@deksden-com/dd-flow-cli 0.9.0-beta.122 → 0.9.0-beta.124

package/dist/harness-runtime/lib/dd-droid-daemon.mjs

@@ -4,7 +4,7 @@ import { spawn } from "node:child_process";
 import net from "node:net";
 import os from "node:os";
 import { DroidError, DroidRuntime, absolute, droidPaths, droidVersion, prepareDroidHome, readJson, removeDroidAuth, processSnapshot } from "./dd-droid.mjs";
-import { durableDaemonDispatch, inspectDaemonOperation } from "./daemon-operations.mjs";
+import { durableDaemonDispatch, inspectDaemonOperation, completeDaemonShutdown, commitDaemonShutdown, assertDaemonDispatch, assertDaemonRestartable, trackDaemonConnections, retireDaemonEndpoint, retireFailedDaemon } from "./daemon-operations.mjs";
 import { observedTimeout } from "./observation-clock.mjs";
 import { writeJsonAtomic } from "./runner-events.mjs";
 import { errorRecord } from "./operation-errors.mjs";
@@ -33,8 +33,9 @@ export async function startDaemon(options) {
   config.cwd = await realpath(config.cwd); config.projectRoot = await realpath(config.projectRoot);
   if (config.mode !== "auto-high") throw new DroidError("droid_mode_unsupported", "The qualified Droid profile requires auto-high");
   await mkdir(paths.dir, { recursive: true, mode: 0o700 }); await chmod(paths.dir, 0o700);
+  let previous = await readJson(paths.state); assertDaemonRestartable(previous);
   try { const live = await callDaemon(paths.dir, "daemon.status", {}, 1_000); if (!sameConfig(live.config, config)) throw new DroidError("daemon_config_mismatch", "Droid daemon already uses another configuration"); return { ...live, already_running: true }; } catch (error) { if (!["daemon_not_running", "daemon_timeout"].includes(error.code)) throw error; }
-  const previous = await readJson(paths.state);
+  previous = await readJson(paths.state); assertDaemonRestartable(previous);
   if (previous) {
     const rows = await processSnapshot();
     if (rows.some(row => !row.zombie && ([previous.pid, previous.provider_pid].includes(row.pid) || previous.owned_processes?.some(old => old.pid === row.pid && old.started === row.started)))) throw new DroidError("invalid_harness_crash", "Previous daemon or provider ownership remains; refusing another owner");
@@ -53,11 +54,11 @@ export async function startDaemon(options) {
     throw new DroidError("daemon_start_failed", "Droid daemon did not become ready");
   } catch (error) { await cleanupFailedStart(config, resource, child, error); throw error; }
 }
-export async function stopDaemon(options) { return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, 30_000), droidPaths(options.stateDir).socket, 30_000); }
+export async function stopDaemon(options) { return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, 30_000), droidPaths(options.stateDir).socket, 30_000, { readState: () => readJson(droidPaths(options.stateDir).state) }); }
 export async function serveDaemon(stateDir) {
   const paths = droidPaths(stateDir), state = await readJson(paths.state); if (!state?.config) throw new DroidError("daemon_state_missing", "Missing Droid daemon state");
   let saving = Promise.resolve(), busy = null, stopping = false;
-  const save = patch => { Object.assign(state, patch, { pid: process.pid, updated_at: new Date().toISOString() }); const snapshot = structuredClone(state); saving = saving.then(() => writeJsonAtomic(paths.state, snapshot)); return saving; };
+  const save = patch => { Object.assign(state, patch, { pid: process.pid, updated_at: new Date().toISOString() }); const snapshot = structuredClone(state); saving = saving.catch(() => {}).then(() => writeJsonAtomic(paths.state, snapshot)); return saving; };
   const runtime = new DroidRuntime(state.config, state, save);
   const dispatch = async request => {
     const params = request.params ?? {}, operation = request.operation;
@@ -71,24 +72,29 @@ export async function serveDaemon(stateDir) {
       return { ...inspected, cancelled: true, settled: cleanup.settled, cleanup };
     }
     if (operation === "daemon.stop") {
-      if (stopping) throw new DroidError("operation_busy", "Droid shutdown is already in progress");
-      if (!params.cancelTree && (busy || (runtime.rootId && !(await runtime.inspect(runtime.rootId)).settled))) throw new DroidError("tree_not_settled", "Droid still owns unfinished work");
       stopping = true;
       try {
+        return await completeDaemonShutdown(paths.dir, state, save, async () => {
+        if (!params.cancelTree && (busy || (runtime.rootId && !(await runtime.inspect(runtime.rootId)).settled))) throw new DroidError("tree_not_settled", "Droid still owns unfinished work");
+        commitDaemonShutdown(paths.dir);
         const cleanup = await runtime.closeTree(params.cancelTree === true);
         if (runtime.rootId) try { await runtime.inspect(runtime.rootId, { ignoreActive: true }); } catch (error) { cleanup.inspection_error = errorRecord(error); }
         // Native evidence failures remain visible, but cannot strand copied
         // credentials or the daemon after the owned process tree is gone.
-        await removeDroidAuth(state.config); const clean = !cleanup.forced && !cleanup.topology_error && !cleanup.inspection_error;
-        await save({ shutdown_state: clean ? "clean" : "unclean", active_tree: false, cleanup }); await finishDaemonProcess(state.config, state.resource_process);
-        return { stopped: true, clean, settled: true, cleanup, _shutdown: true };
+        const clean = !cleanup.forced && !cleanup.topology_error && !cleanup.inspection_error;
+        return { clean, cleanup };
+        }, [
+          ["auth", () => removeDroidAuth(state.config)],
+          ["daemon_resource", () => finishDaemonProcess(state.config, state.resource_process)]
+        ]);
       }
-      catch (error) { stopping = false; await save({ shutdown_state: "cleanup_failed", cleanup_error: errorRecord(error) }); throw error; }
+      catch (error) { if (error.code === "tree_not_settled" && !runtime.closed) stopping = false; throw error; }
     }
     if (busy || stopping) throw new DroidError("operation_busy", "Another Droid operation is active");
     busy = operation;
     try {
       await save({ active_operation: busy });
+      assertDaemonDispatch(paths.dir);
       if (operation === "session.create") { if (params.prompt) throw new DroidError("invalid_create", "create does not execute a prompt"); return await runtime.create(request.id); }
       if (operation === "session.resume") return await runtime.resume(params.sessionId);
       if (["session.prompt", "session.start"].includes(operation)) return await runtime.prompt(params.sessionId, params.prompt, request.id);
@@ -114,10 +120,11 @@ export async function serveDaemon(stateDir) {
       // durable stop receipt while the Unix pathname is atomically retired.
       // Unlinking it from the request handler races that final response and
       // can leave a stop caller observing the socket for its whole timeout.
-      if (_shutdown) server.close();
+      if (_shutdown) retireDaemonEndpoint(server);
       socket.end(`${JSON.stringify({ schema_id: RESPONSE, id: request.id, ok: true, result: receipt })}\n`);
-      } catch (error) { socket.end(`${JSON.stringify({ schema_id: RESPONSE, id: request?.id, ok: false, error: errorRecord(error) })}\n`); } })();
+      } catch (error) { await retireFailedDaemon(server, state, error, save); socket.end(`${JSON.stringify({ schema_id: RESPONSE, id: request?.id, ok: false, error: errorRecord(error) })}\n`); } })();
   }); });
+  trackDaemonConnections(server);
   try { await runtime.start(); scheduleDaemonProcessHeartbeat(state.config, state.resource_process); await save({ shutdown_state: "running" }); await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); }); await chmod(paths.socket, 0o600); }
   catch (error) { let cleanup; try { cleanup = await runtime.closeTree(true); await removeDroidAuth(state.config); } catch (failure) { await save({ shutdown_state: "cleanup_failed", startup_error: errorRecord(error), cleanup_error: errorRecord(failure) }); throw error; } await save({ shutdown_state: "failed", startup_error: errorRecord(error), cleanup }); await finishDaemonProcess(state.config, state.resource_process, "failed", "startup_failed"); throw error; }
   await new Promise(resolve => server.once("close", resolve));

package/dist/harness-runtime/lib/dd-grok-daemon.mjs

@@ -5,7 +5,7 @@ import { createHash, randomUUID } from "node:crypto";
 import { chmod, copyFile, cp, lstat, mkdir, open, readFile, readdir, rename, unlink, writeFile } from "node:fs/promises";
 import { spawn } from "node:child_process";
 import net from "node:net";
-import { durableDaemonDispatch as dispatchOperation } from "./daemon-operations.mjs";
+import { durableDaemonDispatch as dispatchOperation, completeDaemonShutdown, trackDaemonConnections, retireDaemonEndpoint, retireFailedDaemon, assertDaemonDispatch, assertDaemonRestartable } from "./daemon-operations.mjs";
 import { waitForSettlement } from "./session-settlement.mjs";
 import os from "node:os";
 import path from "node:path";
@@ -13,7 +13,7 @@ import path from "node:path";
 import { AcpBridge } from "./dd-zcode.mjs";
 import { observeModel, readModelObservations } from "./model-observations.mjs";
 import { cancelSessionWithBridge, createSessionWithBridge, doctor, forkSessionWithBridge, inspectSessionWithBridge, promptSessionWithBridge } from "./dd-grok.mjs";
-import { prepareRuntimeOwner, cleanupFailedDaemonStart, closeManagedBridge, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, scheduleDaemonProcessHeartbeat, registerDaemonProcess, startManagedBridge, stopProcessGroup } from "./managed-daemon.mjs";
+import { prepareRuntimeOwner, cleanupFailedDaemonStart, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, scheduleDaemonProcessHeartbeat, registerDaemonProcess, startManagedBridge, stopProcessGroup } from "./managed-daemon.mjs";
 
 const REQUEST_SCHEMA = "dd-grok/daemon-request@1";
 const RESPONSE_SCHEMA = "dd-grok/daemon-response@1";
@@ -169,8 +169,10 @@ export async function startDaemon(options) {
   await prepareRuntimeOwner(config, paths.dir);
   if (!config.model || !config.reasoning) throw new DaemonError("profile_required", "--model and --reasoning are required");
   await mkdir(paths.dir, { recursive: true, mode: 0o700 }); await chmod(paths.dir, 0o700);
+  let previous = await readState(paths.dir); assertDaemonRestartable(previous);
   try { const status = await callDaemon(paths.dir, "daemon.ready", {}, 5000); if (!sameConfig(status, config)) throw new DaemonError("daemon_config_mismatch", "a daemon is already running with different configuration"); if (!status.ready) throw new DaemonError("daemon_start_in_progress", "an existing daemon is still starting", true); return { ...status, already_running: true }; } catch (error) { if (error.code !== "daemon_not_running") throw error; }
-  const previous = await readState(paths.dir); if (previous?.shutdown_state === "clean") await authorizeRetainedDaemonResume(paths.dir, previous, options.sessionId, config); if (previous?.shutdown_state === "running" && previous.active_tree) throw new DaemonError("invalid_harness_crash", "previous daemon died with an active or unproven Session tree", false, { daemon_id: previous.daemon_id });
+  previous = await readState(paths.dir); assertDaemonRestartable(previous);
+  if (previous?.shutdown_state === "clean") await authorizeRetainedDaemonResume(paths.dir, previous, options.sessionId, config); if (previous?.shutdown_state === "running" && previous.active_tree) throw new DaemonError("invalid_harness_crash", "previous daemon died with an active or unproven Session tree", false, { daemon_id: previous.daemon_id });
   await removeSocket(paths.socket, paths.dir); await mkdir(paths.home, { recursive: true, mode: 0o700 }); const importedSession = await materializeSessionArchive(paths, config.sessionArchive); const copiedAuth = await copyAuth(paths, authSource); await writeIsolatedConfig(paths); if (!noFlow) await writeHook(paths, config, absolute(options.entryPath, "dd-grok entry")); const versions = await doctor({ bin: config.bin }); if (importedSession && importedSession.grok_version !== versions.versions.grok) throw new DaemonError("session_archive_version_mismatch", "Grok Build session archive was created by a different CLI version", false, { archive_version: importedSession.grok_version, current_version: versions.versions.grok }); const configIsolation = await inspectIsolatedConfig(config);
   const resourceProcess = await registerDaemonProcess(config, { kind: "grok-daemon", owner: `grok:${path.basename(paths.dir)}`, operation: `grok-daemon:${path.basename(paths.dir)}`, stdout: paths.log, stderr: paths.log });
   const daemonId = randomUUID();
@@ -179,11 +181,11 @@ export async function startDaemon(options) {
   try { await confirmDaemonProcess(config, resourceProcess, child); const deadline = Date.now() + 120_000; while (Date.now() < deadline) { await new Promise((resolve) => setTimeout(resolve, 100)); const failure = grokStartupFailure(await readState(paths.dir)); if (failure) throw failure; try { const status = await callDaemon(paths.dir, "daemon.ready", {}, 5000); if (status.ready) return status; } catch (error) { if (error.code !== "daemon_not_running" && error.code !== "daemon_timeout") throw error; } } throw new DaemonError("daemon_start_failed", "daemon did not become ready", true); }
   catch (error) { await cleanupFailedDaemonStart(config, resourceProcess, child, error, () => readState(paths.dir), daemonId); throw error; }
 }
-export async function stopDaemon(options) { const timeoutMs = options.timeoutMs ?? 30_000; return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), locations(options.stateDir).socket, timeoutMs); }
+export async function stopDaemon(options) { const timeoutMs = options.timeoutMs ?? 30_000; return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), locations(options.stateDir).socket, timeoutMs, { readState: () => readState(options.stateDir) }); }
 
 export class Runtime {
   constructor(paths, state, bridge, initialized) { this.paths = paths; this.state = state; this.bridge = bridge; this.initialized = initialized; this.sessions = new Map((state.sessions ?? []).map((item) => [item.provider_session_id, item])); this.descendants = new Map((state.descendants ?? []).map(child => [child.provider_session_id, child])); for (const session of this.sessions.values()) if (session.parent_provider_session_id && !this.descendants.has(session.provider_session_id)) this.descendants.set(session.provider_session_id, { provider_session_id: session.provider_session_id, parent_provider_session_id: session.parent_provider_session_id, status: "unknown", source: "legacy.daemon_state", outcome_error: { code: "native_child_outcome_unavailable", message: "Retained Grok child has no durable outcome receipt" } }); this.toolUsage = new Map(); this.loadedSessionId = null; this.active = null; this.server = null; this.ready = false; this.persisting = Promise.resolve(); }
-  persist(patch = {}) { Object.assign(this.state, patch, { pid: process.pid, updated_at: new Date().toISOString(), sessions: [...this.sessions.values()], descendants: [...this.descendants.values()] }); const snapshot = structuredClone(this.state); this.persisting = this.persisting.then(() => writeState(this.paths.state, snapshot)); return this.persisting; }
+  persist(patch = {}) { Object.assign(this.state, patch, { pid: process.pid, updated_at: new Date().toISOString(), sessions: [...this.sessions.values()], descendants: [...this.descendants.values()] }); const snapshot = structuredClone(this.state); this.persisting = this.persisting.catch(() => {}).then(() => writeState(this.paths.state, snapshot)); return this.persisting; }
   options(params) { const sessionId = params.sessionId; const session = sessionId ? this.sessions.get(sessionId) : null; const root = session?.root_provider_session_id ?? sessionId; return { ...this.state.config, ...params, cwd: params.cwd ?? session?.cwd ?? this.state.config.cwd, initialized: this.initialized, daemonId: this.state.daemon_id, rootProviderSessionId: root, toolUsage: this.toolUsage, allowBackground: true, liveSession: params.liveSession ?? Boolean(sessionId && sessionId === this.loadedSessionId) }; }
   track(result, before = null) { if (!result?.provider_session_id) return; const session = this.sessions.get(result.provider_session_id); const root = result.parent_provider_session_id ? (this.sessions.get(result.parent_provider_session_id)?.root_provider_session_id ?? result.parent_provider_session_id) : (session?.root_provider_session_id ?? result.provider_session_id); this.sessions.set(result.provider_session_id, { native_root_receipt: session?.native_root_receipt ?? null, provider_session_id: result.provider_session_id, adapter_session_id: result.adapter_session_id ?? result.provider_session_id, parent_provider_session_id: result.parent_provider_session_id ?? session?.parent_provider_session_id ?? null, root_provider_session_id: root, cwd: result.cwd ?? result.target?.newCwd ?? result.info?.cwd ?? session?.cwd ?? null }); for (const child of this.running(result)) this.recordDescendant({ ...child, stale_activity: Boolean(before && before.get(child.sessionId ?? child.subagentId ?? child.provider_session_id) !== this.descendants.get(child.sessionId ?? child.subagentId ?? child.provider_session_id)) }, root, "x.ai/subagent/list_running", "running"); for (const child of result.descendants ?? []) this.recordDescendant(child, root, child.source ?? "x.ai/subagent/event", child.status ?? "unknown"); }
   recordDescendant(value, root, source, fallback) {
@@ -339,6 +341,7 @@ export class Runtime {
   }
   async requireSettled() { const running = await this.refreshTree(); if (running.length) throw new DaemonError("tree_not_settled", "daemon still owns a running Session tree", false, { sessions: running }); }
   async productive(name, task, sessionId) {
+    assertDaemonDispatch(this.paths?.dir);
     if (this.active) throw new DaemonError("operation_busy", `${this.active} is already running`, true);
     this.active = name;
     // Only a new native turn invalidates its own root proof. Creation, fork
@@ -346,6 +349,7 @@ export class Runtime {
     if (name === "session.prompt" && this.sessions.has(sessionId)) this.sessions.get(sessionId).native_root_receipt = null;
     try {
       await this.persist({ active_tree: true, active_operation: name });
+      assertDaemonDispatch(this.paths?.dir);
       const before = new Map(this.descendants);
       const result = await task();
       this.track(result, before);
@@ -371,8 +375,9 @@ export class Runtime {
   }
   async dispatch(operation, params) {
     const assertDispatch = captureDispatchGuard(this);
+    if (["session.create", "session.prompt", "session.fork", "session.archive"].includes(operation)) assertDaemonDispatch(this.paths.dir);
     if (operation === "daemon.ready") return { ready: this.ready, daemon_id: this.state.daemon_id, pid: process.pid, socket: this.paths.socket, versions: this.state.versions, config_isolation: this.state.config_isolation, shutdown_state: this.state.shutdown_state, recovery_status: this.state.recovery_status, auth_status: this.state.auth_status, active_tree: this.state.active_tree, active_operation: this.active, sessions: [...this.sessions.values()], config: this.state.config };
-    if (operation === "daemon.status") { let observationError; if (!this.active) try { await this.refreshTree(); } catch (error) { observationError = errorPayload(error); } return { ...await this.dispatch("daemon.ready", {}), observation_error: observationError ?? this.state.observation_error ?? null }; }
+    if (operation === "daemon.status") { let observationError; if (!this.active && !this.state.shutdown?.phases?.tree) try { await this.refreshTree(); } catch (error) { observationError = errorPayload(error); } return { ...await this.dispatch("daemon.ready", {}), observation_error: observationError ?? this.state.observation_error ?? null }; }
     if (operation === "hook.resolve") { const sessionId = String(params.sessionId ?? ""); if (!sessionId) throw new DaemonError("hook_identity_missing", "hook has no sessionId"); let session = this.sessions.get(sessionId); if (!session) { const roots = [...this.sessions.values()].filter((item) => !item.parent_provider_session_id); if (roots.length !== 1) throw new DaemonError("hook_identity_unknown", "unknown hook Session has no unique root"); const observed = await inspectSessionWithBridge(this.bridge, this.options({ sessionId, liveSession: true })); this.track(observed); session = this.sessions.get(sessionId); if (!session || session.root_provider_session_id !== roots[0].provider_session_id) throw new DaemonError("hook_identity_unknown", "hook Session is not a member of the retained native tree"); await this.persist(); } return { daemonId: this.state.daemon_id, rootProviderSessionId: session.root_provider_session_id, parentProviderSessionId: session.parent_provider_session_id ?? null, observedProfile: this.bridge.modelProfiles?.get(sessionId) ?? {} }; }
     if (operation === "session.create") return await this.productive(operation, async () => { const result = await createSessionWithBridge(this.bridge, { ...this.options(params), onSessionCreated: async (session) => { this.track(session); this.loadedSessionId = session.provider_session_id; await this.persist(); } }, this.initialized); this.loadedSessionId = result.provider_session_id; return result; });
     if (operation === "session.prompt") { await this.requireSettled(); return await this.productive(operation, async () => { const result = await promptSessionWithBridge(this.bridge, { ...this.options(params), assertDispatch }); this.loadedSessionId = params.sessionId; return result; }, params.sessionId); }
@@ -381,18 +386,23 @@ export class Runtime {
     if (operation === "session.cancel") { if (!this.sessions.has(params.sessionId)) throw new DaemonError("session_identity_mismatch", "cancellation Session is not owned by this daemon"); cancelPendingDispatch(this); const result = await cancelSessionWithBridge(this.bridge, this.options({ ...params, liveSession: true })); await this.persist({ active_tree: Boolean(this.active) || !result.settled }); return result; }
     if (operation === "session.archive") { await this.requireSettled(); return await this.productive(operation, async () => { let session = this.sessions.get(params.sessionId); if (!session?.cwd) { const result = await inspectSessionWithBridge(this.bridge, this.options({ ...params, liveSession: true })); this.track(result); session = this.sessions.get(result.provider_session_id); } return await exportSessionArchive(this.paths, session.provider_session_id, absolute(session.cwd, "Grok Build session cwd"), params.archiveDir, this.state.versions); }); }
     if (operation === "daemon.stop") {
+      return await completeDaemonShutdown(this.paths.dir, this.state, patch => this.persist(patch), async () => {
       if (params.cancelTree) cancelPendingDispatch(this);
       await waitForSettlement({
         knownSessions: () => [...this.sessions.values()].filter(s => !s.parent_provider_session_id).map(s => s.provider_session_id),
         observe: async () => ({ sessions: await this.refreshTree(), active: Boolean(this.active), active_operation: this.active }),
         ...(params.cancelTree ? { cancel: id => cancelSessionWithBridge(this.bridge, this.options({ sessionId: id, liveSession: true })) } : {})
       });
-      await this.persist({ active_tree: false, shutdown_state: "stopping" });
-      return { daemon_id: this.state.daemon_id, stopped: true, clean: true, _shutdown: true };
+      return { clean: true };
+      }, [
+        ["provider", () => this.bridge.close()],
+        ["provider_resource", () => finishDaemonProcess(this.state.config, this.state.provider_process)],
+        ["auth", () => unlink(path.join(this.paths.home, "auth.json")).catch(error => { if (error.code !== "ENOENT") throw error; })],
+        ["daemon_resource", () => finishDaemonProcess(this.state.config, this.state.resource_process)]
+      ]);
     }
     throw new DaemonError("unknown_operation", `unknown daemon operation: ${operation}`);
   }
-  async shutdown() { await closeManagedBridge(this.state.config, this.state, this.bridge); await finishDaemonProcess(this.state.config, this.state.resource_process); await unlink(path.join(this.paths.home, "auth.json")).catch(error => { if (error.code !== "ENOENT") throw error; }); await this.persist({ shutdown_state: "clean", stopped_at: new Date().toISOString() }); await new Promise((resolve) => this.server.close(resolve)); await removeSocket(this.paths.socket, this.paths.dir); }
 }
 
 export async function serveDaemon(stateDir) {
@@ -426,6 +436,28 @@ export async function serveDaemon(stateDir) {
     }
   }
   scheduleDaemonProcessHeartbeat(state.config, state.resource_process);
-  await removeSocket(paths.socket, paths.dir); const server = net.createServer((connection) => { connection.setEncoding("utf8"); let buffer = ""; connection.on("data", (chunk) => { buffer += chunk; const newline = buffer.indexOf("\n"); if (newline < 0) return; const line = buffer.slice(0, newline); buffer = buffer.slice(newline + 1); void (async () => { let request; try { request = JSON.parse(line); if (request.schema_id !== REQUEST_SCHEMA || !request.id || !request.operation) throw new DaemonError("daemon_protocol_mismatch", "invalid daemon request"); const result = await durableDaemonDispatch(paths.dir, request, () => runtime.dispatch(request.operation, request.params ?? {}), () => runtime.dispatch("session.inspect", request.params ?? {})); connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request.id, ok: true, result: { ...result, _shutdown: undefined } })}\n`); if (result?._shutdown) setImmediate(() => void runtime.shutdown().then(() => process.exit(0)).catch(error => runtime.persist({ shutdown_state: "cleanup_failed", cleanup_error: errorPayload(error) }))); } catch (error) { connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request?.id ?? null, ok: false, error: errorPayload(error) })}\n`); } })(); }); }); runtime.server = server; await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); }); await chmod(paths.socket, 0o600); await runtime.persist({ shutdown_state: "running", active_tree: false }); runtime.ready = true; return await new Promise(() => {});
+  await removeSocket(paths.socket, paths.dir);
+  const server = net.createServer(connection => {
+    connection.setEncoding("utf8"); let buffer = "", consumed = false;
+    connection.on("data", chunk => {
+      if (consumed) return;
+      buffer += chunk; const at = buffer.indexOf("\n"); if (at < 0) return; consumed = true;
+      void (async () => {
+        let request;
+        try {
+          request = JSON.parse(buffer.slice(0, at));
+          if (request.schema_id !== REQUEST_SCHEMA || !request.id || !request.operation) throw new DaemonError("daemon_protocol_mismatch", "invalid daemon request");
+          const result = await durableDaemonDispatch(paths.dir, request, () => runtime.dispatch(request.operation, request.params ?? {}), () => runtime.dispatch("session.inspect", request.params ?? {}));
+          if (result._shutdown) retireDaemonEndpoint(server);
+          connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request.id, ok: true, result: { ...result, _shutdown: undefined } })}\n`);
+        } catch (error) {
+          await retireFailedDaemon(server, state, error, patch => runtime.persist(patch));
+          connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request?.id ?? null, ok: false, error: errorPayload(error) })}\n`);
+        }
+      })();
+    });
+  });
+  trackDaemonConnections(server); runtime.server = server;
+  await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); }); await chmod(paths.socket, 0o600); await runtime.persist({ shutdown_state: "running", active_tree: false }); runtime.ready = true; return await new Promise(() => {});
 }
 import { authorizeRetainedDaemonResume } from "./driver-recovery.mjs";

package/dist/harness-runtime/lib/dd-opencode-daemon.mjs

@@ -5,8 +5,9 @@ import { createHash, randomBytes, randomUUID } from "node:crypto";
 import { appendFile, chmod, copyFile, lstat, mkdir, open, readFile, realpath, rename, unlink, writeFile } from "node:fs/promises";
 import { spawn } from "node:child_process";
 import net from "node:net";
-import { durableDaemonDispatch } from "./daemon-operations.mjs";
+import { durableDaemonDispatch, completeDaemonShutdown, trackDaemonConnections, retireDaemonEndpoint, retireFailedDaemon, assertDaemonDispatch, assertDaemonRestartable } from "./daemon-operations.mjs";
 import { waitForSettlement } from "./session-settlement.mjs";
+import { errorRecord } from "./operation-errors.mjs";
 import { checkObservedProfile, observeModel } from "./model-observations.mjs";
 import os from "node:os";
 import path from "node:path";
@@ -23,7 +24,25 @@ async function readState(dir) { try { return JSON.parse(await readFile(locations
 async function removeSocket(paths) { try { if (!(await lstat(paths.socket)).isSocket()) throw new DaemonError("unsafe_socket", "refusing to replace non-socket daemon path"); await unlink(paths.socket); } catch (error) { if (error.code !== "ENOENT") throw error; } }
 function executable(bin, args = []) { return /\.[cm]?js$/.test(bin) ? { command: process.execPath, args: [bin, ...args] } : { command: bin, args }; }
 function isolatedEnv(config) { return { ...process.env, XDG_DATA_HOME: config.dataHome, XDG_CONFIG_HOME: config.configHome, XDG_CACHE_HOME: config.cacheHome, OPENCODE_CONFIG_DIR: config.adapterDir, OPENCODE_DISABLE_PROJECT_CONFIG: "1", OPENCODE_SERVER_PASSWORD: config.password, ...(config.ddFlowHome ? { DD_FLOW_HOME: config.ddFlowHome } : {}), ...(config.resourceHome ? { DD_FLOW_RESOURCE_HOME: config.resourceHome } : {}), ...(config.ddFlowBin ? { DD_OPENCODE_DD_FLOW_BIN: config.ddFlowBin, DD_OPENCODE_NODE_BIN: process.execPath, DD_OPENCODE_PROJECT_ROOT: config.projectRoot } : {}), DD_OPENCODE_DAEMON_ID: config.daemonId, DD_OPENCODE_PROVIDER: config.provider, DD_OPENCODE_MODEL: config.model, DD_OPENCODE_VARIANT: config.variant, DD_OPENCODE_AGENT: config.agent, DD_OPENCODE_SERVER_VERSION: config.version ?? "unknown", DD_OPENCODE_PLUGIN_SHA256: config.pluginSha256 ?? "unknown" }; }
-async function run(bin, args, config, { stdin, timeoutMs = 30_000 } = {}) { const target = executable(bin, args); return await new Promise((resolve, reject) => { const child = spawn(target.command, target.args, { cwd: config.cwd, env: isolatedEnv(config), stdio: ["pipe", "pipe", "pipe"] }); let stdout = "", stderr = ""; const timer = setTimeout(() => { child.kill("SIGTERM"); reject(new DaemonError("opencode_timeout", `${args[0]} timed out`, true)); }, timeoutMs); child.stdout.setEncoding("utf8").on("data", chunk => stdout += chunk); child.stderr.setEncoding("utf8").on("data", chunk => stderr += chunk); child.on("error", reject); child.on("close", code => { clearTimeout(timer); code === 0 ? resolve(stdout) : reject(new DaemonError("opencode_cli_failed", stderr.trim() || `${target.command} exited ${code}`)); }); if (stdin) child.stdin.end(stdin); else child.stdin.end(); }); }
+async function run(bin, args, config, { stdin, timeoutMs = 30_000 } = {}) {
+  const target = executable(bin, args);
+  return await new Promise((resolve, reject) => {
+    const child = spawn(target.command, target.args, { cwd: config.cwd, env: isolatedEnv(config), detached: process.platform !== "win32", stdio: ["pipe", "pipe", "pipe"] });
+    let stdout = "", stderr = "", timeoutError, inputError, cleanup;
+    const timer = setTimeout(() => {
+      timeoutError = new DaemonError("opencode_timeout", `${args[0]} timed out`, true);
+      cleanup = stopProcessGroup(child, 1_000).catch(error => { timeoutError.details = { cleanup_error: errorRecord(error) }; reject(timeoutError); });
+    }, timeoutMs);
+    child.stdout.setEncoding("utf8").on("data", chunk => stdout += chunk); child.stderr.setEncoding("utf8").on("data", chunk => stderr += chunk);
+    child.on("error", error => { clearTimeout(timer); reject(error); });
+    child.on("close", async code => { clearTimeout(timer); await cleanup; if (timeoutError) reject(timeoutError); else if (inputError) reject(inputError); else if (code === 0) resolve(stdout); else reject(new DaemonError("opencode_cli_failed", stderr.trim() || `${target.command} exited ${code}`)); });
+    child.stdin.on("error", error => {
+      inputError ??= error;
+      cleanup ??= stopProcessGroup(child, 1_000).catch(cause => { clearTimeout(timer); inputError.details = { ...inputError.details, cleanup_error: errorRecord(cause) }; reject(timeoutError ?? inputError); });
+    });
+    child.stdin.end(stdin ?? "");
+  });
+}
 async function forwardUsage(config, sessionId, snapshot) {
   if (!config.ddFlowBin) return;
   const payload = { provider_session_id: sessionId, daemon_id: config.daemonId, observed_at: new Date().toISOString(), usage_scope: "physical_session", completeness: "complete", usage: { inputTokens: snapshot.input_tokens, outputTokens: snapshot.output_tokens, reasoningTokens: snapshot.reasoning_tokens, cacheReadInputTokens: snapshot.cached_input_tokens, cacheWriteInputTokens: snapshot.cache_write_tokens, totalTokens: snapshot.total_tokens, costUsd: snapshot.cost_usd }, tool_calls: snapshot.tool_calls };
@@ -53,8 +72,10 @@ export async function startDaemon(options) {
   const paths = locations(options.stateDir); await mkdir(paths.dir, { recursive: true, mode: 0o700 }); await chmod(paths.dir, 0o700);
   const previous = await readState(paths.dir), daemonId = randomUUID(), noFlow = options.noFlow === true, cwd = await realpath(absolute(options.cwd, "--cwd")); const config = { cwd, journal: options.journal ? absolute(options.journal, "--journal") : paths.journal, bin: options.bin ?? process.env.DD_OPENCODE_BIN ?? "opencode", provider: String(options.provider ?? ""), model: String(options.model ?? ""), variant: String(options.variant ?? "default"), agent: String(options.agent ?? "build"), projectRoot: noFlow ? cwd : await realpath(absolute(options.projectRoot, "--project-root")), ddFlowBin: noFlow ? null : absolute(options.ddFlowBin, "--dd-flow-bin"), ddFlowHome: noFlow ? null : absolute(options.ddFlowHome, "--dd-flow-home"), resourceHome: noFlow ? null : (process.env.DD_FLOW_RESOURCE_HOME ? absolute(process.env.DD_FLOW_RESOURCE_HOME, "DD_FLOW_RESOURCE_HOME") : null), env: noFlow ? {} : { DD_FLOW_HOME: absolute(options.ddFlowHome, "--dd-flow-home") }, sessionArchive: options.sessionArchive ? absolute(options.sessionArchive, "--session-archive") : null, dataHome: paths.data, configHome: paths.config, cacheHome: paths.cache, adapterDir: path.join(paths.config, "opencode"), password: randomBytes(24).toString("hex"), daemonId, noFlow };
   await prepareRuntimeOwner(config, paths.dir);
+  assertDaemonRestartable(previous);
   if (!config.provider || !config.model) throw new DaemonError("profile_required", "--provider and --model are required");
   try { const status = await callDaemon(paths.dir, "daemon.status", {}, 1000); if (!sameConfig(status, config)) throw new DaemonError("daemon_config_mismatch", "a daemon is already running with different configuration"); return { ...status, already_running: true }; } catch (error) { if (!["daemon_not_running","daemon_timeout"].includes(error.code)) throw error; }
+  assertDaemonRestartable(await readState(paths.dir));
   if (previous?.shutdown_state === "clean") await authorizeRetainedDaemonResume(paths.dir, previous, options.sessionId, config); if (previous?.shutdown_state === "running" && previous.active_tree) throw new DaemonError("invalid_harness_crash", "previous daemon died with an active or unproven Session tree");
   await removeSocket(paths); const authSource = options.authPath ? absolute(options.authPath, "--auth-path") : path.join(os.homedir(), ".local", "share", "opencode", "auth.json"); await prepare(paths, config, authSource); config.version = (await run(config.bin, ["--version"], config)).trim(); if (config.version !== OPENCODE_BASELINE.version) throw new DaemonError("unsupported_version", `unsupported OpenCode version ${config.version}; expected ${OPENCODE_BASELINE.version}`); const imported = await importArchive(paths, config, config.sessionArchive);
   const resourceProcess = await registerDaemonProcess(config, { kind: "opencode-daemon", owner: `opencode:${path.basename(paths.dir)}`, operation: `opencode-daemon:${path.basename(paths.dir)}`, stdout: paths.log, stderr: paths.log });
@@ -63,11 +84,11 @@ export async function startDaemon(options) {
   try { await confirmDaemonProcess(config, resourceProcess, child); const deadline = Date.now() + 20_000; while (Date.now() < deadline) { await new Promise(resolve => setTimeout(resolve, 100)); try { return await callDaemon(paths.dir, "daemon.status", {}, 1000); } catch (error) { if (!["daemon_not_running","daemon_timeout"].includes(error.code)) throw error; } } throw new DaemonError("daemon_start_failed", "daemon did not become ready", true); }
   catch (error) { await cleanupFailedStart(config, resourceProcess, child, error); throw error; }
 }
-export const stopDaemon = async options => { const timeoutMs = options.timeoutMs ?? 30_000; return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), locations(options.stateDir).socket, timeoutMs); };
+export const stopDaemon = async options => { const timeoutMs = options.timeoutMs ?? 30_000; return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), locations(options.stateDir).socket, timeoutMs, { readState: () => readState(options.stateDir) }); };
 
 export class Runtime {
   constructor(paths, state, client, child, providerProcess) { this.paths = paths; this.state = state; this.client = client; this.child = child; this.providerProcess = providerProcess; this.sessions = new Map((state.sessions ?? []).map(x => [x.provider_session_id, x])); this.active = null; this.persisting = Promise.resolve(); }
-  persist(patch = {}) { Object.assign(this.state, patch, { pid: process.pid, sessions: [...this.sessions.values()] }); const snapshot = structuredClone(this.state); this.persisting = this.persisting.then(() => writeJson(this.paths.state, snapshot)); return this.persisting; }
+  persist(patch = {}) { Object.assign(this.state, patch, { pid: process.pid, sessions: [...this.sessions.values()] }); const snapshot = structuredClone(this.state); this.persisting = this.persisting.catch(() => {}).then(() => writeJson(this.paths.state, snapshot)); return this.persisting; }
   async journal(operation, result) { const line = `${JSON.stringify({ schema_id: "dd-opencode/daemon-event@1", event_id: randomUUID(), observed_at: new Date().toISOString(), operation, active_tree: this.state.active_tree, session_id: result?.provider_session_id ?? null })}\n`; for (const journal of new Set([this.paths.journal, this.state.config.journal].filter(Boolean))) { await mkdir(path.dirname(journal), { recursive: true, mode: 0o700 }); await appendFile(journal, line, { mode: 0o600 }); } }
   async childTree(id, seen = new Set()) { if (seen.has(id)) return []; seen.add(id); const direct = await this.client.children(id) ?? []; const nested = []; for (const child of direct) nested.push(child, ...await this.childTree(child.id, seen)); return nested; }
   async describe(id) { const [info, messages, children, statuses] = await Promise.all([this.client.inspect(id), this.client.messages(id), this.childTree(id), this.client.statuses()]); if (path.resolve(info.directory) !== path.resolve(this.state.config.cwd)) throw new DaemonError("foreign_session", "OpenCode Session directory is outside the controlled workspace", false, { session_directory: info.directory, workspace: this.state.config.cwd }); const status = statuses?.[id]?.type ?? "idle", parent = info.parentID ?? null, existing = this.sessions.get(id); const record = { provider_session_id: id, adapter_session_id: id, parent_provider_session_id: parent, root_provider_session_id: parent ? (this.sessions.get(parent)?.root_provider_session_id ?? parent) : (existing?.root_provider_session_id ?? id), cwd: info.directory }; this.sessions.set(id, record); for (const child of children) if (!this.sessions.has(child.id)) this.sessions.set(child.id, { provider_session_id: child.id, parent_provider_session_id: child.parentID ?? id, root_provider_session_id: record.root_provider_session_id, cwd: child.directory }); const unsettledChildren = children.filter(child => (statuses?.[child.id]?.type ?? "idle") !== "idle").map(child => child.id); const lastAssistant = messages.findLast(message => message.info?.role === "assistant"); const assistantText = (lastAssistant?.parts ?? []).filter(part => part.type === "text").map(part => part.text ?? "").join(""); await this.observeModels(id, messages, parent); this.journaledTools ??= new Map(); journalToolObservations(this.state.config.journal, messages.flatMap(message => (message.parts ?? []).map(part => openCodeToolObservation(part, id, message.info?.id))), this.journaledTools); return { ...record, observed_profile: { provider: lastAssistant?.info?.providerID ?? null, model: lastAssistant?.info?.modelID ?? null, reasoning: lastAssistant?.info?.variant ?? null, mode: info.agent ?? null }, assistant_text: assistantText, info, status, settled: status === "idle" && unsettledChildren.length === 0, unsettled_children: unsettledChildren, tree_observations: children.map(child => ({ provider_session_id: child.id, status: statuses?.[child.id]?.type ?? "idle", settled: (statuses?.[child.id]?.type ?? "idle") === "idle" })), children, usage: usageSnapshot(messages, id), history_sha256: historyDigest(messages), message_count: messages.length }; }
@@ -86,10 +107,12 @@ export class Runtime {
   async requireSettled(id) { const value = await this.describe(id); if (!value.settled) throw new DaemonError("tree_not_settled", "OpenCode Session is not idle", false, { session_id: id, status: value.status }); return value; }
   async reconcileUsage(observed) { const errors = []; try { await forwardUsage(this.state.config, observed.provider_session_id, observed.usage); } catch (error) { errors.push({ session_id: observed.provider_session_id, phase: "usage", ...errorPayload(error) }); } for (const child of observed.children.filter(item => (item.parentID ?? observed.provider_session_id) === observed.provider_session_id)) { try { errors.push(...await this.reconcileUsage(await this.describe(child.id))); } catch (error) { errors.push({ session_id: child.id, phase: "usage_child", ...errorPayload(error) }); } } return errors; }
   async productive(name, task) {
+    assertDaemonDispatch(this.paths?.dir);
     if (this.active) throw new DaemonError("operation_busy", `${this.active} is already running`, true);
     this.active = name; let result, completed = false;
     try {
       await this.persist({ active_tree: true, active_operation: name });
+      assertDaemonDispatch(this.paths?.dir);
       result = await task(); completed = true;
       await this.persist({ active_tree: !result?.settled, active_operation: null });
       await this.journal(name, result); return result;
@@ -118,6 +141,7 @@ export class Runtime {
   }
   async dispatch(operation, params) {
     const assertDispatch = captureDispatchGuard(this);
+    if (["session.create", "session.prompt", "session.start", "session.fork", "session.resume", "session.archive"].includes(operation)) assertDaemonDispatch(this.paths.dir);
     if (operation === "daemon.status") { const { password, ...config } = this.state.config; return { daemon_id: this.state.daemon_id, pid: process.pid, socket: this.paths.socket, versions: this.state.versions, auth_status: this.state.auth_status, shutdown_state: this.state.shutdown_state, active_tree: this.state.active_tree, active_operation: this.active, sessions: [...this.sessions.values()], config }; }
     if (operation === "daemon.doctor") { const doc = await this.client.request("GET", "/doc", undefined, { directory: null, timeoutMs: 5_000 }); const apiFingerprint = createHash("sha256").update(typeof doc === "string" ? doc : JSON.stringify(doc)).digest("hex"); if (apiFingerprint !== OPENCODE_BASELINE.api_fingerprint) throw new DaemonError("api_fingerprint_mismatch", "OpenCode API differs from the qualified baseline", false, { expected: OPENCODE_BASELINE.api_fingerprint, actual: apiFingerprint }); return { compatible: true, versions: this.state.versions, api_fingerprint: apiFingerprint, plugin_sha256: this.state.config.pluginSha256, auth_status: this.state.auth_status }; }
     if (operation === "session.create") return this.productive(operation, async () => {
@@ -177,6 +201,7 @@ export class Runtime {
     }
     if (operation === "session.archive") { const observed = await this.requireSettled(params.sessionId); const payload = await run(this.state.config.bin, ["export", params.sessionId], this.state.config); const manifest = { schema_id: ARCHIVE_SCHEMA, provider_session_id: params.sessionId, source_cwd: observed.cwd, opencode_version: this.state.versions.opencode, history_sha256: observed.history_sha256, exported_at: new Date().toISOString(), payload: Buffer.from(payload).toString("base64") }; const file = absolute(params.archiveFile, "--archive-file"); await writeJson(file, manifest); return { archive_file: file, ...manifest, payload: undefined, sha256: createHash("sha256").update(await readFile(file)).digest("hex") }; }
     if (operation === "daemon.stop") {
+      return await completeDaemonShutdown(this.paths.dir, this.state, patch => this.persist(patch), async () => {
       if (params.cancelTree) cancelPendingDispatch(this);
       await waitForSettlement({
         knownSessions: () => [...this.sessions.keys()],
@@ -187,18 +212,15 @@ export class Runtime {
         },
         ...(params.cancelTree ? { cancel: id => this.cancelTree(id) } : {})
       });
-      // The durable clean receipt must follow the owned physical stop barrier.
-      await stopProcessGroup(this.child);
-      await this.persist({ active_tree: false, shutdown_state: "stopping" });
-      return { stopped: true, clean: true, _shutdown: true };
+      return { clean: true };
+      }, [
+        ["provider", () => stopProcessGroup(this.child)],
+        ["provider_resource", () => finishDaemonProcess(this.state.config, this.providerProcess)],
+        ["daemon_resource", () => finishDaemonProcess(this.state.config, this.state.resource_process)]
+      ]);
     }
     throw new DaemonError("unknown_operation", `unknown daemon operation: ${operation}`);
   }
-  async shutdown() {
-    try { await stopProcessGroup(this.child); await finishDaemonProcess(this.state.config, this.providerProcess, "stopped", "daemon_stopped"); }
-    catch (error) { await this.persist({ shutdown_state: "cleanup_failed", cleanup_error: errorPayload(error) }); throw error; }
-    await finishDaemonProcess(this.state.config, this.state.resource_process); await this.persist({ shutdown_state: "clean", stopped_at: new Date().toISOString() }); await new Promise(resolve => this.server.close(resolve)); await removeSocket(this.paths);
-  }
 }
 export async function serveDaemon(stateDir) {
   const paths = locations(stateDir), state = await readState(paths.dir); if (state?.schema_id !== STATE_SCHEMA) throw new DaemonError("daemon_state_missing", "daemon state is missing");
@@ -211,7 +233,28 @@ export async function serveDaemon(stateDir) {
     scheduleDaemonProcessHeartbeat(state.config, state.resource_process);
     scheduleDaemonProcessHeartbeat(state.config, providerProcess);
     state.provider_process = providerProcess; state.provider_pid = child.pid; await writeJson(paths.state, state);
-    const client = new OpenCodeClient({ baseUrl: `http://127.0.0.1:${port}`, password: state.config.password, directory: state.config.cwd }); const deadline = Date.now() + 15_000; while (Date.now() < deadline) { try { await client.health(); break; } catch { await new Promise(resolve => setTimeout(resolve, 100)); } } await client.health(); const runtime = new Runtime(paths, state, client, child, providerProcess); await removeSocket(paths); const server = net.createServer(connection => { connection.setEncoding("utf8"); let buffer = ""; connection.on("data", chunk => { buffer += chunk; const i = buffer.indexOf("\n"); if (i < 0) return; void (async () => { let request; try { request = JSON.parse(buffer.slice(0, i)); const result = await durableDaemonDispatch(paths.dir, request, () => runtime.dispatch(request.operation, request.params ?? {}), () => runtime.dispatch("session.inspect", request.params ?? {})); connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request.id, ok: true, result: { ...result, _shutdown: undefined } })}\n`); if (result?._shutdown) setImmediate(() => runtime.shutdown().then(() => process.exit(0)).catch(error => runtime.persist({ shutdown_state: "cleanup_failed", cleanup_error: errorPayload(error) }))); } catch (error) { connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request?.id ?? null, ok: false, error: errorPayload(error) })}\n`); } })(); }); }); runtime.server = server; await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); }); await chmod(paths.socket, 0o600); await runtime.persist({ shutdown_state: "running", active_tree: false, server_port: port }); return await new Promise(() => {});
+    const client = new OpenCodeClient({ baseUrl: `http://127.0.0.1:${port}`, password: state.config.password, directory: state.config.cwd }); const deadline = Date.now() + 15_000; while (Date.now() < deadline) { try { await client.health(); break; } catch { await new Promise(resolve => setTimeout(resolve, 100)); } } await client.health(); const runtime = new Runtime(paths, state, client, child, providerProcess); await removeSocket(paths);
+    const server = net.createServer(connection => {
+      connection.setEncoding("utf8"); let buffer = "", consumed = false;
+      connection.on("data", chunk => {
+        if (consumed) return;
+        buffer += chunk; const at = buffer.indexOf("\n"); if (at < 0) return; consumed = true;
+        void (async () => {
+          let request;
+          try {
+            request = JSON.parse(buffer.slice(0, at));
+            const result = await durableDaemonDispatch(paths.dir, request, () => runtime.dispatch(request.operation, request.params ?? {}), () => runtime.dispatch("session.inspect", request.params ?? {}));
+            if (result._shutdown) retireDaemonEndpoint(server);
+            connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request.id, ok: true, result: { ...result, _shutdown: undefined } })}\n`);
+          } catch (error) {
+            await retireFailedDaemon(server, state, error, patch => runtime.persist(patch));
+            connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request?.id ?? null, ok: false, error: errorPayload(error) })}\n`);
+          }
+        })();
+      });
+    });
+    trackDaemonConnections(server); runtime.server = server;
+    await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); }); await chmod(paths.socket, 0o600); await runtime.persist({ shutdown_state: "running", active_tree: false, server_port: port }); return await new Promise(() => {});
   } catch (error) {
     await cleanupFailedStart(state.config, providerProcess, child, error); throw error;
   }

package/dist/harness-runtime/lib/dd-zcode-daemon.mjs

@@ -3,9 +3,9 @@ import { spawn } from "node:child_process";
 import { createHash, randomUUID } from "node:crypto";
 import { chmod, lstat, mkdir, open, readFile, rename, unlink, writeFile } from "node:fs/promises";
 import net from "node:net";
-import { durableDaemonDispatch } from "./daemon-operations.mjs";
+import { durableDaemonDispatch, completeDaemonShutdown, commitDaemonShutdown, assertDaemonDispatch, assertDaemonRestartable, trackDaemonConnections, retireDaemonEndpoint, retireFailedDaemon } from "./daemon-operations.mjs";
 import path from "node:path";
-import { prepareRuntimeOwner, cleanupFailedDaemonStart, closeManagedBridge, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, scheduleDaemonProcessHeartbeat, registerDaemonProcess, startManagedBridge, stopProcessGroup } from "./managed-daemon.mjs";
+import { prepareRuntimeOwner, cleanupFailedDaemonStart, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, scheduleDaemonProcessHeartbeat, registerDaemonProcess, startManagedBridge, stopProcessGroup } from "./managed-daemon.mjs";
 
 import {
   AcpBridge,
@@ -163,13 +163,14 @@ export async function startDaemon(options) {
   await prepareRuntimeOwner(config, paths.dir);
   await mkdir(paths.dir, { recursive: true, mode: 0o700 });
   await chmod(paths.dir, 0o700);
+  let previous = await readState(paths.dir); assertDaemonRestartable(previous);
   try {
     const status = await callDaemon(paths.dir, "daemon.status", {}, 1000);
     if (!sameConfig(status, config)) throw new DaemonError("daemon_config_mismatch", "a daemon is already running with different configuration");
     return { ...status, already_running: true };
   } catch (error) { if (error.code !== "daemon_not_running" && error.code !== "daemon_timeout") throw error; }
 
-  const previous = await readState(paths.dir);
+  previous = await readState(paths.dir); assertDaemonRestartable(previous);
   if (previous?.shutdown_state === "clean") {
     if (!sameNativeRouting(previous.config, config)) throw new DaemonError("daemon_config_mismatch", "retained ZCode native storage routing changed");
     await authorizeRetainedDaemonResume(paths.dir, previous, options.sessionId, config);
@@ -227,7 +228,7 @@ export async function startDaemon(options) {
 export async function stopDaemon(options) {
   const timeoutMs = options.timeoutMs ?? 30_000;
   const { socket } = locations(options.stateDir);
-  return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), socket, timeoutMs);
+  return await confirmDaemonStopped(() => callDaemon(options.stateDir, "daemon.stop", { cancelTree: options.cancelTree === true }, timeoutMs), socket, timeoutMs, { readState: () => readState(options.stateDir) });
 }
 
 export class DaemonRuntime {
@@ -241,7 +242,7 @@ export class DaemonRuntime {
   persist(patch = {}) {
     Object.assign(this.state, patch, { pid: process.pid, updated_at: new Date().toISOString(), sessions: [...this.sessions.values()] });
     const snapshot = structuredClone(this.state);
-    this.persisting = this.persisting.then(() => writeState(this.paths.state, snapshot));
+    this.persisting = this.persisting.catch(() => {}).then(() => writeState(this.paths.state, snapshot));
     return this.persisting;
   }
 
@@ -448,6 +449,7 @@ export class DaemonRuntime {
   }
 
   async dispatch(operation, params) {
+    if (["session.create", "session.prompt", "session.start", "session.fork", "session.resume"].includes(operation)) assertDaemonDispatch(this.paths.dir);
     if (operation === "daemon.status") return {
       daemon_id: this.state.daemon_id, pid: process.pid, socket: this.paths.socket,
       cwd: this.state.cwd, journal: this.state.journal, versions: this.state.versions,
@@ -548,12 +550,17 @@ export class DaemonRuntime {
       await this.persist({ active_tree: this.treeRunning(result) || Boolean(this.activeProductive) });
       return result;
     }
-    if (operation === "daemon.stop") return await this.prepareStop(params.cancelTree === true);
+    if (operation === "daemon.stop") return await completeDaemonShutdown(this.paths.dir, this.state, patch => this.persist(patch), () => this.prepareStop(params.cancelTree === true), [
+      ["provider_resource", () => finishDaemonProcess(this.state.config, this.state.provider_process)],
+      ["daemon_resource", () => finishDaemonProcess(this.state.config, this.state.resource_process)]
+    ]);
     throw new DaemonError("unknown_operation", `unknown daemon operation: ${operation}`);
   }
 
   async prepareStop(cancel) {
+    if (this.activeProductive && !cancel) throw new DaemonError("tree_not_settled", "productive operation is running");
     if (cancel) this.controlGeneration = (this.controlGeneration ?? 0) + 1;
+    if (!this.state.native_stop_proof || this.state.native_stop_proof.daemon_id !== this.state.daemon_id) {
     const unsettled = [];
     for (const session of this.sessions.values()) {
       if (session.close_receipt) {
@@ -603,6 +610,10 @@ export class DaemonRuntime {
       }
     }
     if (unsettled.length) throw new DaemonError("tree_not_settled", "daemon still owns running Sessions", false, { sessions: unsettled });
+    commitDaemonShutdown(this.paths.dir);
+    await this.persist({ native_stop_proof: { daemon_id: this.state.daemon_id, sessions: [...this.sessions.keys()], observed_at: new Date().toISOString() } });
+    }
+    commitDaemonShutdown(this.paths.dir);
     // A cooperative close can make the native Session non-resident while its
     // original ACP prompt is still pending.  For an explicitly whole-RUN
     // cancellation this daemon owns the bridge process group, so retire that
@@ -614,7 +625,7 @@ export class DaemonRuntime {
         if (typeof this.bridge.close === "function") await this.bridge.close();
         bridgeClosed = true;
       } catch (error) {
-        throw new DaemonError(error.code ?? "provider_process_stop_incomplete", error.message ?? String(error), false);
+        throw new DaemonError(error.code ?? "provider_process_stop_incomplete", error.message ?? String(error), false, error.details);
       }
     }
     const deadline = Date.now() + 5000;
@@ -625,24 +636,11 @@ export class DaemonRuntime {
     // A bridge without a close method occurs only in in-memory diagnostics: it
     // owns no OS process boundary. Production AcpBridge always has one.
     try { if (!bridgeClosed && typeof this.bridge.close === "function") await this.bridge.close(); }
-    catch (error) { throw new DaemonError(error.code ?? "provider_process_stop_incomplete", error.message ?? String(error), false); }
+    catch (error) { throw new DaemonError(error.code ?? "provider_process_stop_incomplete", error.message ?? String(error), false, error.details); }
     await this.persist({ active_tree: false, shutdown_state: "stopping", active_operation: null, active_request: null, physical_settlement: { confirmed_at: new Date().toISOString(), bridge_pid: this.bridge.child?.pid ?? null } });
     return { daemon_id: this.state.daemon_id, stopped: true, clean: true, physical_settlement: this.state.physical_settlement, _shutdown: true };
   }
 
-  async shutdown() {
-    // prepareStop has already closed the owned bridge. Signal shutdown may get
-    // here directly, so close remains idempotent.
-    await closeManagedBridge(this.state.config, this.state, this.bridge);
-    await this.persist({ active_tree: false, shutdown_state: "clean", stopped_at: new Date().toISOString() });
-    const resourceProcess = this.state.resource_process;
-    if (resourceProcess?.id && resourceProcess.lease_token) {
-      try { await finishDaemonProcess(this.state.config, resourceProcess, "stopped", "daemon_stopped"); }
-      catch { /* Keep the state; the next reconciler owns stale processes. */ }
-    }
-    await new Promise((resolve) => this.server.close(resolve));
-    await removeSocket(this.paths.socket, this.paths.dir);
-  }
 }
 
 export function zcodeBridgeEnv(config, daemonId) {
@@ -660,11 +658,13 @@ export async function serveDaemon(stateDir) {
   scheduleDaemonProcessHeartbeat(state.config, state.resource_process);
   await removeSocket(paths.socket, paths.dir);
   const server = net.createServer((connection) => {
-    connection.setEncoding("utf8"); let buffer = "";
+    connection.setEncoding("utf8"); let buffer = "", consumed = false;
     connection.on("data", (chunk) => {
+      if (consumed) return;
       buffer += chunk;
       const newline = buffer.indexOf("\n");
       if (newline < 0) return;
+      consumed = true;
       const line = buffer.slice(0, newline); buffer = buffer.slice(newline + 1);
       void (async () => {
         let request;
@@ -672,15 +672,17 @@ export async function serveDaemon(stateDir) {
           request = JSON.parse(line);
           if (request.schema_id !== REQUEST_SCHEMA || !request.id || !request.operation) throw new DaemonError("daemon_protocol_mismatch", "invalid daemon request");
           const result = await durableDaemonDispatch(paths.dir, request, () => runtime.dispatch(request.operation, request.params ?? {}), () => runtime.dispatch("session.inspect", request.params ?? {}));
+          if (result?._shutdown) retireDaemonEndpoint(server);
           connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request.id, ok: true, result: { ...result, _shutdown: undefined } })}\n`);
-          if (result?._shutdown) setImmediate(() => void runtime.shutdown().then(() => process.exit(0)));
         } catch (error) {
+          await retireFailedDaemon(server, state, error, patch => runtime.persist(patch));
           connection.end(`${JSON.stringify({ schema_id: RESPONSE_SCHEMA, id: request?.id ?? null, ok: false, error: errorPayload(error) })}\n`);
         }
       })();
     });
   });
   runtime.server = server;
+  trackDaemonConnections(server);
   await new Promise((resolve, reject) => { server.once("error", reject); server.listen(paths.socket, resolve); });
   await chmod(paths.socket, 0o600);
   await runtime.persist({ shutdown_state: "running", active_tree: false });