@deksden-com/dd-flow-cli 0.9.0-beta.38 → 0.9.0-beta.40

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

@@ -5,7 +5,7 @@ import { chmod, lstat, mkdir, open, readFile, rename, unlink, writeFile } from "
 import net from "node:net";
 import { durableDaemonDispatch } from "./daemon-operations.mjs";
 import path from "node:path";
-import { cleanupFailedStart, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, heartbeatDaemonProcess, registerDaemonProcess, stopProcessGroup } from "./managed-daemon.mjs";
+import { prepareRuntimeOwner, cleanupFailedStart, confirmDaemonProcess, confirmDaemonStopped, finishDaemonProcess, heartbeatDaemonProcess, registerDaemonProcess, stopProcessGroup } from "./managed-daemon.mjs";
 
 import {
   AcpBridge,
@@ -135,6 +135,7 @@ export async function startDaemon(options) {
     env: options.ddFlowHome ? { DD_FLOW_HOME: absolute(options.ddFlowHome, "--dd-flow-home") } : {},
     livenessTimeoutMs: options.livenessTimeoutMs ?? DEFAULT_LIVENESS_TIMEOUT_MS,
   };
+  await prepareRuntimeOwner(config, paths.dir);
   await mkdir(paths.dir, { recursive: true, mode: 0o700 });
   await chmod(paths.dir, 0o700);
   try {
@@ -243,7 +244,9 @@ export class DaemonRuntime {
   }
 
   treeRunning(result) {
-    return result?.read?.projection?.status === "running" || (result?.subagents?.running ?? result?.evidence?.subagents?.running ?? result?.after?.running ?? []).length > 0;
+    const evidence = result?.evidence ?? result;
+    const children = evidence?.subagents?.running;
+    return evidence?.read?.projection?.status !== "idle" || !Array.isArray(children) || children.length > 0 || Boolean(evidence?.subagents?.observation_error);
   }
 
   async productive(name, task) {
@@ -293,7 +296,7 @@ export class DaemonRuntime {
       await this.persist(); return result;
     }
     if (operation === "session.fork") return await this.productive(operation, async () => { await this.requireSettled(params); return await forkSessionWithBridge(this.bridge, this.options(params)); });
-    if (operation === "session.inspect") {
+    if (operation === "session.inspect" || operation === "session.resume") {
       const adapter = params.adapterSessionId ?? params.sessionId;
       if (!this.liveAdapters.has(adapter) && this.activeProductive) {
         throw new DaemonError(
@@ -311,7 +314,7 @@ export class DaemonRuntime {
       if (this.sessions.has(result.provider_session_id)) this.track(result, result.subagents);
       else if (result.adapter_session_id) this.liveAdapters.add(result.adapter_session_id);
       await this.persist({ active_tree: this.treeRunning(result) || Boolean(this.activeProductive) });
-      return result;
+      return { ...result, settled: !this.treeRunning(result) && !this.activeProductive };
     }
     if (operation === "session.cancel") {
       if (!this.sessions.has(params.sessionId)) throw new DaemonError("session_identity_mismatch", "cancellation Session is not owned by this daemon");
@@ -393,7 +396,7 @@ export async function serveDaemon(stateDir) {
         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 ?? {}));
+          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) {

package/dist/harness-runtime/lib/managed-daemon.mjs

@@ -1,25 +1,143 @@
-import { lstat } from "node:fs/promises";
+import { lstat, readFile, readdir, mkdir } from "node:fs/promises";
+import { randomUUID } from "node:crypto";
+import path from "node:path";
 import { commandJson } from "./process-json.mjs";
 import { errorRecord } from "./operation-errors.mjs";
+import { writeJsonAtomic } from "./runner-events.mjs";
 
 const leases = new Map();
 
-export function assertDaemonOwnership() {
+/** Freeze resource ownership independently of whether Subject hooks are enabled. */
+export function bindRuntimeOwner(config, env = process.env) {
+  const raw = Object.hasOwn(config, "runtime_owner") ? config.runtime_owner === null ? undefined : JSON.stringify(config.runtime_owner) : env.DD_FLOW_RUNTIME_OWNER;
+  if (raw === undefined) return config.runtime_owner = null;
+  let owner;
+  try { if (raw.length > 16_384) throw new Error(); owner = JSON.parse(raw); }
+  catch { throw Object.assign(new Error("runtime ownership descriptor is invalid"), { code: "runtime_owner_invalid" }); }
+  const fields = ["schema_id", "owner_id", "role", "project_id", "run_id", "work_id", "operation_id", "state_dir", "dd_flow_home", "dd_flow_bin", "resource_home", "adapter_executable", "budget"];
+  const text = value => typeof value === "string" && value.trim().length > 0;
+  if (!owner || typeof owner !== "object" || Array.isArray(owner) || Object.keys(owner).some(key => !fields.includes(key))
+    || owner.schema_id !== "dd-flow/runtime-owner@1" || !text(owner.owner_id) || !text(owner.operation_id)
+    || !["subject", "worker", "merge", "judge", "probe"].includes(owner.role)
+    || ["state_dir", "dd_flow_home", "dd_flow_bin", "resource_home"].some(key => !text(owner[key]) || !path.isAbsolute(owner[key]))
+    || (owner.adapter_executable !== undefined && (!text(owner.adapter_executable) || !path.isAbsolute(owner.adapter_executable)))
+    || ["project_id", "run_id", "work_id"].some(key => owner[key] !== undefined && !text(owner[key]))
+    || (["subject", "worker", "merge"].includes(owner.role) && (!text(owner.project_id) || !text(owner.run_id)))
+    || (["worker", "merge"].includes(owner.role) && !text(owner.work_id))
+    || (["judge", "probe"].includes(owner.role) && owner.work_id !== undefined)) {
+    throw Object.assign(new Error("runtime ownership descriptor does not match its role and scope"), { code: "runtime_owner_invalid" });
+  }
+  if (owner.budget !== undefined) {
+    const budget = owner.budget, names = ["codex-desktop", "zcode-acp", "grok-acp", "antigravity-cli", "opencode-server", "droid-cli"];
+    if (!budget || typeof budget !== "object" || Array.isArray(budget) || Object.keys(budget).some(key => !["schema_id", "scope_id", "per_harness"].includes(key)) || budget.schema_id !== "dd-flow/runtime-budget@1" || !/^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/.test(budget.scope_id ?? "") || !budget.per_harness || typeof budget.per_harness !== "object" || Array.isArray(budget.per_harness) || Object.entries(budget.per_harness).some(([name, limit]) => !names.includes(name) || !Number.isSafeInteger(limit) || limit < 1 || limit > 1024)) throw Object.assign(new Error("runtime owner budget is invalid"), { code: "runtime_budget_invalid" });
+  }
+  config.runtime_owner = Object.fromEntries(fields.filter(key => owner[key] !== undefined).map(key => [key, owner[key]]));
+  return config.runtime_owner;
+}
+
+/** Reject rebinding before provider setup, including the already-running path. */
+export async function prepareRuntimeOwner(config, stateDir, env = process.env) {
+  const owner = bindRuntimeOwner(config, env);
+  if (owner && path.resolve(owner.state_dir) !== path.resolve(stateDir)) {
+    throw Object.assign(new Error("runtime owner belongs to another state directory"), { code: "runtime_owner_mismatch" });
+  }
+  let previous;
+  try { previous = JSON.parse(await readFile(path.join(stateDir, "daemon.json"), "utf8")); }
+  catch (error) { if (error.code === "ENOENT") return owner; throw error; }
+  const retained = bindRuntimeOwner({ runtime_owner: previous?.config?.runtime_owner ?? null }, {});
+  if (JSON.stringify(retained) !== JSON.stringify(owner)) {
+    throw Object.assign(new Error("retained daemon ownership differs; use explicit recovery instead of rebinding its sessions"), { code: "runtime_owner_mismatch" });
+  }
+  return owner;
+}
+
+export async function assertDaemonOwnership(stateDir, dispatch = {}, assertCurrent = () => {}) {
   for (const lease of leases.values()) if (lease.error) throw lease.error;
+  const state = JSON.parse(await readFile(path.join(stateDir, "daemon.json"), "utf8"));
+  const owner = bindRuntimeOwner({ runtime_owner: state.config?.runtime_owner ?? null }, {});
+  if (!owner) return;
+  const record = state.resource_process;
+  if (owner.state_dir !== stateDir || !record?.id || !record.lease_token) throw Object.assign(new Error("owned daemon has no confirmed resource binding"), { code: "process_ownership_unknown" });
+  for (;;) {
+    assertCurrent();
+    const result = await runtimeProcess(state.config, "admission", { id: record.id, "lease-token": record.lease_token, owner: owner.owner_id, "state-dir": stateDir, "daemon-id": state.daemon_id ?? state.config?.daemonId, ...dispatch });
+    assertCurrent();
+    if (result?.ok === true && result.admitted === true && result.process_id === record.id) return;
+    if (result?.ok !== true || result.reason !== "provider_capacity" || result.process_id !== record.id) throw Object.assign(new Error("productive runtime admission was not confirmed"), { code: "process_ownership_unknown" });
+    // The daemon, not its disposable socket client, owns this queue. Recheck
+    // the RUN fence and process lease on every admission attempt.
+    await new Promise(resolve => setTimeout(resolve, 250));
+  }
+}
+
+/** Snapshot terminal operations BEFORE observing native idle. A later Turn
+ * cannot be released by an older observation of the same Session. */
+export async function budgetObservationScope(stateDir, sessionId) {
+  const state = await budgetState(stateDir);
+  if (!state?.config?.runtime_owner?.budget || !sessionId) return null;
+  const daemonId = state.daemon_id ?? state.config?.daemonId, operationIds = [];
+  for (const name of await readdir(path.join(stateDir, "operations")).catch(error => { if (error.code === "ENOENT") return []; throw error; })) {
+    const directory = path.join(stateDir, "operations", name);
+    try {
+      const requested = JSON.parse(await readFile(path.join(directory, "requested.json"), "utf8"));
+      const result = JSON.parse(await readFile(path.join(directory, "result.json"), "utf8"));
+      if (requested.daemon_id === daemonId && requested.session_id === sessionId && ["completed", "failed"].includes(result.state) && ["session.prompt", "session.start", "session.resume"].includes(requested.operation)) operationIds.push(requested.operation_id);
+    } catch (error) { if (error.code !== "ENOENT") throw error; }
+  }
+  return { stateDir, state, daemonId, sessionId, operationIds };
+}
+
+export async function releaseBudgetObservation(scope, observation) {
+  if (!scope || !scope.operationIds.length || observation?.settled !== true) return;
+  const id = observation.provider_session_id ?? observation.session_id ?? observation.adapter_session_id;
+  if (id !== scope.sessionId) throw Object.assign(new Error("budget observation belongs to another Session"), { code: "session_identity_mismatch" });
+  const directory = path.join(scope.stateDir, "budget-observations");
+  await mkdir(directory, { recursive: true, mode: 0o700 });
+  const file = path.join(directory, `${randomUUID()}.json`);
+  await writeJsonAtomic(file, { daemon_id: scope.daemonId, session_id: scope.sessionId, settled: true, operation_ids: scope.operationIds, observed_at: new Date().toISOString() });
+  await releaseBudget(scope.state, { "observation-file": file });
+}
+
+export async function completeBudgetOperation(stateDir, request, result, observe) {
+  const state = await budgetState(stateDir);
+  if (!state?.config?.runtime_owner?.budget) return;
+  if (["session.create", "session.fork"].includes(request.operation)) return await releaseBudget(state, { "operation-id": request.id });
+  const scope = await budgetObservationScope(stateDir, request.params?.sessionId);
+  if (!scope?.operationIds.length) return;
+  const observation = observe ? await observe() : result;
+  await releaseBudgetObservation(scope, observation);
+}
+
+async function budgetState(stateDir) {
+  try { return JSON.parse(await readFile(path.join(stateDir, "daemon.json"), "utf8")); }
+  catch (error) { if (error.code === "ENOENT") return null; throw error; }
+}
+
+async function releaseBudget(state, options) {
+  const record = state.resource_process, owner = state.config.runtime_owner;
+  const receipt = await runtimeProcess(state.config, "release-turn", { id: record.id, "lease-token": record.lease_token, owner: owner.owner_id, "daemon-id": state.daemon_id ?? state.config?.daemonId, ...options });
+  if (receipt?.ok !== true) throw Object.assign(new Error("runtime budget release was not confirmed"), { code: "runtime_budget_release_unconfirmed" });
 }
 
 export async function runtimeProcess(config, action, options = {}) {
-  if (!config.resourceHome || !config.ddFlowHome || !config.ddFlowBin) return null;
+  const owner = config.runtime_owner;
+  const resourceHome = owner?.resource_home ?? config.resourceHome;
+  const ddFlowHome = owner?.dd_flow_home ?? config.ddFlowHome;
+  const ddFlowBin = owner?.dd_flow_bin ?? config.ddFlowBin;
+  if (!resourceHome || !ddFlowHome || !ddFlowBin) return null;
   const args = ["runtime", "process", action];
   for (const [key, value] of Object.entries(options)) if (value !== null && value !== undefined) args.push(`--${key}`, String(value));
-  return await commandJson(config.ddFlowBin, args, {
+  return await commandJson(ddFlowBin, args, {
     cwd: config.cwd,
-    env: { ...(config.env ?? {}), DD_FLOW_HOME: config.ddFlowHome, DD_FLOW_RESOURCE_HOME: config.resourceHome }
+    env: { ...(config.env ?? {}), DD_FLOW_HOME: ddFlowHome, DD_FLOW_RESOURCE_HOME: resourceHome }
   });
 }
 
 export async function registerDaemonProcess(config, input) {
-  const result = await runtimeProcess(config, "register", { ...input, "owner-pid": process.pid, "lease-ms": 300_000 });
+  const owner = bindRuntimeOwner(config);
+  const harness = ({ "codex-daemon": "codex-desktop", "zcode-daemon": "zcode-acp", "grok-daemon": "grok-acp", "agy-daemon": "antigravity-cli", "opencode-daemon": "opencode-server", "droid-daemon": "droid-cli" })[input.kind];
+  const scope = owner ? { owner: owner.owner_id, "project-id": owner.project_id, run: owner.run_id, work: owner.work_id, role: owner.role, "state-dir": owner.state_dir, "adapter-executable": owner.adapter_executable, ...(owner.budget ? { "budget-json": JSON.stringify(owner.budget), harness } : {}), ...(owner.project_id && owner.run_id || owner.budget ? { "dispatch-barrier": "run-generation@1" } : {}), operation: `${owner.operation_id}:${input.operation ?? input.kind}` } : {};
+  const result = await runtimeProcess(config, "register", { ...input, ...scope, "owner-pid": process.pid, "lease-ms": 300_000 });
   return result?.process ? { id: result.process.id, lease_token: result.process.lease_token, kind: input.kind } : null;
 }
 
@@ -98,6 +216,7 @@ export async function stopProcessGroup(child, graceMs = 1_000) {
   process.kill(target, "SIGTERM");
   await new Promise((resolve) => setTimeout(resolve, graceMs));
   if (!alive()) return;
+  if (child.exitCode != null || child.signalCode != null) return stopProcessGroup(child, graceMs);
   process.kill(target, "SIGKILL");
   // Grace controls escalation, not the OS acknowledgement budget. A caller
   // testing immediate escalation must still allow the kernel to reap children.

package/dist/harness-runtime/lib/process-json.mjs

@@ -7,7 +7,7 @@ function failure(message, code) {
 
 function reportedFailure(stdout, stderr, fallback) {
   for (const text of [stdout, stderr]) {
-    for (const line of text.trim().split(/\n/).reverse()) {
+    for (const line of [text.trim(), ...text.trim().split(/\n/).reverse()]) {
       try {
         const value = JSON.parse(line);
         const error = value?.error;

package/dist/harness-runtime/lib/session-settlement.mjs

@@ -18,11 +18,13 @@ export async function waitForSettlement({ observe, cancel, knownSessions = () =>
       state = { sessions: knownSessions(), active: true, observation_error: errorRecord(error) };
     }
     if (!state.active && state.sessions.length === 0) return { ...state, cleanup_errors: errors };
-    if (!cancel || performance.now() >= deadline) throw Object.assign(new Error("daemon still owns an unsettled Session tree"), { code: "tree_not_settled", details: { ...state, cleanup_errors: errors } });
-    for (const id of state.sessions) if (!cancelled.has(id)) {
+    // The observation budget must not skip a newly discovered owned target.
+    // Attempt its cancellation once, then leave physical settlement pending.
+    for (const id of state.sessions) if (cancel && !cancelled.has(id)) {
       cancelled.add(id);
       try { await cancel(id); } catch (error) { errors.push({ phase: "cancel", session_id: id, ...errorRecord(error) }); }
     }
+    if (!cancel || performance.now() >= deadline) throw Object.assign(new Error("daemon still owns an unsettled Session tree"), { code: "tree_not_settled", details: { ...state, cleanup_errors: errors } });
     await new Promise(resolve => setTimeout(resolve, 50));
   }
 }

package/dist/schemas/run-control-receipt.schema.json

@@ -0,0 +1,99 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "dd-flow/run-control-receipt@1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["schema_id", "ok", "request_id", "control_id", "scope", "generation", "requested_mode", "applied_mode", "state", "settled", "nodes", "unsettled_reasons", "resumability"],
+  "properties": {
+    "schema_id": { "const": "dd-flow/run-control-receipt@1" },
+    "ok": { "const": true },
+    "request_id": { "type": ["string", "null"], "minLength": 1, "maxLength": 128 },
+    "control_id": { "type": ["string", "null"], "pattern": "^CTL-[A-Za-z0-9][A-Za-z0-9._-]*$" },
+    "scope": {
+      "type": "object", "additionalProperties": false,
+      "required": ["project_id", "run_id"],
+      "properties": {
+        "project_id": { "type": "string", "minLength": 1 },
+        "run_id": { "type": "string", "pattern": "^RUN-[A-Za-z0-9][A-Za-z0-9._-]*$" }
+      }
+    },
+    "generation": { "type": "integer", "minimum": 0 },
+    "requested_mode": { "enum": [null, "pause", "stop", "resume"] },
+    "applied_mode": { "enum": [null, "pause", "stop", "resume"] },
+    "state": { "enum": ["not_requested", "pause_requested", "paused", "stop_requested", "stopped", "resuming", "running", "terminal_noop", "superseded", "recovery_required"] },
+    "settled": { "type": "boolean" },
+    "nodes": {
+      "type": "array",
+      "items": {
+        "type": "object", "additionalProperties": false,
+        "required": ["id", "role", "status", "acknowledged", "settled", "reason"],
+        "properties": {
+          "id": { "type": "string", "minLength": 1 },
+          "role": { "enum": ["subject", "worker", "merge", "judge", "probe", "check", "process", "unknown"] },
+          "status": { "enum": ["pending", "unknown", "idle", "stopped", "completed"] },
+          "acknowledged": { "type": "boolean" },
+          "settled": { "type": "boolean" },
+          "reason": { "type": ["string", "null"] },
+          "receipt_path": { "type": "string", "minLength": 1 },
+          "details": { "type": "object" }
+        },
+        "allOf": [{
+          "if": { "properties": { "settled": { "const": true } } },
+          "then": { "properties": { "status": { "enum": ["idle", "stopped", "completed"] } } }
+        }]
+      }
+    },
+    "unsettled_reasons": { "type": "array", "uniqueItems": true, "items": { "type": "string", "minLength": 1 } },
+    "resumability": { "enum": ["pending", "same_session", "replacement_allowed", "recovery_required", "not_applicable"] },
+    "recovery_id": { "type": ["string", "null"] },
+    "reused": { "type": "boolean" },
+    "next_cursor": { "type": "integer", "minimum": 0 },
+    "run_status": { "type": "string" },
+    "control": { "type": ["object", "null"] },
+    "operations": { "type": "array", "items": { "type": "object" } },
+    "worker": { "type": ["object", "null"] },
+    "resume_operation": { "type": ["object", "null"] },
+    "controller": { "type": "object" },
+    "operation_id": { "type": "string", "minLength": 1 },
+    "status": { "type": "string" },
+    "action": { "enum": ["pause", "stop", "resume"] },
+    "progress": {
+      "type": "array",
+      "items": { "type": "object", "additionalProperties": false, "required": ["phase", "message", "at"], "properties": { "phase": { "enum": ["start", "running", "done", "failed"] }, "message": { "type": "string" }, "at": { "type": "string" }, "detail": { "type": "object" } } }
+    },
+    "warnings": { "type": "array", "items": { "type": "object" } },
+    "response_file": { "type": "string", "minLength": 1 },
+    "response_file_error": { "type": "string" }
+  },
+  "allOf": [
+    {
+      "if": { "properties": { "state": { "not": { "enum": ["not_requested", "terminal_noop"] } } } },
+      "then": { "properties": { "control_id": { "type": "string" }, "request_id": { "type": "string" }, "requested_mode": { "type": "string" } } }
+    },
+    {
+      "if": { "properties": { "state": { "enum": ["paused", "stopped"] } } },
+      "then": { "properties": { "settled": { "const": true } } }
+    },
+    {
+      "if": { "properties": { "state": { "const": "paused" } } },
+      "then": { "properties": { "applied_mode": { "const": "pause" } } }
+    },
+    {
+      "if": { "properties": { "state": { "const": "stopped" } } },
+      "then": { "properties": { "applied_mode": { "const": "stop" } } }
+    },
+    {
+      "if": { "properties": { "settled": { "const": true } } },
+      "then": {
+        "properties": {
+          "nodes": { "items": { "properties": { "settled": { "const": true } } } },
+          "unsettled_reasons": { "maxItems": 0 }
+        }
+      }
+    },
+    {
+      "if": { "properties": { "settled": { "const": false } } },
+      "then": { "properties": { "unsettled_reasons": { "minItems": 1 }, "resumability": { "enum": ["pending", "recovery_required", "not_applicable"] } } }
+    }
+  ]
+}

package/dist/schemas/run-control-request.schema.json

@@ -0,0 +1,36 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "dd-flow/run-control-request@1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["schema_id", "request_id", "project_root", "run_id", "action"],
+  "properties": {
+    "schema_id": { "const": "dd-flow/run-control-request@1" },
+    "request_id": { "type": "string", "minLength": 1, "maxLength": 128, "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]*$" },
+    "project_root": { "type": "string", "minLength": 1, "pattern": "^(?:/|[A-Za-z]:[\\\\/]|\\\\\\\\)" },
+    "run_id": { "type": "string", "pattern": "^RUN-[A-Za-z0-9][A-Za-z0-9._-]*$" },
+    "action": { "enum": ["pause", "stop", "resume"] },
+    "wait_ms": { "type": "integer", "minimum": 0, "maximum": 60000, "default": 0 },
+    "grace_ms": { "type": "integer", "minimum": 0, "maximum": 3600000, "default": 30000 },
+    "escalate": { "const": "interrupt" },
+    "force": { "type": "boolean" },
+    "from": { "type": "string", "pattern": "^CTL-[A-Za-z0-9][A-Za-z0-9._-]*$" }
+  },
+  "allOf": [
+    {
+      "if": { "properties": { "action": { "const": "pause" } } },
+      "then": { "not": { "anyOf": [{ "required": ["force"] }, { "required": ["from"] }] } }
+    },
+    {
+      "if": { "properties": { "action": { "const": "stop" } } },
+      "then": { "not": { "anyOf": [{ "required": ["grace_ms"] }, { "required": ["escalate"] }, { "required": ["from"] }] } }
+    },
+    {
+      "if": { "properties": { "action": { "const": "resume" } } },
+      "then": {
+        "required": ["from"],
+        "not": { "anyOf": [{ "required": ["grace_ms"] }, { "required": ["escalate"] }, { "required": ["force"] }] }
+      }
+    }
+  ]
+}

package/dist/services/cleanup.js

@@ -4,10 +4,10 @@ import path from "node:path";
 import { AppError } from "../shared/errors.js";
 import { parseJsonObject } from "../shared/json.js";
 import { ensureReadableFile } from "../storage/database.js";
-import { projectCheckoutRoot, resolveProjectRoot } from "../storage/paths.js";
+import { projectCheckoutRoot, resolveProjectRoot, runtimeStateJsonPath } from "../storage/paths.js";
 import { appendAudit } from "./audit.js";
 import { requireProjectByRoot } from "./projects.js";
-import { protocolRunDiagnostics, readProtocolRuntimeState, requireProtocol } from "./protocols.js";
+import { protocolRunDiagnostics, readProtocolRuntimeState, repairProtocolRuntimeState, requireProtocol } from "./protocols.js";
 import { publicSessionIdentity } from "./session-identity.js";
 const actionKinds = new Set([
     "repair_protocol_state",
@@ -25,7 +25,8 @@ export function cleanupScan(context, input) {
     const protocols = context.db.all("SELECT id, status, stage, state_path FROM protocols WHERE project_id = ? ORDER BY id ASC", [project.id]);
     for (const protocol of protocols) {
         const terminal = ["closed", "cancelled"].includes(protocol.status) || ["closed", "cancelled"].includes(protocol.stage);
-        if (!fs.existsSync(protocol.state_path)) {
+        const requiresMigration = protocol.state_path !== runtimeStateJsonPath(context.ddFlowHome, project.id, protocol.id);
+        if (!fs.existsSync(protocol.state_path) || requiresMigration) {
             pushFinding(findings, actions, "missing_protocol_runtime_state", "warning", {
                 protocol_id: protocol.id,
                 path: protocol.state_path,
@@ -43,6 +44,8 @@ export function cleanupScan(context, input) {
             }
         }
         const fullProtocol = requireProtocol(context, protocol.id, project.id);
+        if (requiresMigration)
+            continue;
         const state = readProtocolRuntimeState(context, fullProtocol).state;
         const diagnostics = protocolRunDiagnostics(context, fullProtocol, state).diagnostics;
         for (const diagnostic of diagnostics.filter((entry) => entry.code === "protocol_run_stage_mismatch")) {
@@ -283,7 +286,7 @@ function applyAction(context, action, reason, force, now) {
         const protocol = requireProtocol(context, action.protocol_id, action.project_id);
         const beforePath = protocol.state_path;
         const existed = fs.existsSync(beforePath);
-        readProtocolRuntimeState(context, protocol);
+        repairProtocolRuntimeState(context, protocol);
         return {
             kind: action.kind,
             changed: !existed || beforePath !== protocol.state_path,

package/dist/services/cli-operation-classifier.js

@@ -101,6 +101,8 @@ function isRouterNative(args) {
         return true;
     if (["engine", "version", "schema"].includes(first))
         return true;
+    if (first === "storage" && args[1] === "migrate-writer")
+        return true;
     return (first === "codex" && args[1] === "hook" && args[2] === "handle")
         || (first === "zcode" && args[1] === "event" && args[2] === "handle")
         || (first === "zcode" && args[1] === "usage" && args[2] === "ingest")
@@ -127,7 +129,7 @@ function isReadOnlyDiagnostic(family, command, action) {
     if (family === "plan")
         return command === "status";
     if (family === "run")
-        return command === "status" || command === "list";
+        return command === "status" || command === "list" || (["drive", "control"].includes(command ?? "") && action === "status");
     if (family === "lane") {
         return (command === "status" ||
             command === "waiters" ||

package/dist/services/code-checks.js

@@ -2,8 +2,10 @@ import crypto from "node:crypto";
 import { spawn, spawnSync } from "node:child_process";
 import fs from "node:fs";
 import path from "node:path";
+import { assertRunMutationAllowed } from "./run-recovery.js";
 import { AppError } from "../shared/errors.js";
 import { confirmManagedProcess, finishManagedProcess, heartbeatManagedProcess, managedProcessStatus, managedOwnerIsAlive, processTreeIsAlive, registerManagedProcess, reservePorts } from "./managed-processes.js";
+import { admitRuntimeCheck, runtimeBudgetForRun } from "./runtime-budget.js";
 const profileRelativePath = path.join(".memory-bank", "spec", "engineering", "code-check-profile.json");
 // Receipt reservation precedes child spawn so an interrupted CLI leaves an
 // auditable attempt. A second caller must not mistake that small handoff window
@@ -79,6 +81,7 @@ export function validateCheckDeclaration(workspaceRoot, check) {
 }
 export async function runCodeChecks(context, input) {
     reconcileUnfinishedChecks(context, input);
+    const generation = assertRunMutationAllowed(context, input.projectId, input.runId);
     for (const check of input.checks)
         validateCheckDeclaration(input.workspaceRoot, check);
     const { hash: profileHash, profile } = readCodeCheckProfile(input.workspaceRoot);
@@ -92,21 +95,35 @@ export async function runCodeChecks(context, input) {
         const startedAt = context.now();
         const inputHash = crypto.createHash("sha256").update(JSON.stringify({ gate: group.gate, command: group.command, environment: String(codeExecutionEnvironment(input.workspaceRoot).DD_FLOW_LOCAL_DATABASE_SUFFIX), required_artifacts: group.requiredArtifacts, ports: [...group.ports].sort() })).digest("hex");
         const epoch = crypto.createHash("sha256").update(`${group.gate}\0${before.fingerprint}`).digest("hex");
-        const reserved = reserveReceipt(context, { ...input, group, before, profileHash, inputHash, epoch, startedAt });
+        const reserved = reserveReceipt(context, { ...input, generation, group, before, profileHash, inputHash, epoch, startedAt });
         const { id, evidenceDir, stdoutPath, stderrPath, receiptPath, completionPath } = reserved;
         let resources = {};
         let result = { exitCode: null, error: "check did not start", aborted: true };
         try {
             fs.mkdirSync(evidenceDir, { recursive: true });
             writeReceipt(receiptPath, receiptFor({ ...reserved, resources, result, status: "running", after: before, mutationPaths: [], artifacts: [] }));
-            const managed = registerManagedProcess(context, { kind: "check", ownerPid: process.pid, ownerId: `check:${id}`, projectId: input.projectId, runId: input.runId, workId: input.workId ?? null, checkId: id, stdoutPath, stderrPath, metadata: { command: group.command } });
+            const budget = runtimeBudgetForRun(context, input.projectId, input.runId);
+            const managed = registerManagedProcess(context, { kind: "check", ownerPid: process.pid, ownerId: `check:${id}`, projectId: input.projectId, runId: input.runId, workId: input.workId ?? null, checkId: id, stdoutPath, stderrPath, metadata: { command: group.command, role: "check", budget } });
             const allocation = await reservePorts(context, { ownerId: managed.owner_id, processId: managed.id, names: group.ports });
             resources = allocation.ports;
-            input.progress?.(`check ${index + 1}/${groups.length} started: ${group.command}`);
+            try {
+                input.progress?.(`check ${index + 1}/${groups.length} preparing: ${group.command}`);
+                assertRunMutationAllowed(context, input.projectId, input.runId, generation);
+            }
+            catch (error) {
+                allocation.release();
+                finishManagedProcess(context, { id: managed.id, leaseToken: managed.lease_token, state: "failed", reason: "dispatch_fenced_before_spawn" });
+                throw error;
+            }
             const stdout = fs.openSync(stdoutPath, "a");
             const stderr = fs.openSync(stderrPath, "a");
             try {
-                result = await runCheck(group.command, input.workspaceRoot, { ...codeExecutionEnvironment(input.workspaceRoot), DD_FLOW_EVIDENCE_DIR: evidenceDir, DD_FLOW_CHECK_COMPLETION_FILE: completionPath, ...portEnvironment(resources) }, stdout, stderr, (elapsed) => input.progress?.(`check ${index + 1}/${groups.length} still running (${elapsed}s): ${group.command}`), (pid) => confirmManagedProcess(context, { id: managed.id, leaseToken: managed.lease_token, pid, processGroupId: process.platform === "win32" ? null : pid }), () => heartbeatManagedProcess(context, { id: managed.id, leaseToken: managed.lease_token }));
+                result = await runCheck(group.command, input.workspaceRoot, { ...codeExecutionEnvironment(input.workspaceRoot), DD_FLOW_EVIDENCE_DIR: evidenceDir, DD_FLOW_CHECK_COMPLETION_FILE: completionPath, ...portEnvironment(resources) }, stdout, stderr, (elapsed) => input.progress?.(`check ${index + 1}/${groups.length} still running (${elapsed}s): ${group.command}`), (pid) => {
+                    confirmManagedProcess(context, { id: managed.id, leaseToken: managed.lease_token, pid, processGroupId: process.platform === "win32" ? null : pid });
+                    assertRunMutationAllowed(context, input.projectId, input.runId, generation);
+                    if (budget)
+                        admitRuntimeCheck(context, managed.id, managed.lease_token);
+                }, () => heartbeatManagedProcess(context, { id: managed.id, leaseToken: managed.lease_token }));
             }
             finally {
                 fs.closeSync(stdout);
@@ -149,15 +166,19 @@ export async function runCodeChecks(context, input) {
  * a dead/terminal process becomes an explicit aborted receipt that can retry.
  */
 export function reconcileUnfinishedChecks(context, input) {
-    const where = input.workId ? "project_id = ? AND run_id = ? AND work_id = ? AND status = 'running'" : "project_id = ? AND run_id = ? AND status = 'running'";
+    let where = input.workId ? "project_id = ? AND run_id = ? AND work_id = ? AND status = 'running'" : "project_id = ? AND run_id = ? AND status = 'running'";
     const params = input.workId ? [input.projectId, input.runId, input.workId] : [input.projectId, input.runId];
+    if (input.receiptId) {
+        where += " AND id = ?";
+        params.push(input.receiptId);
+    }
     const pending = context.db.all(`SELECT id, receipt_path, started_at FROM check_receipts WHERE ${where}`, params);
     if (pending.length === 0)
         return [];
     const processes = managedProcessStatus(context);
     const reconciled = [];
     for (const pendingReceipt of pending) {
-        const process = processes.filter((item) => item.check_id === pendingReceipt.id).at(-1);
+        const process = processes.filter((item) => item.check_id === pendingReceipt.id && item.project_id === input.projectId && item.run_id === input.runId).at(-1);
         const receipt = readReceipt(pendingReceipt.receipt_path);
         const completion = receipt ? readCompletion(receipt.completion_path || path.join(path.dirname(pendingReceipt.receipt_path), "completion.json")) : null;
         // A stale lease after host sleep is not permission to duplicate a living
@@ -189,6 +210,7 @@ export function reconcileUnfinishedChecks(context, input) {
 function reserveReceipt(context, input) {
     context.db.exec("BEGIN IMMEDIATE");
     try {
+        assertRunMutationAllowed(context, input.projectId, input.runId, input.generation);
         const rows = context.db.all("SELECT id FROM check_receipts WHERE work_id IS ? AND run_id = ?", [input.workId ?? null, input.runId]);
         const ordinal = rows.reduce((max, row) => Math.max(max, Number(/\/RCP-(\d+)$/.exec(row.id)?.[1] ?? 0)), 0) + 1;
         const localId = `RCP-${String(ordinal).padStart(3, "0")}`;
@@ -321,7 +343,7 @@ function listFiles(root, current = root) { return fs.readdirSync(current, { with
 export function runCheck(command, cwd, environment, stdout, stderr, heartbeat, onSpawn, renewLease) {
     return new Promise((resolve) => {
         const started = Date.now();
-        const child = spawn("/bin/sh", ["-lc", checkShellScript()], { cwd, env: { ...environment, DD_FLOW_CHECK_COMMAND: command }, stdio: ["ignore", stdout, stderr], detached: process.platform !== "win32" });
+        const child = spawn("/bin/sh", ["-c", checkShellScript()], { cwd, env: { ...environment, DD_FLOW_CHECK_COMMAND: command }, stdio: ["pipe", stdout, stderr], detached: process.platform !== "win32" });
         let error = null;
         let stopping = false;
         let progress;
@@ -343,6 +365,7 @@ export function runCheck(command, cwd, environment, stdout, stderr, heartbeat, o
         // Register handlers before checking pid or calling persistence callbacks.
         // Failure is not completion until the spawned process has actually closed.
         child.once("error", (value) => { error = value.message; });
+        child.stdin.on("error", stop);
         child.once("close", (exitCode, signal) => {
             if (progress)
                 clearInterval(progress);
@@ -363,6 +386,7 @@ export function runCheck(command, cwd, environment, stdout, stderr, heartbeat, o
             stop(cause);
             return;
         }
+        child.stdin.end("admitted\n");
         progress = setInterval(() => {
             try {
                 if (!renewLease()) {
@@ -379,6 +403,8 @@ export function runCheck(command, cwd, environment, stdout, stderr, heartbeat, o
 }
 function checkShellScript() {
     return [
+        // The actual command cannot run before PID ownership and admission persist.
+        'IFS= read -r admission && [ "$admission" = admitted ] || exit 125',
         'completion="${DD_FLOW_CHECK_COMPLETION_FILE:-}"',
         '/bin/sh -lc "$DD_FLOW_CHECK_COMMAND"',
         "code=$?",