function taskSessionSummary(session) {
         thinkingEffort: session.thinkingEffort || "off",
         // 会话实际跑的执行模式；旧会话没有该字段时留空，前端回落到任务上的配置。
         mode: session.mode || "",
+        // 执行引擎：core = Wand Agent（进程内 SDK）；pi 会话没跑过时按 CLI 读。
+        engine: session.provider === "pi" ? (session.structuredState?.engine === "core" ? "sdk" : "cli") : undefined,
         employeeId: session.employeeId,
         employeeName: session.employeeName,
         employeeAvatar: session.employeeAvatar,

package/dist/server-workspace-routes.js

@@ -38,6 +38,7 @@ function workspaceSessionSummary(session, names = {}, registry, team = {}) {
         completionRevision: live.completionRevision ?? 0,
         viewedCompletionRevision: live.viewedCompletionRevision ?? 0,
         ptyBusy: live.ptyBusy === true,
+        ptyCommandRunning: live.ptyCommandRunning === true,
         archived: live.archived === true,
         archivedAt: live.archivedAt ?? null,
         providerCliActive: live.providerCliActive,
@@ -84,6 +85,7 @@ function cheapTasksRevision(storage, registry, includeArchived = false) {
             session.viewedCompletionRevision ?? 0,
             session.archived === true ? 1 : 0,
             session.ptyBusy === true ? 1 : 0,
+            session.ptyCommandRunning === true ? 1 : 0,
             session.providerCliActive ? 1 : 0,
             session.structuredState?.inFlight === true ? 1 : 0,
             session.title ?? "",

package/dist/server.js

@@ -12,7 +12,7 @@ import process from "node:process";
 import { WebSocketServer } from "ws";
 import { AuthService, BROWSER_ADMIN_PRINCIPAL, CONNECTED_APP_PRINCIPAL, principalHasScope, readSessionCookie, SESSION_COOKIE_HTTP, SESSION_COOKIE_HTTPS, SESSION_COOKIE_LEGACY, } from "./auth.js";
 import { ensureCertificates } from "./cert.js";
-import { getDefaultModelForProvider, getProviderDefaultModels, isExecutionMode, resolveConfigDir, } from "./config.js";
+import { getDefaultModelForProvider, getProviderDefaultModels, isExecutionMode, resolveConfigDir, writePreferenceToStorage, } from "./config.js";
 import { ModelCatalogService } from "./models.js";
 import { defaultModelGroupSelector } from "./model-groups.js";
 import { ProcessManager } from "./process-manager.js";
@@ -28,6 +28,10 @@ import { registerDecisionRoutes } from "./server-decision-routes.js";
 import { recordRecentPath, registerFileRoutes } from "./server-file-routes.js";
 import { registerLocalPreviewRoutes } from "./server-local-preview-routes.js";
 import { registerSettingsRoutes } from "./server-settings-routes.js";
+import { registerSpeechRoutes } from "./server-speech-routes.js";
+import { registerLocalModelRoutes } from "./server-local-model-routes.js";
+import { LocalModelSetupService } from "./local-model-setup.js";
+import { SpeechService } from "./speech-service.js";
 import { appTokenLoginPayload, buildStructuredChatPersonaPayload, isBrowserExtensionOrigin, resolveAppConnectCode, resolveStructuredChatAvatarPath, verifyAppToken, } from "./server-app-connect.js";
 import { firstHeaderValue, sendRouteError } from "./server-request.js";
 import { inferProviderFromCommand, isSessionProvider } from "./session-provider.js";
@@ -78,6 +82,9 @@ import { getStylesAsset } from "./web-ui/styles.js";
 import { WsBroadcastManager } from "./ws-broadcast.js";
 import { TerminalDaemonClient } from "./terminal-daemon-client.js";
 import { createUpgradeAwareTerminalHost } from "./render-host.js";
+import { DaemonAdmission, DaemonMaintenance } from "./daemon-maintenance.js";
+import { createDaemonMaintenanceTargets } from "./daemon-maintenance-targets.js";
+import { terminalDaemonBuildIsCurrent } from "./terminal-daemon-build.js";
 import { createUpgradeAwareStructuredHost } from "./render-structured-host.js";
 import { checkPasswordRateLimit, recordFailedPassword, resetPasswordRateLimit } from "./middleware/rate-limit.js";
 import { updateProviderClis, verifyProviderCliUpdateResults, } from "./provider-cli-updater.js";
@@ -88,7 +95,7 @@ const RUNTIME_ROOT_DIR = path.resolve(SERVER_MODULE_DIR, "..");
 const PKG_JSON = JSON.parse(readFileSync(path.join(RUNTIME_ROOT_DIR, "package.json"), "utf8"));
 const PKG_NAME = PKG_JSON.name;
 const PKG_VERSION = PKG_JSON.version;
-const PKG_NODE_REQ = PKG_JSON.engines?.node ?? ">=26.10.0";
+const PKG_NODE_REQ = PKG_JSON.engines?.node ?? ">=24.21.0";
 const PKG_REPO_URL = "https://github.com/co0ontty/wand";
 /** 结构化聊天头像允许的图片类型；未知扩展名回 415。 */
 const AVATAR_CONTENT_TYPES = {
@@ -182,6 +189,7 @@ const CONNECTED_APP_PREFERENCE_KEYS = new Set([
     "defaultThinkingEffort",
     "defaultProvider",
     "defaultSessionKind",
+    "defaultEngine",
     "defaultTaskWorktree",
 ]);
 function requireAdminOrSessionPreferences(req, res, next) {
@@ -351,6 +359,18 @@ export async function startServer(config, configPath, options = {}) {
     // Clients receive its persisted snapshot via GET /api/models.
     const modelCatalog = new ModelCatalogService(getModelRefreshOptions);
     const configDir = resolveConfigDir(configPath);
+    const speech = new SpeechService(storage, configDir);
+    const localModels = new LocalModelSetupService({ configDir, speech, decisions,
+        decisionConfig: () => config.localDecision ?? { enabled: false, pythonPath: "", modelPath: "" },
+        configureDecision(next) {
+            decisions.assertConfigurable();
+            const candidate = runtimeConfig.createCandidate();
+            writePreferenceToStorage(candidate, storage, "localDecision", next);
+            decisions.configure(next);
+            runtimeConfig.commit(candidate, ["localDecision"]);
+            refreshDecisionRuntime();
+        },
+    });
     const distributionManager = new DistributionManager({
         configDir,
         configPath,
@@ -369,7 +389,8 @@ export async function startServer(config, configPath, options = {}) {
             binaryPath: config.render?.binaryPath,
             knownSessionIds: knownPtySessionIds,
         });
-    const terminalHost = ptyHosts.host;
+    const daemonAdmission = new DaemonAdmission();
+    const terminalHost = daemonAdmission.terminal(ptyHosts.host);
     let decisionRuntime = null;
     let autoAssignEvaluate;
     const processes = new ProcessManager(config, storage, configDir, terminalHost, { resolveDecisionEvaluate: () => autoAssignEvaluate });
@@ -377,9 +398,9 @@ export async function startServer(config, configPath, options = {}) {
     // Production startup provides a daemon-backed host for structured CLI runs
     // even when Render owns every PTY. In-process hosts are for test injection.
     const legacyStructuredHost = ptyHosts.legacyHost
-        ?? (terminalHost instanceof TerminalDaemonClient ? terminalHost : null);
+        ?? (ptyHosts.host instanceof TerminalDaemonClient ? ptyHosts.host : null);
     const structuredHosts = await createUpgradeAwareStructuredHost(configPath, legacyStructuredHost, config.structured?.processHost);
-    const structuredSessions = new StructuredSessionManager(storage, config, structuredLogger, {}, structuredHosts.host, () => decisionRuntime, openRouter, () => autoAssignEvaluate);
+    const structuredSessions = new StructuredSessionManager(storage, config, structuredLogger, {}, structuredHosts.host ? daemonAdmission.structured(structuredHosts.host) : undefined, () => decisionRuntime, openRouter, () => autoAssignEvaluate);
     // core harness 预热：只读本机 Pi 认证与模型目录，不联网。失败不影响启动，
     // 但要在开始接输入前得到确定结论，否则首轮引擎裁决会因缓存未就绪而退回 CLI。
     const coreHarnessStatus = await warmCoreHarness(config.harness);
@@ -436,6 +457,16 @@ export async function startServer(config, configPath, options = {}) {
         structuredSessions.registerRelay(prefix, (sessionId, input) => forwardConversationRelay(conversationService, sessionId, input));
     }
     const updateState = new ServerUpdateState();
+    const daemonMaintenance = new DaemonMaintenance({
+        targets: testMode ? [] : await createDaemonMaintenanceTargets(configPath, ptyHosts, structuredHosts.rustClient, config.render?.binaryPath),
+        admission: daemonAdmission,
+        busy: () => structuredSessions.getCoreTurnStatus().hasActiveTurns
+            || [...processes.list(), ...structuredSessions.list()].some((session) => session.status === "running" || session.structuredState?.inFlight === true || (session.queuedMessages?.length ?? 0) > 0),
+        beginCoreDrain: () => structuredSessions.beginCoreRestartDrain(),
+        available: () => !shuttingDown && !updateState.updateInFlight && !updateState.providerCliUpdateInFlight
+            && terminalDaemonBuildIsCurrent(),
+        log: (error) => wandError("底层组件自动更新暂缓，将稍后重试", getErrorMessage(error)),
+    });
     const getUpdateChannel = () => normalizeUpdateChannel(storage.getConfigValue("updateChannel"));
     let disconnectAuthenticatedSockets = () => { };
     const refreshProviderCliUpdates = async () => {
@@ -660,6 +691,9 @@ export async function startServer(config, configPath, options = {}) {
     });
     registerDecisionRoutes(app, { storage, decisions, requireAuth, requireSessions });
     app.use("/api", requireAuth);
+    app.get("/api/daemon-maintenance", (_req, res) => {
+        res.set("Cache-Control", "no-store").json(daemonMaintenance.status());
+    });
     // Connected apps receive only the route families used by native clients and
     // the browser extension. Browser-admin sessions implicitly satisfy all scopes.
     app.use([
@@ -728,6 +762,7 @@ export async function startServer(config, configPath, options = {}) {
             port: config.port,
             defaultProvider: config.defaultProvider ?? "claude",
             defaultSessionKind: config.defaultSessionKind ?? "structured",
+            defaultEngine: config.defaultEngine ?? "cli",
             defaultTaskWorktree: config.defaultTaskWorktree !== false,
             defaultMode: config.defaultMode,
             defaultCwd: config.defaultCwd,
@@ -833,6 +868,8 @@ export async function startServer(config, configPath, options = {}) {
     registerWorkspaceRoutes(app, storage, sessionRegistry, { config });
     registerMissionRoutes(app, missions);
     registerUploadRoutes(app, sessionRegistry);
+    registerSpeechRoutes(app, { speech, requireSessions, requireAdmin });
+    registerLocalModelRoutes(app, { models: localModels, requireSessions, requireAdmin });
     app.post("/api/optimize-prompt", asyncRoute(async (req, res) => {
         const body = (req.body ?? {});
         const text = typeof body.text === "string" ? body.text : "";
@@ -1099,6 +1136,17 @@ export async function startServer(config, configPath, options = {}) {
         }, 600);
         restartTimer.unref?.();
     }));
+    function refreshDecisionRuntime() {
+        const address = server.address();
+        const port = typeof address === "object" && address ? address.port : config.port;
+        const status = decisions.status();
+        const localHost = config.host === "::1" ? "[::1]" : config.host === "0.0.0.0" || config.host === "::" ? "127.0.0.1" : config.host;
+        decisionRuntime = config.localDecision?.enabled ? { url: `${protocol}://${localHost}:${port}`,
+            ...(activeSslCertPath ? { caPath: activeSslCertPath } : {}),
+            evaluate: (value, caller, signal) => decisions.evaluate(value, caller, signal) } : null;
+        autoAssignEvaluate = config.localDecision?.enabled && status.supported && status.configured
+            ? (value, caller, signal) => decisions.evaluate(value, caller, signal) : undefined;
+    }
     let bindAddr = config.host === "0.0.0.0" ? "0.0.0.0" : config.host;
     const collectedUrls = [];
     await new Promise((resolve, reject) => {
@@ -1111,6 +1159,8 @@ export async function startServer(config, configPath, options = {}) {
             }
             catch { /* noop */ }
             conversationService.dispose();
+            localModels.dispose();
+            speech.dispose();
             try {
                 structuredSessions.dispose();
             }
@@ -1157,19 +1207,7 @@ export async function startServer(config, configPath, options = {}) {
             const address = server.address();
             const actualPort = typeof address === "object" && address ? address.port : config.port;
             bindAddr = `${config.host}:${actualPort}`;
-            const decisionStatus = decisions.status();
-            if (config.localDecision?.enabled) {
-                const localHost = config.host === "::1" ? "[::1]" : config.host === "0.0.0.0" || config.host === "::" ? "127.0.0.1" : config.host;
-                decisionRuntime = { url: `${protocol}://${localHost}:${actualPort}`,
-                    ...(activeSslCertPath ? { caPath: activeSslCertPath } : {}),
-                    // core 会话直接调进程内决策服务，不再绕出 HTTP + env token。
-                    evaluate: (value, caller, signal) => decisions.evaluate(value, caller, signal) };
-            }
-            // 「智能分配」只要真正能跑（启用 + 平台/运行环境就绪）才由本地决策负责；
-            // 否则转给系统员工的一次性文本回退，而不是每次白等一个必然失败的调用。
-            if (config.localDecision?.enabled && decisionStatus.supported && decisionStatus.configured) {
-                autoAssignEvaluate = (value, caller, signal) => decisions.evaluate(value, caller, signal);
-            }
+            refreshDecisionRuntime();
             const scheme = useHttps ? "HTTPS" : "HTTP";
             // 主 URL：本机回环；若绑定 0.0.0.0 再补一个对外提示。
             collectedUrls.push({ url: `${protocol}://127.0.0.1:${actualPort}`, scheme });
@@ -1244,6 +1282,10 @@ export async function startServer(config, configPath, options = {}) {
     // async route above is wrapped with asyncRoute, and this final middleware
     // keeps parser, synchronous middleware, and async failures JSON-shaped.
     app.use(jsonErrorHandler);
+    // Daemon-only maintenance does not restart Server, depend on npm auto-update
+    // preferences, or stop a live execution. Startup recovery is already attached.
+    if (!testMode)
+        daemonMaintenance.start();
     // ── Auto-update logic ──
     async function performAutoUpdate() {
         if (shuttingDown)
@@ -1384,6 +1426,7 @@ export async function startServer(config, configPath, options = {}) {
             return closePromise;
         closePromise = (async () => {
             shuttingDown = true;
+            await daemonMaintenance.stop();
             decisions.dispose();
             openRouter.dispose();
             await userMemory.dispose();
@@ -1436,6 +1479,8 @@ export async function startServer(config, configPath, options = {}) {
             }
             catch { /* best-effort shutdown */ }
             conversationService.dispose();
+            localModels.dispose();
+            speech.dispose();
             try {
                 structuredSessions.dispose();
             }

package/dist/session-completion-state.d.ts

@@ -16,6 +16,7 @@ export declare function isSessionJustCompleted(session: SessionCompletionState &
     status?: string;
     archived?: boolean;
     ptyBusy?: boolean;
+    ptyCommandRunning?: boolean;
     inFlight?: boolean;
     structuredState?: {
         inFlight?: boolean;

package/dist/session-completion-state.js

@@ -21,6 +21,7 @@ export function isSessionJustCompleted(session) {
     return (session.completionRevision ?? 0) > (session.viewedCompletionRevision ?? 0)
         && !session.archived
         && !session.ptyBusy
+        && !session.ptyCommandRunning
         && !session.inFlight
         && !session.structuredState?.inFlight
         && !session.permissionBlocked

package/dist/session-transport.js

@@ -36,6 +36,7 @@ function sessionBase(snapshot) {
         providerCliActive: snapshot.providerCliActive,
         providerCliExitCode: snapshot.providerCliExitCode,
         ptyBusy: snapshot.ptyBusy === true,
+        ptyCommandRunning: snapshot.ptyCommandRunning === true,
         // 与 ptyBusy 同生命周期的本轮锚点（运行时事实，存储里没有对应列）。和 ptyBusy 一起
         // 读成 null，客户端才不会在一条已经 idle 的会话上算出「已运行 N 分钟」。
         ptyTurnStartedAt: snapshot.ptyBusy === true ? snapshot.ptyTurnStartedAt ?? null : null,

package/dist/silicon-employee-dispatch.js

@@ -1,13 +1,16 @@
 import { providerCliInstalled } from "./session-provider.js";
+import { isLocalDecisionModel } from "./decision-expert-identity.js";
 /** Match child-process PATH without starting a shell or trusting a client availability claim. */
 export function employeeCliAvailable(agent, pathValue = process.env.PATH ?? "") {
     return providerCliInstalled(agent.provider, pathValue);
 }
 /** First installed candidate wins; when none is installed retain the preferred candidate for a recoverable failure. */
 export function selectEmployeeCandidate(employee, available = employeeCliAvailable, options = {}) {
-    const candidates = options.skipSdk ? employee.agents.filter((agent) => agent.engine !== "sdk") : employee.agents;
+    // LAYA is invoked only by the bounded decision evaluator; never pass its selector to a chat CLI/harness.
+    const generative = employee.agents.filter((agent) => !isLocalDecisionModel(agent.model));
+    const candidates = options.skipSdk ? generative.filter((agent) => agent.engine !== "sdk") : generative;
     if (!candidates.length)
-        throw new Error("硅基员工没有可用的 CLI 执行候选。");
+        throw new Error("LAYA 仅支持有界决策，或员工没有可用 CLI 候选；请配置「决策专家」备用调用链。");
     if (!candidates.length || candidates.some((agent) => agent.kind !== "structured")) {
         throw new Error("硅基员工没有可用的结构化执行候选。");
     }

package/dist/silicon-employee-draft.js

@@ -115,7 +115,7 @@ function parseEmployeeCandidate(raw, allowed) {
         throw new SiliconEmployeeDraftError("执行引擎无效。", "DRAFT_INVALID_ENGINE", "engine");
     }
     if (engine === "sdk" && provider !== "pi") {
-        throw new SiliconEmployeeDraftError("SDK 执行引擎仅支持 Pi 结构化会话。", "DRAFT_INVALID_ENGINE", "engine");
+        throw new SiliconEmployeeDraftError("Wand Agent 仅支持结构化会话。", "DRAFT_INVALID_ENGINE", "engine");
     }
     return {
         provider,

package/dist/speech-models.d.ts

@@ -0,0 +1,24 @@
+/** Multilingual models (not .en); pinned upstream LFS SHA-256 and byte size. */
+export declare const SPEECH_MODEL_REVISION = "5359861c739e955e79d9a303bcbc70fb988958b1";
+export declare const SPEECH_MODELS: readonly [{
+    readonly id: "tiny";
+    readonly label: "Whisper Tiny";
+    readonly description: "中英及多语言 · 低内存 CPU，速度优先";
+    readonly size: 77691713;
+    readonly sha256: "be07e048e1e599ad46341c8d2a135645097a538221678b7acdd1b1919c6e1b21";
+}, {
+    readonly id: "base";
+    readonly label: "Whisper Base";
+    readonly description: "中英及多语言 · 通用 CPU / Mac mini，均衡推荐";
+    readonly size: 147951465;
+    readonly sha256: "60ed5bc3dd14eea856493d334349b405782ddcaf0028d4b5df4088345fba2efe";
+}, {
+    readonly id: "small";
+    readonly label: "Whisper Small";
+    readonly description: "中英及多语言 · 更高准确度，适合 Metal / CUDA 或较强 CPU";
+    readonly size: 487601967;
+    readonly sha256: "1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b";
+}];
+export type SpeechModel = typeof SPEECH_MODELS[number];
+export declare function speechModel(id: string): SpeechModel | undefined;
+export declare function speechModelUrl(model: SpeechModel): string;

package/dist/speech-models.js

@@ -0,0 +1,16 @@
+/** Multilingual models (not .en); pinned upstream LFS SHA-256 and byte size. */
+export const SPEECH_MODEL_REVISION = "5359861c739e955e79d9a303bcbc70fb988958b1";
+export const SPEECH_MODELS = [
+    { id: "tiny", label: "Whisper Tiny", description: "中英及多语言 · 低内存 CPU，速度优先", size: 77_691_713,
+        sha256: "be07e048e1e599ad46341c8d2a135645097a538221678b7acdd1b1919c6e1b21" },
+    { id: "base", label: "Whisper Base", description: "中英及多语言 · 通用 CPU / Mac mini，均衡推荐", size: 147_951_465,
+        sha256: "60ed5bc3dd14eea856493d334349b405782ddcaf0028d4b5df4088345fba2efe" },
+    { id: "small", label: "Whisper Small", description: "中英及多语言 · 更高准确度，适合 Metal / CUDA 或较强 CPU", size: 487_601_967,
+        sha256: "1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b" },
+];
+export function speechModel(id) {
+    return SPEECH_MODELS.find((model) => model.id === id);
+}
+export function speechModelUrl(model) {
+    return `https://huggingface.co/ggerganov/whisper.cpp/resolve/${SPEECH_MODEL_REVISION}/ggml-${model.id}.bin`;
+}

package/dist/speech-service.d.ts

@@ -0,0 +1,57 @@
+import { type SpeechModel } from "./speech-models.js";
+import { type SpeechBackend, type SpeechResult, type SpeechSettings, type SpeechStatus } from "./speech-types.js";
+import type { WandStorage } from "./storage.js";
+export declare const SPEECH_SETTINGS_KEY = "pref:speechRecognition";
+export declare class SpeechError extends Error {
+    readonly code: string;
+    readonly status: number;
+    constructor(code: string, message: string, status?: number);
+}
+export declare function defaultSpeechSettings(): SpeechSettings;
+export declare function parseSpeechSettings(value: unknown, previous?: SpeechSettings): SpeechSettings;
+/** Strict canonical WAV; all clients encode this exact shape. No decoder/ffmpeg needed on the host. */
+export declare function validateSpeechWav(audio: Buffer): {
+    samples: number;
+    silent: boolean;
+};
+export interface SpeechServiceOptions {
+    /** Trusted deployment configuration only, never HTTP request input. */
+    executable?: string;
+    backend?: SpeechBackend;
+    fetch?: typeof fetch;
+    timeoutMs?: number;
+    models?: readonly SpeechModel[];
+}
+/** A single bounded job per host avoids CPU/RAM amplification. Audio is temporary, never history. */
+export declare class SpeechService {
+    private readonly storage;
+    private readonly options;
+    private readonly root;
+    private readonly models;
+    private readonly verified;
+    private readonly verifying;
+    private download;
+    private downloadAbort;
+    private active;
+    private disposed;
+    private maintenance;
+    private readonly initialized;
+    constructor(storage: Pick<WandStorage, "getPreference" | "setPreference">, configDir: string, options?: SpeechServiceOptions);
+    settings(): SpeechSettings;
+    configure(patch: unknown): SpeechSettings;
+    acquireMaintenance(): void;
+    releaseMaintenance(): void;
+    /** Explicit admin health check loads the selected model using synthetic silence, without enabling it. */
+    initializeModel(id: string, signal: AbortSignal): Promise<void>;
+    private modelPath;
+    private modelReady;
+    private verifyModel;
+    private runtime;
+    status(): Promise<SpeechStatus>;
+    cancelDownload(): void;
+    startDownload(id: string): Promise<void>;
+    private downloadModel;
+    transcribe(audio: Buffer, signal?: AbortSignal): Promise<SpeechResult>;
+    private run;
+    dispose(): void;
+}