fix(daemon): address PR #22 review — 6 issues fixed

Critical:
1. replayAndResume: remove double moderate() call, reuse loop result
2. drainAndRespawn: check workflow still in config before respawn
3. drain: mark in-flight runs as 'interrupted' in DB before clearing

Should fix:
4. crash recovery: dedup runId before re-queuing/re-activating
5. drain timeout: DEFAULT_DRAIN_TIMEOUT_MS > WORKER_SHUTDOWN_TIMEOUT_MS
6. crash-loop protection: max 5 crashes in 60s window, then stop respawn

5 new tests added. All 173 tests pass.

小橘 🍊（NEKO Team）

packages/cli/src/commands/init.ts

@@ -48,9 +48,9 @@ export const cpuUsage = sqliteTable("cpu_usage", {
 });
 `;
 
-const CPU_INDEX_TS = `import { cpus } from "node:os";
+const CPU_INDEX_JS = `import { cpus } from "node:os";
 
-export async function compute(): Promise<unknown> {
+export async function compute() {
   const cpuList = cpus();
 
   let totalIdle = 0;
@@ -135,13 +135,14 @@ export const initCommand = defineCommand({
     }
 
     mkdirSync(join(nerveRoot, "data"), { recursive: true });
+    mkdirSync(join(nerveRoot, "data", "senses"), { recursive: true });
     mkdirSync(join(nerveRoot, "senses", "cpu-usage", "migrations"), { recursive: true });
 
     writeFile(join(nerveRoot, "nerve.yaml"), NERVE_YAML);
     writeFile(join(nerveRoot, "package.json"), PACKAGE_JSON);
     writeFile(join(nerveRoot, ".gitignore"), GITIGNORE);
     writeFile(join(nerveRoot, "senses", "cpu-usage", "schema.ts"), CPU_SCHEMA_TS);
-    writeFile(join(nerveRoot, "senses", "cpu-usage", "index.ts"), CPU_INDEX_TS);
+    writeFile(join(nerveRoot, "senses", "cpu-usage", "index.js"), CPU_INDEX_JS);
     writeFile(
       join(nerveRoot, "senses", "cpu-usage", "migrations", "0001_init.sql"),
       CPU_MIGRATION_SQL,

packages/core/src/index.ts

@@ -8,6 +8,14 @@ export type {
   QueueOverflowConfig,
   WorkflowConfig,
   NerveConfig,
+  CommandEvent,
+  ThreadState,
+  ModerateResult,
+  WorkflowContext,
+  RoleExecuteFn,
+  Role,
+  ModerateFn,
+  WorkflowDefinition,
 } from "./types.js";
 export type { Result } from "./result.js";
 export { ok, err } from "./result.js";

packages/core/src/types.ts

@@ -45,3 +45,58 @@ export type NerveConfig = {
   reflexes: ReflexConfig[];
   workflows: Record<string, WorkflowConfig> | null;
 };
+
+// ---------------------------------------------------------------------------
+// Workflow Engine types (RFC-002)
+// ---------------------------------------------------------------------------
+
+/** A single event in the command event stream that drives a workflow thread. */
+export type CommandEvent = {
+  type: string;
+  [key: string]: unknown;
+};
+
+/** Accumulated state of a running thread — the event history for moderate(). */
+export type ThreadState = {
+  runId: string;
+  /** All events so far, including the initial thread_start event. */
+  events: CommandEvent[];
+};
+
+/** The result of moderate() — which role to hand to next, and what prompt to pass. */
+export type ModerateResult = {
+  role: string;
+  prompt: unknown;
+};
+
+/** Context injected into every role execute() call. */
+export type WorkflowContext = {
+  runId: string;
+  workflowName: string;
+  /** Emit a log message back to the parent process. */
+  log: (message: string) => void;
+};
+
+/**
+ * A role's execute function. Has side effects (API calls, file I/O, etc.).
+ * Returns a CommandEvent that is fed back into moderate().
+ */
+export type RoleExecuteFn = (prompt: unknown, ctx: WorkflowContext) => Promise<CommandEvent>;
+
+/** A role in a workflow — a named unit of execution with side effects. */
+export type Role = {
+  execute: RoleExecuteFn;
+};
+
+/**
+ * The moderator function — pure, no side effects.
+ * Decides which role to pass control to next.
+ * Returns null to signal thread completion.
+ */
+export type ModerateFn = (thread: ThreadState, event: CommandEvent) => ModerateResult | null;
+
+/** The complete definition of a workflow, as authored by users. */
+export type WorkflowDefinition = {
+  roles: Record<string, Role>;
+  moderate: ModerateFn;
+};

packages/daemon/src/__tests__/crash-recovery.test.ts

@@ -0,0 +1,429 @@
+/**
+ * Phase 3 — Worker crash recovery tests.
+ *
+ * Verifies that WorkflowManager correctly:
+ * - Marks in-flight threads as "crashed" in the DB when a worker exits unexpectedly
+ * - Respawns the worker after a crash
+ * - Resumes "started" threads from persisted event history (resume-thread IPC)
+ * - Re-queues "queued" threads so they are dispatched on the new worker
+ */
+
+import { EventEmitter } from "node:events";
+
+import type { NerveConfig, WorkflowConfig } from "@uncaged/nerve-core";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+type MockChild = EventEmitter & {
+  send: ReturnType<typeof vi.fn>;
+  kill: ReturnType<typeof vi.fn>;
+  connected: boolean;
+  exitCode: number | null;
+  pid: number;
+};
+
+const mockChildren: MockChild[] = [];
+
+function makeMockChild(pid = 1): MockChild {
+  const child = new EventEmitter() as MockChild;
+  child.connected = true;
+  child.exitCode = null;
+  child.pid = pid;
+  child.send = vi.fn((msg: unknown) => {
+    if (
+      msg !== null &&
+      typeof msg === "object" &&
+      (msg as Record<string, unknown>).type === "shutdown"
+    ) {
+      setImmediate(() => {
+        child.exitCode = 0;
+        child.connected = false;
+        child.emit("exit", 0, null);
+      });
+    }
+  });
+  child.kill = vi.fn((_signal?: string) => {
+    child.exitCode = 1;
+    child.connected = false;
+    child.emit("exit", null, _signal ?? "SIGKILL");
+  });
+  return child;
+}
+
+vi.mock("node:child_process", () => ({
+  fork: vi.fn((_script: string, _args: string[], _opts: unknown) => {
+    const child = makeMockChild(mockChildren.length + 1);
+    mockChildren.push(child);
+    return child;
+  }),
+}));
+
+const { createWorkflowManager } = await import("../workflow-manager.js");
+
+function makeConfig(workflows: Record<string, WorkflowConfig> = {}): NerveConfig {
+  return {
+    senses: {},
+    reflexes: [],
+    workflows,
+  };
+}
+
+function makeLogStore(
+  activeRuns: Array<{
+    runId: string;
+    workflow: string;
+    status: "queued" | "started";
+    ts: number;
+  }> = [],
+) {
+  const store = {
+    append: vi.fn(),
+    query: vi.fn(() => []),
+    getMeta: vi.fn(() => null),
+    setMeta: vi.fn(),
+    upsertWorkflowRun: vi.fn(),
+    appendWithWorkflowUpdate: vi.fn(),
+    getWorkflowRun: vi.fn(() => null),
+    getActiveWorkflowRuns: vi.fn((_workflowName?: string) => {
+      if (_workflowName !== undefined) {
+        return activeRuns.filter((r) => r.workflow === _workflowName);
+      }
+      return activeRuns;
+    }),
+    getTriggerPayload: vi.fn(() => ({ value: 42 })),
+    getThreadEvents: vi.fn(() => [{ type: "thread_start", triggerPayload: {} }]),
+    close: vi.fn(),
+  };
+  return store;
+}
+
+describe("WorkflowManager — crash recovery (Phase 3)", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+    vi.clearAllMocks();
+  });
+
+  describe("worker crash marks active threads as crashed", () => {
+    it("logs 'crashed' status for each active thread when worker exits unexpectedly", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-wf": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", { n: 1 });
+      mgr.startWorkflow("my-wf", { n: 2 });
+      expect(mgr.activeCount("my-wf")).toBe(2);
+
+      // Simulate unexpected exit (not shutdown)
+      const child = mockChildren[0];
+      child.exitCode = 1;
+      child.connected = false;
+      child.emit("exit", 1, null);
+
+      const crashedCalls = logStore.upsertWorkflowRun.mock.calls.filter(
+        ([entry]: [{ type: string }]) => entry.type === "crashed",
+      );
+      expect(crashedCalls).toHaveLength(2);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("clears active count after crash", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-wf": { concurrency: 3, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", {});
+      expect(mgr.activeCount("my-wf")).toBe(2);
+
+      const child = mockChildren[0];
+      child.exitCode = 1;
+      child.connected = false;
+      child.emit("exit", 1, null);
+
+      expect(mgr.activeCount("my-wf")).toBe(0);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("worker crash triggers respawn", () => {
+    it("spawns a new worker after crash", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", {});
+      expect(mockChildren).toHaveLength(1);
+
+      const child = mockChildren[0];
+      child.exitCode = 1;
+      child.connected = false;
+      child.emit("exit", 1, null);
+
+      // setImmediate to allow respawn
+      await vi.runAllTimersAsync();
+
+      expect(mockChildren).toHaveLength(2);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("sends resume-thread for 'started' runs from DB after respawn", async () => {
+      const activeRuns = [
+        { runId: "run-started-1", workflow: "my-wf", status: "started" as const, ts: 1000 },
+      ];
+      const logStore = makeLogStore(activeRuns);
+      logStore.getThreadEvents.mockReturnValue([
+        { type: "thread_start", triggerPayload: {} },
+        { type: "scan_complete", items: ["a"] },
+      ]);
+      logStore.getTriggerPayload.mockReturnValue({ trigger: "initial" });
+
+      const config = makeConfig({
+        "my-wf": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", {});
+      const firstChild = mockChildren[0];
+      firstChild.exitCode = 1;
+      firstChild.connected = false;
+      firstChild.emit("exit", 1, null);
+
+      await vi.runAllTimersAsync();
+
+      // New worker should have been spawned
+      const secondChild = mockChildren[1];
+      expect(secondChild).toBeDefined();
+
+      // resume-thread should have been sent
+      const resumeCalls = (secondChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
+        ([msg]: [unknown]) =>
+          msg !== null &&
+          typeof msg === "object" &&
+          (msg as Record<string, unknown>).type === "resume-thread",
+      );
+      expect(resumeCalls).toHaveLength(1);
+      expect(resumeCalls[0][0]).toMatchObject({
+        type: "resume-thread",
+        runId: "run-started-1",
+        triggerPayload: { trigger: "initial" },
+      });
+      expect(Array.isArray((resumeCalls[0][0] as Record<string, unknown>).events)).toBe(true);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("re-queues 'queued' runs from DB after respawn", async () => {
+      const activeRuns = [
+        { runId: "run-queued-1", workflow: "my-wf", status: "queued" as const, ts: 900 },
+      ];
+      const logStore = makeLogStore(activeRuns);
+      logStore.getTriggerPayload.mockReturnValue({ queued: "payload" });
+
+      const config = makeConfig({
+        "my-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      // Start one thread to fill the concurrency slot (so queued run stays queued on respawn)
+      mgr.startWorkflow("my-wf", {});
+      const firstChild = mockChildren[0];
+      firstChild.exitCode = 1;
+      firstChild.connected = false;
+      firstChild.emit("exit", 1, null);
+
+      await vi.runAllTimersAsync();
+
+      // After respawn, the queue should contain the recovered run
+      expect(mgr.queueLength("my-wf")).toBeGreaterThanOrEqual(1);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("command events are persisted (for crash recovery replay)", () => {
+    it("persists thread_command_event when worker sends thread-command-event IPC", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", { x: 1 });
+
+      const child = mockChildren[0];
+      const startCall = (child.send as ReturnType<typeof vi.fn>).mock.calls[0];
+      const runId = (startCall[0] as Record<string, unknown>).runId as string;
+
+      // Simulate worker sending a command event back
+      child.emit("message", {
+        type: "thread-command-event",
+        runId,
+        event: { type: "scan_complete", items: ["a", "b"] },
+      });
+
+      const appendCalls = logStore.append.mock.calls.filter(
+        ([entry]: [{ type: string }]) => entry.type === "thread_command_event",
+      );
+      expect(appendCalls).toHaveLength(1);
+      expect(appendCalls[0][0]).toMatchObject({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: runId,
+      });
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("triggerPayload is persisted in 'started' log entry", () => {
+    it("stores triggerPayload in the payload field of the started log entry", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      const payload = { task: "build-docker", repo: "myrepo" };
+      mgr.startWorkflow("my-wf", payload);
+
+      const startedCall = logStore.upsertWorkflowRun.mock.calls.find(
+        ([entry]: [{ type: string }]) => entry.type === "started",
+      );
+      expect(startedCall).toBeDefined();
+      const logEntry = startedCall?.[0] as { payload: string | null };
+      expect(logEntry.payload).not.toBeNull();
+      const parsed = JSON.parse(logEntry.payload as string) as Record<string, unknown>;
+      expect(parsed.triggerPayload).toMatchObject(payload);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("runId deduplication in crash recovery", () => {
+    it("does not push duplicate runIds into the queue during crash recovery", async () => {
+      const activeRuns = [
+        { runId: "run-queued-dup", workflow: "my-wf", status: "queued" as const, ts: 900 },
+      ];
+      const logStore = makeLogStore(activeRuns);
+      logStore.getTriggerPayload.mockReturnValue({ q: 1 });
+
+      const config = makeConfig({
+        "my-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      // Start one thread to fill the concurrency slot
+      mgr.startWorkflow("my-wf", {});
+      const firstChild = mockChildren[0];
+
+      // Crash once → respawn → crash again → second respawn
+      firstChild.exitCode = 1;
+      firstChild.connected = false;
+      firstChild.emit("exit", 1, null);
+      await vi.runAllTimersAsync();
+
+      const secondChild = mockChildren[1];
+      secondChild.exitCode = 1;
+      secondChild.connected = false;
+      secondChild.emit("exit", 1, null);
+      await vi.runAllTimersAsync();
+
+      // The recovered queued run should appear at most once in the queue
+      expect(mgr.queueLength("my-wf")).toBeLessThanOrEqual(1);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("does not add duplicate active runIds during crash recovery", async () => {
+      const activeRuns = [
+        { runId: "run-started-dup", workflow: "my-wf", status: "started" as const, ts: 1000 },
+      ];
+      const logStore = makeLogStore(activeRuns);
+      logStore.getThreadEvents.mockReturnValue([{ type: "thread_start", triggerPayload: {} }]);
+      logStore.getTriggerPayload.mockReturnValue({ s: 1 });
+
+      const config = makeConfig({
+        "my-wf": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-wf", {});
+      const firstChild = mockChildren[0];
+      firstChild.exitCode = 1;
+      firstChild.connected = false;
+      firstChild.emit("exit", 1, null);
+      await vi.runAllTimersAsync();
+
+      const secondChild = mockChildren[1];
+      secondChild.exitCode = 1;
+      secondChild.connected = false;
+      secondChild.emit("exit", 1, null);
+      await vi.runAllTimersAsync();
+
+      // The active set should not double-count the recovered run
+      expect(mgr.activeCount("my-wf")).toBeLessThanOrEqual(1);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("crash-loop backoff", () => {
+    it("stops respawning after exceeding max crashes in the window", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "crash-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("crash-wf", {});
+
+      // Crash the worker 6 times in rapid succession (within CRASH_WINDOW_MS = 60s)
+      for (let i = 0; i < 6; i++) {
+        const child = mockChildren[mockChildren.length - 1];
+        child.exitCode = 1;
+        child.connected = false;
+        child.emit("exit", 1, null);
+        await vi.runAllTimersAsync();
+      }
+
+      // After 6 crashes, no new worker should be spawned
+      // The 1st crash spawns child[1], ..., 5th crash spawns child[5], 6th should NOT spawn
+      expect(mockChildren.length).toBeLessThanOrEqual(6);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+});

packages/daemon/src/__tests__/file-watcher-workflow.test.ts

@@ -0,0 +1,119 @@
+/**
+ * Phase 3 — FileWatcher workflow change detection tests.
+ *
+ * Verifies that file-watcher.ts detects .ts file changes under workflows/.
+ */
+
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { afterEach, describe, expect, it } from "vitest";
+
+import { createFileWatcher } from "../file-watcher.js";
+import type { FileChange, FileWatcher } from "../file-watcher.js";
+
+function makeTempNerveRoot(): string {
+  const dir = mkdtempSync(join(tmpdir(), "nerve-fw-wf-test-"));
+  mkdirSync(join(dir, "workflows", "my-workflow"), { recursive: true });
+  writeFileSync(join(dir, "nerve.yaml"), "senses: {}\nreflexes: []\n");
+  writeFileSync(
+    join(dir, "workflows", "my-workflow", "index.ts"),
+    "export default { roles: {}, moderate: () => null };",
+  );
+  return dir;
+}
+
+async function waitFor(
+  predicate: () => boolean,
+  timeoutMs: number,
+  intervalMs = 50,
+): Promise<void> {
+  return new Promise((resolve, reject) => {
+    const timer = setTimeout(
+      () => reject(new Error(`waitFor timed out after ${timeoutMs}ms`)),
+      timeoutMs,
+    );
+    const check = setInterval(() => {
+      if (predicate()) {
+        clearTimeout(timer);
+        clearInterval(check);
+        resolve();
+      }
+    }, intervalMs);
+  });
+}
+
+describe("createFileWatcher — workflow file changes (Phase 3)", () => {
+  let watcher: FileWatcher | null = null;
+
+  afterEach(() => {
+    if (watcher !== null) {
+      watcher.close();
+      watcher = null;
+    }
+  });
+
+  it("detects workflow .ts file changes and emits kind=workflow", async () => {
+    const root = makeTempNerveRoot();
+    const changes: FileChange[] = [];
+
+    watcher = createFileWatcher(root, (change) => changes.push(change), 50);
+
+    await new Promise((r) => setTimeout(r, 100));
+    writeFileSync(
+      join(root, "workflows", "my-workflow", "index.ts"),
+      "export default { roles: {}, moderate: () => null }; // updated",
+    );
+
+    await waitFor(() => changes.some((c) => c.kind === "workflow"), 3000);
+
+    const wfChanges = changes.filter((c) => c.kind === "workflow");
+    expect(wfChanges.length).toBeGreaterThanOrEqual(1);
+    const wfChange = wfChanges[0] as { workflowName: string; filePath: string };
+    expect(wfChange.workflowName).toBe("my-workflow");
+  }, 10_000);
+
+  it("does NOT emit workflow change for nerve.yaml", async () => {
+    const root = makeTempNerveRoot();
+    const changes: FileChange[] = [];
+
+    watcher = createFileWatcher(root, (change) => changes.push(change), 50);
+
+    await new Promise((r) => setTimeout(r, 100));
+    writeFileSync(join(root, "nerve.yaml"), "senses: {}\nreflexes: []\n# changed\n");
+
+    await waitFor(() => changes.some((c) => c.kind === "config"), 3000);
+
+    const wfChanges = changes.filter((c) => c.kind === "workflow");
+    expect(wfChanges).toHaveLength(0);
+  }, 10_000);
+
+  it("debounces rapid workflow file changes", async () => {
+    const root = makeTempNerveRoot();
+    const changes: FileChange[] = [];
+
+    watcher = createFileWatcher(root, (change) => changes.push(change), 200);
+
+    await new Promise((r) => setTimeout(r, 100));
+
+    for (let i = 0; i < 5; i++) {
+      writeFileSync(
+        join(root, "workflows", "my-workflow", "index.ts"),
+        `export default {}; // v${i}`,
+      );
+    }
+
+    await waitFor(() => changes.some((c) => c.kind === "workflow"), 3000);
+    // Allow debounce window to pass
+    await new Promise((r) => setTimeout(r, 300));
+
+    const wfChanges = changes.filter((c) => c.kind === "workflow");
+    expect(wfChanges.length).toBe(1);
+  }, 10_000);
+
+  it("cleans up temp dir after test", () => {
+    const root = makeTempNerveRoot();
+    rmSync(root, { recursive: true, force: true });
+  });
+});

packages/daemon/src/__tests__/hot-reload.test.ts

@@ -0,0 +1,353 @@
+/**
+ * Phase 3 — Hot reload tests.
+ *
+ * Verifies that:
+ * - drainAndRespawn() sends shutdown, waits for exit, then respawns the worker
+ * - Kernel dispatches handleWorkflowFileChange when file-watcher emits a workflow change
+ * - Kernel logs a workflow_reload system event on hot reload
+ * - drainAndRespawn on a non-existent worker is a no-op
+ * - drainAndRespawn after the drain sends a fresh worker (not crash-recovery)
+ */
+
+import { EventEmitter } from "node:events";
+
+import type { NerveConfig, WorkflowConfig } from "@uncaged/nerve-core";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+type MockChild = EventEmitter & {
+  send: ReturnType<typeof vi.fn>;
+  kill: ReturnType<typeof vi.fn>;
+  connected: boolean;
+  exitCode: number | null;
+  pid: number;
+};
+
+const mockChildren: MockChild[] = [];
+
+function makeMockChild(pid = 1): MockChild {
+  const child = new EventEmitter() as MockChild;
+  child.connected = true;
+  child.exitCode = null;
+  child.pid = pid;
+  child.send = vi.fn((msg: unknown) => {
+    if (
+      msg !== null &&
+      typeof msg === "object" &&
+      (msg as Record<string, unknown>).type === "shutdown"
+    ) {
+      setImmediate(() => {
+        child.exitCode = 0;
+        child.connected = false;
+        child.emit("exit", 0, null);
+      });
+    }
+  });
+  child.kill = vi.fn((_signal?: string) => {
+    child.exitCode = 1;
+    child.connected = false;
+    child.emit("exit", null, _signal ?? "SIGKILL");
+  });
+  return child;
+}
+
+vi.mock("node:child_process", () => ({
+  fork: vi.fn((_script: string, _args: string[], _opts: unknown) => {
+    const child = makeMockChild(mockChildren.length + 1);
+    mockChildren.push(child);
+    return child;
+  }),
+}));
+
+const { createWorkflowManager } = await import("../workflow-manager.js");
+const { createKernel } = await import("../kernel.js");
+
+function makeWfConfig(workflows: Record<string, WorkflowConfig> = {}): NerveConfig {
+  return { senses: {}, reflexes: [], workflows };
+}
+
+function makeLogStore() {
+  return {
+    append: vi.fn(),
+    query: vi.fn(() => []),
+    getMeta: vi.fn(() => null),
+    setMeta: vi.fn(),
+    upsertWorkflowRun: vi.fn(),
+    appendWithWorkflowUpdate: vi.fn(),
+    getWorkflowRun: vi.fn(() => null),
+    getActiveWorkflowRuns: vi.fn(() => []),
+    getTriggerPayload: vi.fn(() => null),
+    getThreadEvents: vi.fn(() => []),
+    close: vi.fn(),
+  };
+}
+
+describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+    vi.clearAllMocks();
+  });
+
+  it("drainAndRespawn does NOT respawn when workflow is removed from config", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", {});
+    expect(mockChildren).toHaveLength(1);
+
+    // Remove workflow from config before drain completes
+    mgr.updateConfig(makeWfConfig({}));
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    // No new worker should have been spawned (workflow was removed)
+    expect(mockChildren).toHaveLength(1);
+  });
+
+  it("drainAndRespawn marks in-flight runs as interrupted in DB", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 2, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", { n: 1 });
+    mgr.startWorkflow("my-wf", { n: 2 });
+    expect(mgr.activeCount("my-wf")).toBe(2);
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    const interruptedCalls = logStore.upsertWorkflowRun.mock.calls.filter(
+      ([entry]: [{ type: string }]) => entry.type === "interrupted",
+    );
+    expect(interruptedCalls).toHaveLength(2);
+
+    const stopPromise = mgr.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+
+  it("drainAndRespawn on an unknown workflow (no worker) resolves immediately", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    // No thread started — no worker spawned
+    await expect(mgr.drainAndRespawn("my-wf")).resolves.toBeUndefined();
+  });
+
+  it("drainAndRespawn sends shutdown to existing worker and waits for exit", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", {});
+    expect(mockChildren).toHaveLength(1);
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    const firstChild = mockChildren[0];
+    expect(firstChild.send).toHaveBeenCalledWith(expect.objectContaining({ type: "shutdown" }));
+  });
+
+  it("drainAndRespawn spawns a fresh worker after the old one exits", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", {});
+    expect(mockChildren).toHaveLength(1);
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    // A new worker should have been spawned (not crash-recovery, just fresh)
+    expect(mockChildren).toHaveLength(2);
+  });
+
+  it("fresh worker after drainAndRespawn does NOT receive resume-thread messages", async () => {
+    const logStore = makeLogStore();
+    // Even if there are active runs in DB, after drain the worker should NOT get resume
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", {});
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    const newChild = mockChildren[1];
+    const resumeCalls = (newChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
+      ([msg]: [unknown]) =>
+        msg !== null &&
+        typeof msg === "object" &&
+        (msg as Record<string, unknown>).type === "resume-thread",
+    );
+    expect(resumeCalls).toHaveLength(0);
+
+    const stopPromise = mgr.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+
+  it("new threads can be started on the fresh worker after drainAndRespawn", async () => {
+    const logStore = makeLogStore();
+    const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
+    const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+    mgr.startWorkflow("my-wf", { first: true });
+
+    const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    // Start a new thread on the fresh worker
+    mgr.startWorkflow("my-wf", { second: true });
+
+    const newChild = mockChildren[1];
+    const startCalls = (newChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
+      ([msg]: [unknown]) =>
+        msg !== null &&
+        typeof msg === "object" &&
+        (msg as Record<string, unknown>).type === "start-thread",
+    );
+    expect(startCalls).toHaveLength(1);
+
+    const stopPromise = mgr.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+});
+
+describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+  });
+
+  afterEach(async () => {
+    vi.useRealTimers();
+    vi.clearAllMocks();
+  });
+
+  it("handleWorkflowFileChange logs workflow_reload system event", async () => {
+    const logStore = makeLogStore();
+    const config: NerveConfig = {
+      senses: {},
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null }],
+      workflows: { "my-wf": { concurrency: 1, overflow: "drop" } },
+    };
+
+    const kernel = createKernel(config, "/tmp/nerve-hot-reload-test", {
+      workerScript: "fake-worker.js",
+      logStore,
+    });
+
+    // Trigger a workflow thread so a worker is spawned
+    kernel.workflowManager.startWorkflow("my-wf", {});
+
+    // Manually call drainAndRespawn (simulating what kernel does on workflow file change)
+    const drainPromise = kernel.workflowManager.drainAndRespawn("my-wf", 1000);
+    await vi.runAllTimersAsync();
+    await drainPromise;
+
+    // Kernel's handleWorkflowFileChange should log a workflow_reload event
+    // We test this via the kernel itself
+    const appendCalls = logStore.append.mock.calls;
+    const startCall = appendCalls.find(([e]: [{ type: string }]) => e.type === "start");
+    expect(startCall).toBeDefined();
+
+    const stopPromise = kernel.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+
+  it("reloadConfig drains worker for removed workflows", async () => {
+    const logStore = makeLogStore();
+    const initialConfig: NerveConfig = {
+      senses: {},
+      reflexes: [{ kind: "workflow", workflow: "old-wf", on: null }],
+      workflows: { "old-wf": { concurrency: 1, overflow: "drop" } },
+    };
+
+    const kernel = createKernel(initialConfig, "/tmp/nerve-hot-reload-test", {
+      workerScript: "fake-worker.js",
+      logStore,
+    });
+
+    // Spawn a worker for old-wf
+    kernel.workflowManager.startWorkflow("old-wf", {});
+    expect(mockChildren).toHaveLength(1);
+
+    // Reload config without old-wf
+    const newConfig: NerveConfig = {
+      senses: {},
+      reflexes: [],
+      workflows: null,
+    };
+    kernel.reloadConfig(newConfig);
+
+    await vi.runAllTimersAsync();
+
+    // The old worker should have received a shutdown (drain)
+    expect(mockChildren[0].send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "shutdown" }),
+    );
+
+    const stopPromise = kernel.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+
+  it("reloadConfig updates concurrency/overflow without restarting worker", async () => {
+    const logStore = makeLogStore();
+    const initialConfig: NerveConfig = {
+      senses: {},
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null }],
+      workflows: { "my-wf": { concurrency: 1, overflow: "drop" } },
+    };
+
+    const kernel = createKernel(initialConfig, "/tmp/nerve-hot-reload-test", {
+      workerScript: "fake-worker.js",
+      logStore,
+    });
+
+    kernel.workflowManager.startWorkflow("my-wf", {});
+    const workersBefore = mockChildren.length;
+
+    // Reload with updated concurrency — should NOT spawn a new workflow worker
+    const newConfig: NerveConfig = {
+      senses: {},
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null }],
+      workflows: { "my-wf": { concurrency: 5, overflow: "queue", maxQueue: 50 } },
+    };
+    kernel.reloadConfig(newConfig);
+
+    // No extra workflow worker spawn (the config update is in-place)
+    // The worker count may increase if senses change, but the workflow worker should not be respawned
+    expect(mockChildren).toHaveLength(workersBefore);
+
+    // After reload, the new concurrency should be respected
+    expect(kernel.workflowManager.activeCount("my-wf")).toBe(1);
+
+    // Can now start up to 5 concurrent threads (previously only 1)
+    kernel.workflowManager.startWorkflow("my-wf", { n: 2 });
+    kernel.workflowManager.startWorkflow("my-wf", { n: 3 });
+    expect(kernel.workflowManager.activeCount("my-wf")).toBe(3);
+
+    const stopPromise = kernel.stop();
+    await vi.runAllTimersAsync();
+    await stopPromise;
+  });
+});

packages/daemon/src/__tests__/kernel-workflow-integration.test.ts

@@ -0,0 +1,420 @@
+/**
+ * Integration tests for Kernel + WorkflowManager integration.
+ *
+ * Verifies that sense signals trigger workflow runs via workflow reflexes,
+ * that workflow events are logged, that reloadConfig handles workflow changes,
+ * and that graceful shutdown stops workflow workers.
+ *
+ * Uses mocked child_process.fork to avoid real subprocesses.
+ */
+
+import { EventEmitter } from "node:events";
+
+import type { NerveConfig } from "@uncaged/nerve-core";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+// ---------------------------------------------------------------------------
+// Mock child_process.fork before importing kernel
+// ---------------------------------------------------------------------------
+
+type MockChild = EventEmitter & {
+  send: ReturnType<typeof vi.fn>;
+  kill: ReturnType<typeof vi.fn>;
+  connected: boolean;
+  exitCode: number | null;
+  pid: number;
+};
+
+const mockChildren: MockChild[] = [];
+
+function makeMockChild(pid = 1): MockChild {
+  const child = new EventEmitter() as MockChild;
+  child.connected = true;
+  child.exitCode = null;
+  child.pid = pid;
+  child.send = vi.fn((msg: unknown) => {
+    if (
+      msg !== null &&
+      typeof msg === "object" &&
+      (msg as Record<string, unknown>).type === "shutdown"
+    ) {
+      setImmediate(() => {
+        child.exitCode = 0;
+        child.connected = false;
+        child.emit("exit", 0, null);
+      });
+    }
+  });
+  child.kill = vi.fn((_signal?: string) => {
+    child.exitCode = 1;
+    child.connected = false;
+    child.emit("exit", null, _signal ?? "SIGKILL");
+  });
+  return child;
+}
+
+vi.mock("node:child_process", () => ({
+  fork: vi.fn((_script: string, _args: string[], _opts: unknown) => {
+    const child = makeMockChild(mockChildren.length + 1);
+    mockChildren.push(child);
+    return child;
+  }),
+}));
+
+// Import after mock is set up
+const { createKernel } = await import("../kernel.js");
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function makeLogStore() {
+  return {
+    append: vi.fn(),
+    query: vi.fn(() => []),
+    getMeta: vi.fn(() => null),
+    setMeta: vi.fn(),
+    upsertWorkflowRun: vi.fn(),
+    appendWithWorkflowUpdate: vi.fn(),
+    getWorkflowRun: vi.fn(() => null),
+    getActiveWorkflowRuns: vi.fn(() => []),
+    getTriggerPayload: vi.fn(() => null),
+    getThreadEvents: vi.fn(() => []),
+    close: vi.fn(),
+  };
+}
+
+function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
+  return {
+    senses: {
+      "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+    },
+    reflexes: [],
+    workflows: null,
+    ...overrides,
+  };
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+describe("kernel + workflowManager integration", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+  });
+
+  afterEach(async () => {
+    vi.useRealTimers();
+    vi.clearAllMocks();
+  });
+
+  describe("sense signal triggers workflow via reflex", () => {
+    it("calls workflowManager.startWorkflow when a sense signal fires on a workflow reflex", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "my-workflow", on: ["cpu-usage"] }],
+        workflows: { "my-workflow": { concurrency: 2, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      // Emit a signal from "cpu-usage" on the bus
+      const { createSignalBus } = await import("../signal-bus.js");
+      void createSignalBus; // ensure import resolves
+      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: { value: 80 }, ts: Date.now() });
+
+      // The workflow worker should be spawned (one for the sense group, one for workflow)
+      // The sense group worker is mockChildren[0]; the workflow worker is mockChildren[1]
+      // We need to check that a start-thread message was sent to the workflow worker
+      const workflowWorker = mockChildren.find((c) =>
+        (c.send as ReturnType<typeof vi.fn>).mock.calls.some(
+          ([msg]: [unknown]) =>
+            msg !== null &&
+            typeof msg === "object" &&
+            (msg as Record<string, unknown>).type === "start-thread",
+        ),
+      );
+      expect(workflowWorker).toBeDefined();
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("passes the signal payload as triggerPayload to the workflow", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "alert-workflow", on: ["cpu-usage"] }],
+        workflows: { "alert-workflow": { concurrency: 1, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      const payload = { level: "critical", value: 99 };
+      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload, ts: Date.now() });
+
+      // Find the start-thread call and verify triggerPayload
+      const startThreadCall = mockChildren
+        .flatMap((c) => (c.send as ReturnType<typeof vi.fn>).mock.calls as [unknown][])
+        .find(
+          ([msg]) =>
+            msg !== null &&
+            typeof msg === "object" &&
+            (msg as Record<string, unknown>).type === "start-thread",
+        );
+
+      expect(startThreadCall).toBeDefined();
+      expect(startThreadCall?.[0]).toMatchObject({
+        type: "start-thread",
+        workflow: "alert-workflow",
+        triggerPayload: payload,
+      });
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("does not trigger workflow when signal senseId is not in 'on' list", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+          "disk-io": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "my-workflow", on: ["disk-io"] }],
+        workflows: { "my-workflow": { concurrency: 1, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      // Emit signal from cpu-usage — NOT in the workflow's "on" list
+      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: 50, ts: Date.now() });
+
+      // No workflow worker should have been spawned (only the sense group worker)
+      const workflowWorkerSpawned = mockChildren.some((c) =>
+        (c.send as ReturnType<typeof vi.fn>).mock.calls.some(
+          ([msg]: [unknown]) =>
+            msg !== null &&
+            typeof msg === "object" &&
+            (msg as Record<string, unknown>).type === "start-thread",
+        ),
+      );
+      expect(workflowWorkerSpawned).toBe(false);
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("workflow events are logged", () => {
+    it("logs a 'started' event when workflow thread is triggered", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "log-test-workflow", on: ["cpu-usage"] }],
+        workflows: { "log-test-workflow": { concurrency: 2, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: null, ts: Date.now() });
+
+      expect(logStore.upsertWorkflowRun).toHaveBeenCalledWith(
+        expect.objectContaining({ source: "workflow", type: "started" }),
+        expect.objectContaining({ workflow: "log-test-workflow", status: "started" }),
+      );
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("reloadConfig handles workflow changes", () => {
+    it("new workflow reflexes are active after reloadConfig", async () => {
+      const logStore = makeLogStore();
+      const initialConfig = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [],
+        workflows: null,
+      });
+
+      const kernel = createKernel(initialConfig, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      // Reload with a workflow reflex added
+      const newConfig: NerveConfig = {
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "new-workflow", on: ["cpu-usage"] }],
+        workflows: { "new-workflow": { concurrency: 1, overflow: "drop" } },
+      };
+      kernel.reloadConfig(newConfig);
+
+      // Now emit a signal — should trigger the new workflow
+      kernel.bus.emit({ id: 2, senseId: "cpu-usage", payload: "reload-test", ts: Date.now() });
+
+      const startThreadCall = mockChildren
+        .flatMap((c) => (c.send as ReturnType<typeof vi.fn>).mock.calls as [unknown][])
+        .find(
+          ([msg]) =>
+            msg !== null &&
+            typeof msg === "object" &&
+            (msg as Record<string, unknown>).type === "start-thread" &&
+            (msg as Record<string, unknown>).workflow === "new-workflow",
+        );
+
+      expect(startThreadCall).toBeDefined();
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+
+    it("old workflow reflexes are removed after reloadConfig", async () => {
+      const logStore = makeLogStore();
+      const initialConfig = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "old-workflow", on: ["cpu-usage"] }],
+        workflows: { "old-workflow": { concurrency: 1, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(initialConfig, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      // Reload with the workflow reflex removed
+      const newConfig: NerveConfig = {
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [],
+        workflows: null,
+      };
+      kernel.reloadConfig(newConfig);
+
+      // Clear send history
+      for (const c of mockChildren) {
+        (c.send as ReturnType<typeof vi.fn>).mockClear();
+      }
+
+      // Emit a signal — old-workflow should NOT be triggered
+      kernel.bus.emit({ id: 3, senseId: "cpu-usage", payload: "after-reload", ts: Date.now() });
+
+      const startThreadCall = mockChildren
+        .flatMap((c) => (c.send as ReturnType<typeof vi.fn>).mock.calls as [unknown][])
+        .find(
+          ([msg]) =>
+            msg !== null &&
+            typeof msg === "object" &&
+            (msg as Record<string, unknown>).type === "start-thread",
+        );
+
+      expect(startThreadCall).toBeUndefined();
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+
+  describe("graceful shutdown stops workflow workers", () => {
+    it("stop() resolves after workflow workers exit", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "shutdown-test", on: ["cpu-usage"] }],
+        workflows: { "shutdown-test": { concurrency: 1, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      // Trigger a workflow so a worker is spawned
+      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: null, ts: Date.now() });
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await expect(stopPromise).resolves.toBeUndefined();
+    });
+
+    it("workflowManager is exposed on kernel", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        workflows: { "my-wf": { concurrency: 1, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      expect(kernel.workflowManager).toBeDefined();
+      expect(typeof kernel.workflowManager.startWorkflow).toBe("function");
+      expect(typeof kernel.workflowManager.activeCount).toBe("function");
+      expect(typeof kernel.workflowManager.stop).toBe("function");
+
+      kernel.stop().catch(() => {});
+    });
+
+    it("getHealth includes activeWorkflows count", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        senses: {
+          "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
+        },
+        reflexes: [{ kind: "workflow", workflow: "health-wf", on: ["cpu-usage"] }],
+        workflows: { "health-wf": { concurrency: 2, overflow: "drop" } },
+      });
+
+      const kernel = createKernel(config, "/tmp/nerve-test", {
+        workerScript: "fake-worker.js",
+        logStore,
+      });
+
+      const health = kernel.getHealth();
+      expect(health).toHaveProperty("activeWorkflows");
+      expect(typeof health.activeWorkflows).toBe("number");
+
+      const stopPromise = kernel.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+    });
+  });
+});

packages/daemon/src/__tests__/log-store-crash-recovery.test.ts

@@ -0,0 +1,198 @@
+/**
+ * Phase 3 — LogStore crash recovery helpers tests.
+ *
+ * Tests for getThreadEvents() and getTriggerPayload().
+ */
+
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+
+import { createLogStore } from "../log-store.js";
+import type { LogStore } from "../log-store.js";
+
+describe("LogStore — crash recovery helpers (Phase 3)", () => {
+  let tmpDir: string;
+  let store: LogStore;
+
+  beforeEach(() => {
+    tmpDir = mkdtempSync(join(tmpdir(), "nerve-cr-log-test-"));
+    store = createLogStore(join(tmpDir, "data", "logs.db"));
+  });
+
+  afterEach(() => {
+    store.close();
+    rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  describe("getTriggerPayload", () => {
+    it("returns null for an unknown runId", () => {
+      expect(store.getTriggerPayload("no-such-run")).toBeNull();
+    });
+
+    it("returns the triggerPayload stored in the 'started' log entry", () => {
+      const payload = { task: "build", repo: "myrepo" };
+      store.upsertWorkflowRun(
+        {
+          source: "workflow",
+          type: "started",
+          refId: "run-1",
+          payload: JSON.stringify({ triggerPayload: payload }),
+          ts: 1000,
+        },
+        { runId: "run-1", workflow: "my-wf", status: "started", ts: 1000 },
+      );
+
+      const result = store.getTriggerPayload("run-1");
+      expect(result).toMatchObject(payload);
+    });
+
+    it("returns null when started log entry has no payload", () => {
+      store.upsertWorkflowRun(
+        {
+          source: "workflow",
+          type: "started",
+          refId: "run-2",
+          payload: null,
+          ts: 1000,
+        },
+        { runId: "run-2", workflow: "my-wf", status: "started", ts: 1000 },
+      );
+
+      expect(store.getTriggerPayload("run-2")).toBeNull();
+    });
+
+    it("returns the payload from the first 'started' entry (earliest)", () => {
+      const payloadA = { trigger: "first" };
+      const payloadB = { trigger: "second" };
+      // Insert two started entries for the same run
+      store.append({
+        source: "workflow",
+        type: "started",
+        refId: "run-3",
+        payload: JSON.stringify({ triggerPayload: payloadA }),
+        ts: 100,
+      });
+      store.append({
+        source: "workflow",
+        type: "started",
+        refId: "run-3",
+        payload: JSON.stringify({ triggerPayload: payloadB }),
+        ts: 200,
+      });
+
+      const result = store.getTriggerPayload("run-3");
+      // Should return the first (earliest) started entry
+      expect(result).toMatchObject(payloadA);
+    });
+  });
+
+  describe("getThreadEvents", () => {
+    it("returns empty array for an unknown runId", () => {
+      expect(store.getThreadEvents("no-such-run")).toHaveLength(0);
+    });
+
+    it("returns CommandEvents in insertion order", () => {
+      const events = [
+        { type: "thread_start", triggerPayload: {} },
+        { type: "scan_complete", items: ["a", "b"] },
+        { type: "process_done", count: 2 },
+      ];
+
+      for (const event of events) {
+        store.append({
+          source: "workflow",
+          type: "thread_command_event",
+          refId: "run-4",
+          payload: JSON.stringify(event),
+          ts: Date.now(),
+        });
+      }
+
+      const result = store.getThreadEvents("run-4");
+      expect(result).toHaveLength(3);
+      expect(result[0].type).toBe("thread_start");
+      expect(result[1].type).toBe("scan_complete");
+      expect(result[2].type).toBe("process_done");
+    });
+
+    it("skips entries with null payload", () => {
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-5",
+        payload: null,
+        ts: 1000,
+      });
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-5",
+        payload: JSON.stringify({ type: "valid_event" }),
+        ts: 1001,
+      });
+
+      const result = store.getThreadEvents("run-5");
+      expect(result).toHaveLength(1);
+      expect(result[0].type).toBe("valid_event");
+    });
+
+    it("only returns thread_command_event entries (not other workflow log types)", () => {
+      // Insert a mix of workflow log types
+      store.upsertWorkflowRun(
+        {
+          source: "workflow",
+          type: "started",
+          refId: "run-6",
+          payload: JSON.stringify({ triggerPayload: {} }),
+          ts: 1000,
+        },
+        { runId: "run-6", workflow: "my-wf", status: "started", ts: 1000 },
+      );
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-6",
+        payload: JSON.stringify({ type: "step_one" }),
+        ts: 1001,
+      });
+      store.append({
+        source: "workflow",
+        type: "step_complete",
+        refId: "run-6",
+        payload: JSON.stringify({ message: "done step" }),
+        ts: 1002,
+      });
+
+      const result = store.getThreadEvents("run-6");
+      expect(result).toHaveLength(1);
+      expect(result[0].type).toBe("step_one");
+    });
+
+    it("does not return events from a different runId", () => {
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-7",
+        payload: JSON.stringify({ type: "event_for_7" }),
+        ts: 1000,
+      });
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-8",
+        payload: JSON.stringify({ type: "event_for_8" }),
+        ts: 1001,
+      });
+
+      const result7 = store.getThreadEvents("run-7");
+      expect(result7).toHaveLength(1);
+      expect(result7[0].type).toBe("event_for_7");
+
+      const result8 = store.getThreadEvents("run-8");
+      expect(result8).toHaveLength(1);
+      expect(result8[0].type).toBe("event_for_8");
+    });
+  });
+});

packages/daemon/src/__tests__/log-store-workflow.test.ts

@@ -0,0 +1,186 @@
+/**
+ * Tests for workflow-related LogStore functionality (RFC-002 §6.2).
+ */
+
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+
+import { createLogStore } from "../log-store.js";
+import type { LogStore, WorkflowRun } from "../log-store.js";
+
+describe("LogStore — workflow_runs", () => {
+  let tmpDir: string;
+  let store: LogStore;
+
+  beforeEach(() => {
+    tmpDir = mkdtempSync(join(tmpdir(), "nerve-wf-log-test-"));
+    store = createLogStore(join(tmpDir, "data", "logs.db"));
+  });
+
+  afterEach(() => {
+    store.close();
+    rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  describe("upsertWorkflowRun", () => {
+    it("writes a log entry and creates a workflow_runs row atomically", () => {
+      const run: WorkflowRun = {
+        runId: "run-1",
+        workflow: "cleanup",
+        status: "started",
+        ts: 1000,
+      };
+
+      const entry = store.upsertWorkflowRun(
+        { source: "workflow", type: "started", refId: "run-1", payload: null, ts: 1000 },
+        run,
+      );
+
+      expect(entry.id).toBeGreaterThan(0);
+      expect(entry.source).toBe("workflow");
+      expect(entry.type).toBe("started");
+
+      const stored = store.getWorkflowRun("run-1");
+      expect(stored).not.toBeNull();
+      expect(stored?.runId).toBe("run-1");
+      expect(stored?.workflow).toBe("cleanup");
+      expect(stored?.status).toBe("started");
+      expect(stored?.ts).toBe(1000);
+    });
+
+    it("updates existing workflow_runs row on upsert (status transition)", () => {
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "started", refId: "run-2", payload: null, ts: 1000 },
+        { runId: "run-2", workflow: "cleanup", status: "started", ts: 1000 },
+      );
+
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "completed", refId: "run-2", payload: null, ts: 2000 },
+        { runId: "run-2", workflow: "cleanup", status: "completed", ts: 2000 },
+      );
+
+      const stored = store.getWorkflowRun("run-2");
+      expect(stored?.status).toBe("completed");
+      expect(stored?.ts).toBe(2000);
+
+      // Both log entries should be present (event sourcing)
+      const logs = store.query({ refId: "run-2" });
+      expect(logs).toHaveLength(2);
+    });
+
+    it("the log entries act as source of truth for event history", () => {
+      for (const [type, status, ts] of [
+        ["queued", "queued", 1000],
+        ["started", "started", 1001],
+        ["step_complete", "started", 1002],
+        ["completed", "completed", 1005],
+      ] as const) {
+        store.upsertWorkflowRun(
+          { source: "workflow", type, refId: "run-3", payload: null, ts },
+          { runId: "run-3", workflow: "cleanup", status, ts },
+        );
+      }
+
+      const logs = store.query({ source: "workflow", refId: "run-3" });
+      expect(logs).toHaveLength(4);
+      expect(logs[0].type).toBe("queued");
+      expect(logs[3].type).toBe("completed");
+    });
+  });
+
+  describe("getWorkflowRun", () => {
+    it("returns null for a non-existent runId", () => {
+      expect(store.getWorkflowRun("no-such-run")).toBeNull();
+    });
+
+    it("returns the latest state after multiple upserts", () => {
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "queued", refId: "run-4", payload: null, ts: 100 },
+        { runId: "run-4", workflow: "code-review", status: "queued", ts: 100 },
+      );
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "started", refId: "run-4", payload: null, ts: 200 },
+        { runId: "run-4", workflow: "code-review", status: "started", ts: 200 },
+      );
+
+      const run = store.getWorkflowRun("run-4");
+      expect(run?.status).toBe("started");
+      expect(run?.ts).toBe(200);
+    });
+  });
+
+  describe("getActiveWorkflowRuns", () => {
+    beforeEach(() => {
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "queued", refId: "r1", payload: null, ts: 100 },
+        { runId: "r1", workflow: "cleanup", status: "queued", ts: 100 },
+      );
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "started", refId: "r2", payload: null, ts: 200 },
+        { runId: "r2", workflow: "cleanup", status: "started", ts: 200 },
+      );
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "completed", refId: "r3", payload: null, ts: 300 },
+        { runId: "r3", workflow: "cleanup", status: "completed", ts: 300 },
+      );
+      store.upsertWorkflowRun(
+        { source: "workflow", type: "failed", refId: "r4", payload: null, ts: 400 },
+        { runId: "r4", workflow: "deploy", status: "queued", ts: 400 },
+      );
+    });
+
+    it("returns queued and started runs by default", () => {
+      const active = store.getActiveWorkflowRuns();
+      expect(active).toHaveLength(3);
+      const ids = active.map((r) => r.runId);
+      expect(ids).toContain("r1");
+      expect(ids).toContain("r2");
+      expect(ids).toContain("r4");
+    });
+
+    it("excludes completed and failed runs", () => {
+      const active = store.getActiveWorkflowRuns();
+      const ids = active.map((r) => r.runId);
+      expect(ids).not.toContain("r3");
+    });
+
+    it("filters by workflow name", () => {
+      const cleanupRuns = store.getActiveWorkflowRuns("cleanup");
+      expect(cleanupRuns).toHaveLength(2);
+      const ids = cleanupRuns.map((r) => r.runId);
+      expect(ids).toContain("r1");
+      expect(ids).toContain("r2");
+    });
+
+    it("returns only matching workflow when name given", () => {
+      const deployRuns = store.getActiveWorkflowRuns("deploy");
+      expect(deployRuns).toHaveLength(1);
+      expect(deployRuns[0].runId).toBe("r4");
+    });
+
+    it("returns empty array when workflow name matches no active runs", () => {
+      expect(store.getActiveWorkflowRuns("nonexistent")).toHaveLength(0);
+    });
+
+    it("returns runs ordered by ts ascending", () => {
+      const active = store.getActiveWorkflowRuns();
+      expect(active[0].ts).toBeLessThan(active[1].ts);
+    });
+  });
+
+  describe("all statuses are storable", () => {
+    it.each(["queued", "started", "completed", "failed", "crashed", "dropped"] as const)(
+      "stores status=%s",
+      (status) => {
+        const runId = `run-${status}`;
+        store.upsertWorkflowRun(
+          { source: "workflow", type: status, refId: runId, payload: null, ts: 1 },
+          { runId, workflow: "test", status, ts: 1 },
+        );
+        expect(store.getWorkflowRun(runId)?.status).toBe(status);
+      },
+    );
+  });
+});

packages/daemon/src/__tests__/workflow-manager.test.ts

@@ -0,0 +1,364 @@
+/**
+ * Tests for WorkflowManager — concurrency, overflow, queue, and lifecycle.
+ */
+
+import { EventEmitter } from "node:events";
+
+import type { NerveConfig } from "@uncaged/nerve-core";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+// ---------------------------------------------------------------------------
+// Mock child_process.fork before importing workflow-manager
+// ---------------------------------------------------------------------------
+
+type MockChild = EventEmitter & {
+  send: ReturnType<typeof vi.fn>;
+  kill: ReturnType<typeof vi.fn>;
+  connected: boolean;
+  exitCode: number | null;
+  pid: number;
+};
+
+const mockChildren: MockChild[] = [];
+
+function makeMockChild(pid = 1): MockChild {
+  const child = new EventEmitter() as MockChild;
+  child.connected = true;
+  child.exitCode = null;
+  child.pid = pid;
+  child.send = vi.fn((msg: unknown) => {
+    if (
+      msg !== null &&
+      typeof msg === "object" &&
+      (msg as Record<string, unknown>).type === "shutdown"
+    ) {
+      setImmediate(() => {
+        child.exitCode = 0;
+        child.connected = false;
+        child.emit("exit", 0, null);
+      });
+    }
+  });
+  child.kill = vi.fn((_signal?: string) => {
+    child.exitCode = 1;
+    child.connected = false;
+    child.emit("exit", null, _signal ?? "SIGKILL");
+  });
+  return child;
+}
+
+vi.mock("node:child_process", () => ({
+  fork: vi.fn((_script: string, _args: string[], _opts: unknown) => {
+    const child = makeMockChild(mockChildren.length + 1);
+    mockChildren.push(child);
+    return child;
+  }),
+}));
+
+// Import after mock is set up
+const { createWorkflowManager } = await import("../workflow-manager.js");
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function makeLogStore() {
+  return {
+    append: vi.fn(),
+    query: vi.fn(() => []),
+    getMeta: vi.fn(() => null),
+    setMeta: vi.fn(),
+    upsertWorkflowRun: vi.fn(),
+    appendWithWorkflowUpdate: vi.fn(),
+    getWorkflowRun: vi.fn(() => null),
+    getActiveWorkflowRuns: vi.fn(() => []),
+    getTriggerPayload: vi.fn(() => null),
+    getThreadEvents: vi.fn(() => []),
+    close: vi.fn(),
+  };
+}
+
+function makeConfig(overrides: Partial<NerveConfig["workflows"]> = {}): NerveConfig {
+  return {
+    senses: {},
+    reflexes: [],
+    workflows: overrides as NerveConfig["workflows"],
+  };
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+describe("WorkflowManager", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+    vi.clearAllMocks();
+  });
+
+  describe("startWorkflow under concurrency limit dispatches thread", () => {
+    it("forks a worker and sends start-thread when active < concurrency", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-workflow": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-workflow", { event: "test" });
+
+      expect(mockChildren).toHaveLength(1);
+      expect(mockChildren[0].send).toHaveBeenCalledWith(
+        expect.objectContaining({ type: "start-thread", workflow: "my-workflow" }),
+      );
+      expect(mgr.activeCount("my-workflow")).toBe(1);
+    });
+
+    it("reuses the same worker for a second thread under the limit", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-workflow": { concurrency: 3, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-workflow", { n: 1 });
+      mgr.startWorkflow("my-workflow", { n: 2 });
+
+      // Only one forked child — worker is reused
+      expect(mockChildren).toHaveLength(1);
+      expect(mockChildren[0].send).toHaveBeenCalledTimes(2);
+      expect(mgr.activeCount("my-workflow")).toBe(2);
+    });
+
+    it("logs a 'started' event for each dispatched thread", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "my-workflow": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("my-workflow", { x: 1 });
+
+      expect(logStore.upsertWorkflowRun).toHaveBeenCalledWith(
+        expect.objectContaining({ source: "workflow", type: "started" }),
+        expect.objectContaining({ workflow: "my-workflow", status: "started" }),
+      );
+    });
+  });
+
+  describe("startWorkflow at limit with drop overflow drops the request", () => {
+    it("does NOT send start-thread when at concurrency limit with overflow=drop", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "drop-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("drop-wf", { first: true });
+      // now at limit — second call should be dropped
+      mgr.startWorkflow("drop-wf", { second: true });
+
+      expect(mgr.activeCount("drop-wf")).toBe(1);
+      expect(mgr.queueLength("drop-wf")).toBe(0);
+      // Only one start-thread sent
+      expect(mockChildren[0].send).toHaveBeenCalledTimes(1);
+    });
+
+    it("logs a 'dropped' event when overflow=drop", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "drop-wf": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("drop-wf", {});
+      mgr.startWorkflow("drop-wf", {});
+
+      const droppedCall = logStore.upsertWorkflowRun.mock.calls.find(
+        ([entry]) => entry.type === "dropped",
+      );
+      expect(droppedCall).toBeDefined();
+    });
+  });
+
+  describe("startWorkflow at limit with queue overflow queues the request", () => {
+    it("queues the thread when at concurrency limit with overflow=queue", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "queue-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { second: true });
+
+      expect(mgr.activeCount("queue-wf")).toBe(1);
+      expect(mgr.queueLength("queue-wf")).toBe(1);
+    });
+
+    it("logs a 'queued' event for the waiting thread", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "queue-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("queue-wf", {});
+      mgr.startWorkflow("queue-wf", {});
+
+      const queuedCall = logStore.upsertWorkflowRun.mock.calls.find(
+        ([entry]) => entry.type === "queued",
+      );
+      expect(queuedCall).toBeDefined();
+      expect(queuedCall?.[1]).toMatchObject({ workflow: "queue-wf", status: "queued" });
+    });
+  });
+
+  describe("queue overflow with maxQueue drops oldest when full", () => {
+    it("drops oldest queued item when queue reaches maxQueue", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "queue-wf": { concurrency: 1, overflow: "queue", maxQueue: 2 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      // Fill the concurrency slot
+      mgr.startWorkflow("queue-wf", { n: 0 });
+      // Fill the queue to maxQueue
+      mgr.startWorkflow("queue-wf", { n: 1 });
+      mgr.startWorkflow("queue-wf", { n: 2 });
+      // This one should push out { n: 1 }
+      mgr.startWorkflow("queue-wf", { n: 3 });
+
+      // Queue should still be at maxQueue (2)
+      expect(mgr.queueLength("queue-wf")).toBe(2);
+
+      // There should be a dropped event (for the oldest evicted item)
+      const droppedCalls = logStore.upsertWorkflowRun.mock.calls.filter(
+        ([entry]) => entry.type === "dropped",
+      );
+      expect(droppedCalls).toHaveLength(1);
+    });
+  });
+
+  describe("completing a thread dequeues the next one", () => {
+    it("dispatches the next queued thread when the active thread sends completed", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "queue-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { second: true });
+
+      expect(mgr.activeCount("queue-wf")).toBe(1);
+      expect(mgr.queueLength("queue-wf")).toBe(1);
+
+      // Simulate the worker sending a "thread-event: completed"
+      const child = mockChildren[0];
+      const startCalls = (child.send as ReturnType<typeof vi.fn>).mock.calls;
+      const firstRunId = (startCalls[0][0] as { runId: string }).runId;
+
+      child.emit("message", {
+        type: "thread-event",
+        runId: firstRunId,
+        eventType: "completed",
+        payload: null,
+      });
+
+      // The queued thread should now be active
+      expect(mgr.activeCount("queue-wf")).toBe(1);
+      expect(mgr.queueLength("queue-wf")).toBe(0);
+      // A second start-thread message should have been sent
+      expect(child.send).toHaveBeenCalledTimes(2);
+      expect(child.send).toHaveBeenLastCalledWith(
+        expect.objectContaining({ type: "start-thread", workflow: "queue-wf" }),
+      );
+    });
+
+    it("dispatches next queued thread when active thread sends failed", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "queue-wf": { concurrency: 1, overflow: "queue", maxQueue: 5 },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { second: true });
+
+      const child = mockChildren[0];
+      const firstRunId = (child.send as ReturnType<typeof vi.fn>).mock.calls[0][0].runId as string;
+
+      child.emit("message", {
+        type: "thread-event",
+        runId: firstRunId,
+        eventType: "failed",
+        payload: { error: "boom" },
+      });
+
+      expect(mgr.activeCount("queue-wf")).toBe(1);
+      expect(mgr.queueLength("queue-wf")).toBe(0);
+    });
+  });
+
+  describe("stop() shuts down all workers", () => {
+    it("sends shutdown to every worker and resolves when they exit", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "wf-a": { concurrency: 2, overflow: "drop" },
+        "wf-b": { concurrency: 1, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("wf-a", {});
+      mgr.startWorkflow("wf-b", {});
+
+      // Two distinct workers should have been forked
+      expect(mockChildren).toHaveLength(2);
+
+      const stopPromise = mgr.stop();
+      // Let setImmediate callbacks run (the mock exits on shutdown)
+      await vi.runAllTimersAsync();
+      await stopPromise;
+
+      for (const child of mockChildren) {
+        expect(child.send).toHaveBeenCalledWith(expect.objectContaining({ type: "shutdown" }));
+      }
+    });
+
+    it("ignores startWorkflow calls after stop()", async () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({
+        "wf-a": { concurrency: 2, overflow: "drop" },
+      });
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      const stopPromise = mgr.stop();
+      await vi.runAllTimersAsync();
+      await stopPromise;
+
+      mgr.startWorkflow("wf-a", {});
+
+      // No worker should have been spawned
+      expect(mockChildren).toHaveLength(0);
+    });
+  });
+
+  describe("unknown workflow name", () => {
+    it("does nothing for an unknown workflow name", () => {
+      const logStore = makeLogStore();
+      const config = makeConfig({});
+      const mgr = createWorkflowManager("/nerve-root", config, logStore);
+
+      mgr.startWorkflow("no-such-workflow", {});
+
+      expect(mockChildren).toHaveLength(0);
+      expect(logStore.upsertWorkflowRun).not.toHaveBeenCalled();
+    });
+  });
+});

packages/daemon/src/file-watcher.ts

@@ -31,7 +31,13 @@ export type ConfigFileChange = {
   filePath: string;
 };
 
-export type FileChange = SenseFileChange | ConfigFileChange;
+export type WorkflowFileChange = {
+  kind: "workflow";
+  workflowName: string;
+  filePath: string;
+};
+
+export type FileChange = SenseFileChange | ConfigFileChange | WorkflowFileChange;
 
 export type FileChangeHandler = (change: FileChange) => void;
 
@@ -61,6 +67,28 @@ export function createFileWatcher(
     );
   }
 
+  function handleSenseChange(normalized: string, filename: string): void {
+    if (!(normalized.startsWith("senses/") && normalized.endsWith(".ts"))) return;
+    const rel = relative("senses", normalized);
+    const senseName = rel.split("/")[0];
+    if (senseName) {
+      debounced(`sense:${senseName}`, () => {
+        handler({ kind: "sense", senseName, filePath: join(nerveRoot, filename) });
+      });
+    }
+  }
+
+  function handleWorkflowChange(normalized: string, filename: string): void {
+    if (!(normalized.startsWith("workflows/") && normalized.endsWith(".ts"))) return;
+    const rel = relative("workflows", normalized);
+    const workflowName = rel.split("/")[0];
+    if (workflowName) {
+      debounced(`workflow:${workflowName}`, () => {
+        handler({ kind: "workflow", workflowName, filePath: join(nerveRoot, filename) });
+      });
+    }
+  }
+
   function handleFsEvent(_eventType: string, filename: string | null): void {
     if (filename === null) return;
 
@@ -73,19 +101,8 @@ export function createFileWatcher(
       return;
     }
 
-    if (normalized.startsWith("senses/") && normalized.endsWith(".ts")) {
-      const rel = relative("senses", normalized);
-      const senseName = rel.split("/")[0];
-      if (senseName) {
-        debounced(`sense:${senseName}`, () => {
-          handler({
-            kind: "sense",
-            senseName,
-            filePath: join(nerveRoot, filename),
-          });
-        });
-      }
-    }
+    handleSenseChange(normalized, filename);
+    handleWorkflowChange(normalized, filename);
   }
 
   try {

packages/daemon/src/index.ts

@@ -8,6 +8,10 @@ export type {
   ErrorMessage,
   ReadyMessage,
   WorkerToParentMessage,
+  StartThreadMessage,
+  ResumeThreadMessage,
+  ThreadEventMessage,
+  WorkflowErrorMessage,
 } from "./ipc.js";
 
 export type { SignalBus, SignalHandler, Unsubscribe } from "./signal-bus.js";
@@ -29,4 +33,7 @@ export { createFileWatcher } from "./file-watcher.js";
 export type { FileWatcher, FileChange, FileChangeHandler } from "./file-watcher.js";
 
 export { createLogStore } from "./log-store.js";
-export type { LogStore, LogEntry, LogQuery } from "./log-store.js";
+export type { LogStore, LogEntry, LogQuery, WorkflowRun, WorkflowRunStatus } from "./log-store.js";
+
+export { createWorkflowManager } from "./workflow-manager.js";
+export type { WorkflowManager } from "./workflow-manager.js";

packages/daemon/src/ipc.ts

@@ -22,8 +22,36 @@ export type HealthRequestMessage = {
   type: "health-request";
 };
 
+// ---------------------------------------------------------------------------
+// Workflow IPC messages (RFC-002 §5.2)
+// ---------------------------------------------------------------------------
+
+/** Parent → Workflow Worker: start a new thread */
+export type StartThreadMessage = {
+  type: "start-thread";
+  runId: string;
+  workflow: string;
+  /** The trigger payload from the Reflex that initiated this thread. */
+  triggerPayload: unknown;
+};
+
+/** Parent → Workflow Worker: resume an existing thread after crash recovery */
+export type ResumeThreadMessage = {
+  type: "resume-thread";
+  runId: string;
+  /** Serialised CommandEvent history to rebuild ThreadState. */
+  events: Array<{ type: string; [key: string]: unknown }>;
+  /** Serialised trigger payload (the same value as in the original start-thread). */
+  triggerPayload: unknown;
+};
+
 /** Union of all messages the parent sends to a worker */
-export type ParentToWorkerMessage = ComputeMessage | ShutdownMessage | HealthRequestMessage;
+export type ParentToWorkerMessage =
+  | ComputeMessage
+  | ShutdownMessage
+  | HealthRequestMessage
+  | StartThreadMessage
+  | ResumeThreadMessage;
 
 /** Worker → Parent: compute produced a signal */
 export type SignalMessage = {
@@ -51,14 +79,69 @@ export type HealthResponseMessage = {
   inFlightCount: number;
 };
 
+// ---------------------------------------------------------------------------
+// Workflow Worker → Parent messages (RFC-002 §5.2)
+// ---------------------------------------------------------------------------
+
+/** Valid lifecycle event types for a workflow thread. */
+export type ThreadEventType = "queued" | "started" | "step_complete" | "completed" | "failed";
+
+/**
+ * Workflow Worker → Parent: a thread lifecycle event.
+ */
+export type ThreadEventMessage = {
+  type: "thread-event";
+  runId: string;
+  eventType: ThreadEventType;
+  payload: unknown;
+};
+
+/** Workflow Worker → Parent: a thread encountered an unrecoverable error. */
+export type WorkflowErrorMessage = {
+  type: "workflow-error";
+  runId: string;
+  error: string;
+};
+
+/** Workflow Worker → Parent: a thread CommandEvent produced by a role (for crash recovery). */
+export type ThreadCommandEventMessage = {
+  type: "thread-command-event";
+  runId: string;
+  /** The CommandEvent returned by role.execute() — will be persisted for crash recovery. */
+  event: { type: string; [key: string]: unknown };
+};
+
 /** Union of all messages a worker sends to the parent */
 export type WorkerToParentMessage =
   | SignalMessage
   | ErrorMessage
   | ReadyMessage
-  | HealthResponseMessage;
+  | HealthResponseMessage
+  | ThreadEventMessage
+  | WorkflowErrorMessage
+  | ThreadCommandEventMessage;
 
-const PARENT_MSG_TYPES = new Set(["compute", "shutdown", "health-request"]);
+const PARENT_MSG_TYPES = new Set([
+  "compute",
+  "shutdown",
+  "health-request",
+  "start-thread",
+  "resume-thread",
+]);
+
+function validateStartThreadMsg(obj: Record<string, unknown>): string | null {
+  if (typeof obj.runId !== "string") return "'start-thread' message missing string 'runId'";
+  if (typeof obj.workflow !== "string") return "'start-thread' message missing string 'workflow'";
+  if (!("triggerPayload" in obj)) return "'start-thread' message missing 'triggerPayload'";
+  return null;
+}
+
+function validateResumeThreadMsg(obj: Record<string, unknown>): string | null {
+  if (typeof obj.runId !== "string") return "'resume-thread' message missing string 'runId'";
+  if (!Array.isArray(obj.events)) return "'resume-thread' message missing 'events' array";
+  if (!("triggerPayload" in obj)) return "'resume-thread' message missing 'triggerPayload'";
+  return null;
+}
 
 /** Validate and parse an unknown IPC message received from the parent process. */
 export function parseParentMessage(raw: unknown): Result<ParentToWorkerMessage> {
@@ -72,6 +155,14 @@ export function parseParentMessage(raw: unknown): Result<ParentToWorkerMessage>
   if (!PARENT_MSG_TYPES.has(obj.type)) {
     return err(new Error(`Unknown IPC message type: "${obj.type}"`));
   }
+  if (obj.type === "start-thread") {
+    const errMsg = validateStartThreadMsg(obj);
+    if (errMsg !== null) return err(new Error(errMsg));
+  }
+  if (obj.type === "resume-thread") {
+    const errMsg = validateResumeThreadMsg(obj);
+    if (errMsg !== null) return err(new Error(errMsg));
+  }
   return ok(raw as ParentToWorkerMessage);
 }
 
@@ -108,7 +199,71 @@ function parseHealthResponseMsg(
   return ok(raw as HealthResponseMessage);
 }
 
-const WORKER_MSG_TYPES = new Set(["signal", "error", "ready", "health-response"]);
+const THREAD_EVENT_TYPES = new Set<string>([
+  "queued",
+  "started",
+  "step_complete",
+  "completed",
+  "failed",
+]);
+
+function parseThreadEventMsg(
+  obj: Record<string, unknown>,
+  raw: unknown,
+): Result<WorkerToParentMessage> {
+  if (typeof obj.runId !== "string") {
+    return err(new Error("Worker 'thread-event' message missing string 'runId' field"));
+  }
+  if (typeof obj.eventType !== "string" || !THREAD_EVENT_TYPES.has(obj.eventType)) {
+    return err(
+      new Error(`Worker 'thread-event' message has invalid 'eventType': "${obj.eventType}"`),
+    );
+  }
+  if (!("payload" in obj)) {
+    return err(new Error("Worker 'thread-event' message missing 'payload' field"));
+  }
+  return ok(raw as ThreadEventMessage);
+}
+
+function parseWorkflowErrorMsg(
+  obj: Record<string, unknown>,
+  raw: unknown,
+): Result<WorkerToParentMessage> {
+  if (typeof obj.runId !== "string") {
+    return err(new Error("Worker 'workflow-error' message missing string 'runId' field"));
+  }
+  if (typeof obj.error !== "string") {
+    return err(new Error("Worker 'workflow-error' message missing string 'error' field"));
+  }
+  return ok(raw as WorkflowErrorMessage);
+}
+
+const WORKER_MSG_TYPES = new Set([
+  "signal",
+  "error",
+  "ready",
+  "health-response",
+  "thread-event",
+  "workflow-error",
+  "thread-command-event",
+]);
+
+function parseThreadCommandEventMsg(
+  obj: Record<string, unknown>,
+  raw: unknown,
+): Result<WorkerToParentMessage> {
+  if (typeof obj.runId !== "string") {
+    return err(new Error("Worker 'thread-command-event' message missing string 'runId' field"));
+  }
+  if (obj.event === null || typeof obj.event !== "object") {
+    return err(new Error("Worker 'thread-command-event' message missing object 'event' field"));
+  }
+  const event = obj.event as Record<string, unknown>;
+  if (typeof event.type !== "string") {
+    return err(new Error("Worker 'thread-command-event' event missing string 'type' field"));
+  }
+  return ok(raw as ThreadCommandEventMessage);
+}
 
 /** Validate and parse an unknown IPC message received from a worker process. */
 export function parseWorkerMessage(raw: unknown): Result<WorkerToParentMessage> {
@@ -125,5 +280,8 @@ export function parseWorkerMessage(raw: unknown): Result<WorkerToParentMessage>
   if (obj.type === "signal") return parseSignalMsg(obj, raw);
   if (obj.type === "error") return parseErrorMsg(obj, raw);
   if (obj.type === "health-response") return parseHealthResponseMsg(obj, raw);
+  if (obj.type === "thread-event") return parseThreadEventMsg(obj, raw);
+  if (obj.type === "workflow-error") return parseWorkflowErrorMsg(obj, raw);
+  if (obj.type === "thread-command-event") return parseThreadCommandEventMsg(obj, raw);
   return ok({ type: "ready" });
 }

packages/daemon/src/kernel.ts

@@ -31,12 +31,15 @@ import { createReflexScheduler } from "./reflex-scheduler.js";
 import type { ReflexScheduler } from "./reflex-scheduler.js";
 import { createSignalBus } from "./signal-bus.js";
 import type { SignalBus } from "./signal-bus.js";
+import { createWorkflowManager } from "./workflow-manager.js";
+import type { WorkflowManager } from "./workflow-manager.js";
 
 export type KernelHealth = {
   uptime: number;
   activeSenses: number;
   activeGroups: number;
   pendingComputes: number;
+  activeWorkflows: number;
   memoryUsage: NodeJS.MemoryUsage;
 };
 
@@ -46,6 +49,7 @@ export type Kernel = {
   senseCount: number;
   bus: SignalBus;
   logStore: LogStore;
+  workflowManager: WorkflowManager;
   /** Resolves when all workers have sent their initial "ready" message. */
   ready: Promise<void>;
   /** Returns the PID of the worker process for a given group, or null if not found. */
@@ -143,6 +147,8 @@ export function createKernel(
     return _signalIdCounter;
   }
 
+  const workflowManager = createWorkflowManager(nerveRoot, config, logStore);
+
   const groups = new Set<string>();
   for (const senseConfig of Object.values(config.senses)) {
     groups.add(senseConfig.group);
@@ -267,7 +273,12 @@ export function createKernel(
     sendCompute(entry.process, senseName);
   }
 
-  scheduler = createReflexScheduler(config, bus, triggerFn, { logStore });
+  scheduler = createReflexScheduler(config, bus, triggerFn, {
+    logStore,
+    workflowTriggerFn: (workflowName, payload) => {
+      workflowManager.startWorkflow(workflowName, payload);
+    },
+  });
 
   if (groups.size === 0) {
     readyResolve?.();
@@ -346,11 +357,37 @@ export function createKernel(
   function reloadConfig(newConfig: NerveConfig): void {
     const oldGroups = collectGroups(config);
     const oldConfig = config;
+    const oldWorkflows = config.workflows ?? {};
     config = newConfig;
     // Note: pending/throttled computes in the old scheduler are silently dropped here.
     // In-flight state is not preserved across reloadConfig.
     scheduler.stop();
-    scheduler = createReflexScheduler(config, bus, triggerFn, { logStore });
+    scheduler = createReflexScheduler(config, bus, triggerFn, {
+      logStore,
+      workflowTriggerFn: (workflowName, payload) => {
+        workflowManager.startWorkflow(workflowName, payload);
+      },
+    });
+    // Update workflow concurrency/overflow config incrementally — no restart needed
+    workflowManager.updateConfig(newConfig);
+
+    const newWorkflows = newConfig.workflows ?? {};
+
+    // Drain + remove workers for deleted workflows
+    for (const workflowName of Object.keys(oldWorkflows)) {
+      if (!(workflowName in newWorkflows)) {
+        process.stderr.write(
+          `[kernel] workflow "${workflowName}" removed from config, draining worker\n`,
+        );
+        workflowManager.drainAndRespawn(workflowName).catch((e) => {
+          const msg = e instanceof Error ? e.message : String(e);
+          process.stderr.write(
+            `[kernel] drainAndRespawn error for removed workflow "${workflowName}": ${msg}\n`,
+          );
+        });
+      }
+    }
+
     const newGroups = collectGroups(newConfig);
     removeStaleGroups(oldGroups, newGroups);
     addNewGroups(oldGroups, newGroups);
@@ -378,6 +415,7 @@ export function createKernel(
       activeSenses: Object.keys(config.senses).length,
       activeGroups: workers.size,
       pendingComputes: 0,
+      activeWorkflows: workflowManager.totalActiveCount(),
       memoryUsage: process.memoryUsage(),
     };
   }
@@ -401,6 +439,23 @@ export function createKernel(
     });
   }
 
+  function handleWorkflowFileChange(workflowName: string): void {
+    process.stderr.write(
+      `[kernel] workflow file changed: "${workflowName}", draining and respawning worker\n`,
+    );
+    logStore.append({
+      source: "system",
+      type: "workflow_reload",
+      refId: workflowName,
+      payload: null,
+      ts: Date.now(),
+    });
+    workflowManager.drainAndRespawn(workflowName).catch((e) => {
+      const msg = e instanceof Error ? e.message : String(e);
+      process.stderr.write(`[kernel] drainAndRespawn error for "${workflowName}": ${msg}\n`);
+    });
+  }
+
   function handleConfigFileChange(): void {
     process.stderr.write("[kernel] nerve.yaml changed, reloading config\n");
     logStore.append({
@@ -431,6 +486,7 @@ export function createKernel(
     fileWatcher = createFileWatcher(nerveRoot, (change) => {
       if (change.kind === "sense") handleSenseFileChange(change.senseName);
       if (change.kind === "config") handleConfigFileChange();
+      if (change.kind === "workflow") handleWorkflowFileChange(change.workflowName);
     });
   }
 
@@ -441,6 +497,7 @@ export function createKernel(
       fileWatcher = null;
     }
     scheduler.stop();
+    await workflowManager.stop();
     const exitPromises: Promise<void>[] = [];
     for (const entry of workers.values()) {
       sendShutdown(entry.process);
@@ -469,6 +526,7 @@ export function createKernel(
     senseCount,
     bus,
     logStore,
+    workflowManager,
     ready,
     getWorkerPid,
     triggerCompute: triggerFn,

packages/daemon/src/log-store.ts

@@ -30,11 +30,76 @@ export type LogQuery = {
   limit?: number;
 };
 
+// ---------------------------------------------------------------------------
+// Workflow runs materialized view (RFC-002 §6.2)
+// ---------------------------------------------------------------------------
+
+export type WorkflowRunStatus =
+  | "queued"
+  | "started"
+  | "completed"
+  | "failed"
+  | "crashed"
+  | "dropped"
+  | "interrupted";
+
+const VALID_WORKFLOW_STATUSES = new Set<string>([
+  "queued",
+  "started",
+  "completed",
+  "failed",
+  "crashed",
+  "dropped",
+  "interrupted",
+]);
+
+function validateWorkflowRunStatus(status: string): WorkflowRunStatus {
+  if (!VALID_WORKFLOW_STATUSES.has(status)) {
+    throw new Error(`Invalid workflow run status from DB: "${status}"`);
+  }
+  return status as WorkflowRunStatus;
+}
+
+/** One row in the workflow_runs materialized table. */
+export type WorkflowRun = {
+  runId: string;
+  workflow: string;
+  status: WorkflowRunStatus;
+  ts: number;
+};
+
 export type LogStore = {
   append: (entry: Omit<LogEntry, "id">) => LogEntry;
   query: (filter?: LogQuery) => LogEntry[];
   getMeta: (key: string) => string | null;
   setMeta: (key: string, value: string) => void;
+  /**
+   * Append a workflow log event and atomically upsert the workflow_runs
+   * materialized table — both in a single SQLite transaction (RFC-002 §6.2).
+   */
+  upsertWorkflowRun: (entry: Omit<LogEntry, "id">, run: WorkflowRun) => LogEntry;
+  /**
+   * Alias for upsertWorkflowRun — append a log entry and update workflow_runs
+   * in one atomic transaction.
+   */
+  appendWithWorkflowUpdate: (entry: Omit<LogEntry, "id">, run: WorkflowRun) => LogEntry;
+  /** Get the current materialized state of a specific workflow run. */
+  getWorkflowRun: (runId: string) => WorkflowRun | null;
+  /**
+   * Get all workflow runs with status 'queued' or 'started'.
+   * Optionally filter by workflow name.
+   */
+  getActiveWorkflowRuns: (workflowName?: string) => WorkflowRun[];
+  /**
+   * Get the trigger payload for a workflow run (stored in the 'started' log entry).
+   * Returns null if not found.
+   */
+  getTriggerPayload: (runId: string) => unknown;
+  /**
+   * Get all workflow CommandEvents for a specific run, ordered by id ASC.
+   * Used for crash recovery to rebuild ThreadState.
+   */
+  getThreadEvents: (runId: string) => Array<{ type: string; [key: string]: unknown }>;
   close: () => void;
 };
 
@@ -56,6 +121,16 @@ CREATE TABLE IF NOT EXISTS meta (
   key   TEXT PRIMARY KEY,
   value TEXT NOT NULL
 );
+
+CREATE TABLE IF NOT EXISTS workflow_runs (
+  run_id   TEXT PRIMARY KEY,
+  workflow TEXT NOT NULL,
+  status   TEXT NOT NULL,
+  ts       INTEGER NOT NULL
+);
+
+CREATE INDEX IF NOT EXISTS idx_workflow_runs_status ON workflow_runs(status);
+CREATE INDEX IF NOT EXISTS idx_workflow_runs_workflow ON workflow_runs(workflow);
 `;
 
 export function createLogStore(dbPath: string): LogStore {
@@ -74,6 +149,49 @@ export function createLogStore(dbPath: string): LogStore {
     "INSERT INTO meta (key, value) VALUES (@key, @value) ON CONFLICT(key) DO UPDATE SET value = @value",
   );
 
+  const upsertWorkflowRunStmt = sqlite.prepare(
+    "INSERT OR REPLACE INTO workflow_runs (run_id, workflow, status, ts) VALUES (@runId, @workflow, @status, @ts)",
+  );
+
+  const getWorkflowRunStmt = sqlite.prepare(
+    "SELECT run_id, workflow, status, ts FROM workflow_runs WHERE run_id = ?",
+  );
+
+  const getTriggerPayloadStmt = sqlite.prepare(
+    "SELECT payload FROM logs WHERE source = 'workflow' AND type = 'started' AND ref_id = ? ORDER BY id ASC LIMIT 1",
+  );
+
+  const getThreadEventsStmt = sqlite.prepare(
+    "SELECT payload FROM logs WHERE source = 'workflow' AND type = 'thread_command_event' AND ref_id = ? ORDER BY id ASC",
+  );
+
+  const getActiveWorkflowRunsStmt = sqlite.prepare(
+    "SELECT run_id, workflow, status, ts FROM workflow_runs WHERE status IN ('queued', 'started') ORDER BY ts ASC",
+  );
+
+  const getActiveWorkflowRunsByNameStmt = sqlite.prepare(
+    "SELECT run_id, workflow, status, ts FROM workflow_runs WHERE status IN ('queued', 'started') AND workflow = ? ORDER BY ts ASC",
+  );
+
+  const upsertWorkflowRunTx = sqlite.transaction(
+    (entry: Omit<LogEntry, "id">, run: WorkflowRun) => {
+      const info = insertStmt.run({
+        source: entry.source,
+        type: entry.type,
+        refId: entry.refId,
+        payload: entry.payload,
+        ts: entry.ts,
+      });
+      upsertWorkflowRunStmt.run({
+        runId: run.runId,
+        workflow: run.workflow,
+        status: run.status,
+        ts: run.ts,
+      });
+      return { ...entry, id: Number(info.lastInsertRowid) };
+    },
+  );
+
   function append(entry: Omit<LogEntry, "id">): LogEntry {
     const info = insertStmt.run({
       source: entry.source,
@@ -142,9 +260,92 @@ export function createLogStore(dbPath: string): LogStore {
     setMetaStmt.run({ key, value });
   }
 
+  function upsertWorkflowRun(entry: Omit<LogEntry, "id">, run: WorkflowRun): LogEntry {
+    return upsertWorkflowRunTx(entry, run) as LogEntry;
+  }
+
+  function appendWithWorkflowUpdate(entry: Omit<LogEntry, "id">, run: WorkflowRun): LogEntry {
+    return upsertWorkflowRunTx(entry, run) as LogEntry;
+  }
+
+  function getWorkflowRun(runId: string): WorkflowRun | null {
+    const row = getWorkflowRunStmt.get(runId) as
+      | { run_id: string; workflow: string; status: string; ts: number }
+      | undefined;
+    if (row === undefined) return null;
+    return {
+      runId: row.run_id,
+      workflow: row.workflow,
+      status: validateWorkflowRunStatus(row.status),
+      ts: row.ts,
+    };
+  }
+
+  function getActiveWorkflowRuns(workflowName?: string): WorkflowRun[] {
+    const rows = (
+      workflowName !== undefined
+        ? getActiveWorkflowRunsByNameStmt.all(workflowName)
+        : getActiveWorkflowRunsStmt.all()
+    ) as Array<{ run_id: string; workflow: string; status: string; ts: number }>;
+    return rows.map((r) => ({
+      runId: r.run_id,
+      workflow: r.workflow,
+      status: validateWorkflowRunStatus(r.status),
+      ts: r.ts,
+    }));
+  }
+
+  function getTriggerPayload(runId: string): unknown {
+    const row = getTriggerPayloadStmt.get(runId) as { payload: string | null } | undefined;
+    if (row === undefined || row.payload === null) return null;
+    try {
+      const parsed = JSON.parse(row.payload) as unknown;
+      if (parsed !== null && typeof parsed === "object") {
+        const obj = parsed as Record<string, unknown>;
+        return obj.triggerPayload ?? null;
+      }
+    } catch {
+      // malformed
+    }
+    return null;
+  }
+
+  function getThreadEvents(runId: string): Array<{ type: string; [key: string]: unknown }> {
+    const rows = getThreadEventsStmt.all(runId) as Array<{ payload: string | null }>;
+    const result: Array<{ type: string; [key: string]: unknown }> = [];
+    for (const row of rows) {
+      if (row.payload === null) continue;
+      try {
+        const parsed = JSON.parse(row.payload) as unknown;
+        if (
+          parsed !== null &&
+          typeof parsed === "object" &&
+          typeof (parsed as Record<string, unknown>).type === "string"
+        ) {
+          result.push(parsed as { type: string; [key: string]: unknown });
+        }
+      } catch {
+        // skip malformed payloads
+      }
+    }
+    return result;
+  }
+
   function close(): void {
     sqlite.close();
   }
 
-  return { append, query, getMeta, setMeta, close };
+  return {
+    append,
+    query,
+    getMeta,
+    setMeta,
+    upsertWorkflowRun,
+    appendWithWorkflowUpdate,
+    getWorkflowRun,
+    getActiveWorkflowRuns,
+    getTriggerPayload,
+    getThreadEvents,
+    close,
+  };
 }

packages/daemon/src/reflex-scheduler.ts

@@ -16,6 +16,9 @@ import type { SignalBus, Unsubscribe } from "./signal-bus.js";
 /** Sends a compute message to the worker responsible for the given sense. */
 export type TriggerFn = (senseName: string) => void;
 
+/** Triggers a workflow run in response to a signal. */
+export type WorkflowTriggerFn = (workflowName: string, payload: unknown) => void;
+
 /** Per-sense mutable state tracked by the scheduler. */
 type SenseState = {
   lastComputeAt: number;
@@ -37,6 +40,7 @@ function makeSenseState(): SenseState {
 
 export type ReflexSchedulerOptions = {
   logStore?: LogStore;
+  workflowTriggerFn?: WorkflowTriggerFn;
 };
 
 /**
@@ -153,6 +157,21 @@ export function createReflexScheduler(
   }
 
   for (const reflex of config.reflexes) {
+    if (reflex.kind === "workflow") {
+      if (opts?.workflowTriggerFn !== undefined && reflex.on !== null && reflex.on.length > 0) {
+        const workflowTriggerFn = opts.workflowTriggerFn;
+        const workflowName = reflex.workflow;
+        const watchedSenses = new Set(reflex.on);
+        const unsub = bus.subscribe((signal) => {
+          if (watchedSenses.has(signal.senseId)) {
+            workflowTriggerFn(workflowName, signal.payload);
+          }
+        });
+        unsubscribers.push(unsub);
+      }
+      continue;
+    }
+
     if (reflex.kind !== "sense") continue;
     const senseReflex = reflex;
     const senseName = senseReflex.sense;

packages/daemon/src/sense-runtime.ts

@@ -1,5 +1,5 @@
-import { readFileSync, readdirSync } from "node:fs";
-import { join } from "node:path";
+import { mkdirSync, readFileSync, readdirSync } from "node:fs";
+import { dirname, join } from "node:path";
 
 import Database from "better-sqlite3";
 import { drizzle } from "drizzle-orm/better-sqlite3";
@@ -128,6 +128,7 @@ export function openSenseDb(
   let sqlite: Database.Database;
 
   try {
+    mkdirSync(dirname(dbPath), { recursive: true });
     sqlite = new Database(dbPath);
     // WAL mode for better concurrent read performance
     sqlite.pragma("journal_mode = WAL");

packages/daemon/src/workflow-manager.ts

@@ -0,0 +1,556 @@
+/**
+ * WorkflowManager — manages per-workflow worker processes, concurrency control,
+ * and thread lifecycle tracking.
+ *
+ * RFC-002 §7.1: one worker process per workflow name, reused while alive.
+ * Concurrency and overflow (drop/queue) are enforced here in the parent process.
+ */
+
+import { fork } from "node:child_process";
+import type { ChildProcess } from "node:child_process";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+
+import type { NerveConfig, WorkflowConfig } from "@uncaged/nerve-core";
+
+import type {
+  ResumeThreadMessage,
+  ShutdownMessage,
+  StartThreadMessage,
+  ThreadEventMessage,
+} from "./ipc.js";
+import { parseWorkerMessage } from "./ipc.js";
+import type { LogStore } from "./log-store.js";
+import type { WorkflowRunStatus } from "./log-store.js";
+
+export type WorkflowManager = {
+  /** Trigger a new workflow thread (called by Reflex scheduler). */
+  startWorkflow: (workflowName: string, payload: unknown) => void;
+  /** Number of currently active (running) threads for a workflow. */
+  activeCount: (workflowName: string) => number;
+  /** Number of pending queued threads waiting to run for a workflow. */
+  queueLength: (workflowName: string) => number;
+  /** Total active workflow threads across all workflows. */
+  totalActiveCount: () => number;
+  /** Update the config reference (e.g. after hot reload). Active workers are unaffected. */
+  updateConfig: (newConfig: NerveConfig) => void;
+  /**
+   * Drain active threads for a workflow, then respawn its worker process.
+   * Used for hot reload when the workflow .ts file changes.
+   * Waits up to `drainTimeoutMs` for threads to complete before force-killing.
+   */
+  drainAndRespawn: (workflowName: string, drainTimeoutMs?: number) => Promise<void>;
+  /** Gracefully shut down all workflow workers. */
+  stop: () => Promise<void>;
+};
+
+type PendingThread = {
+  runId: string;
+  payload: unknown;
+};
+
+type WorkflowState = {
+  active: Set<string>;
+  queue: PendingThread[];
+};
+
+type WorkerEntry = {
+  workflowName: string;
+  process: ChildProcess;
+  stopping: boolean;
+  /** When set, the worker is draining before a hot-reload respawn. */
+  draining: boolean;
+};
+
+// Crash respawn backoff: track crash timestamps per workflow.
+const MAX_CRASHES_IN_WINDOW = 5;
+const CRASH_WINDOW_MS = 60_000;
+
+/**
+ * Worker shutdown timeout — must stay in sync with SHUTDOWN_TIMEOUT_MS in workflow-worker.ts.
+ * The drain timeout passed to drainAndRespawn must be >= this value so the worker has
+ * enough time to finish in-flight threads before the parent force-kills it.
+ */
+const WORKER_SHUTDOWN_TIMEOUT_MS = 10_000;
+
+const DEFAULT_MAX_QUEUE = 100;
+
+function resolveWorkerScript(): string {
+  const __filename = fileURLToPath(import.meta.url);
+  const __dir = dirname(__filename);
+  return join(__dir, "workflow-worker.js");
+}
+
+function spawnWorkflowWorker(
+  nerveRoot: string,
+  workflowName: string,
+  workerScript: string,
+): ChildProcess {
+  return fork(workerScript, ["--workflow", workflowName, "--root", nerveRoot], {
+    stdio: ["ignore", "inherit", "inherit", "ipc"],
+  });
+}
+
+function sendStartThread(worker: ChildProcess, msg: StartThreadMessage): void {
+  if (worker.connected === false) return;
+  try {
+    worker.send(msg);
+  } catch {
+    // IPC channel closed between connected check and send
+  }
+}
+
+function sendShutdown(worker: ChildProcess, entry: WorkerEntry): void {
+  entry.stopping = true;
+  if (worker.connected === false) return;
+  const msg: ShutdownMessage = { type: "shutdown" };
+  try {
+    worker.send(msg);
+  } catch {
+    // IPC channel closed between connected check and send
+  }
+}
+
+function sendResumeThread(worker: ChildProcess, msg: ResumeThreadMessage): void {
+  if (worker.connected === false) return;
+  try {
+    worker.send(msg);
+  } catch {
+    // IPC channel closed between connected check and send
+  }
+}
+
+function waitForExit(child: ChildProcess, timeoutMs: number): Promise<void> {
+  return new Promise((resolve) => {
+    const timer = setTimeout(() => {
+      child.kill("SIGKILL");
+      resolve();
+    }, timeoutMs);
+    child.once("exit", () => {
+      clearTimeout(timer);
+      resolve();
+    });
+  });
+}
+
+export function createWorkflowManager(
+  nerveRoot: string,
+  initialConfig: NerveConfig,
+  logStore: LogStore,
+): WorkflowManager {
+  const workerScript = resolveWorkerScript();
+
+  const states = new Map<string, WorkflowState>();
+  const workers = new Map<string, WorkerEntry>();
+  const crashTimestamps = new Map<string, number[]>();
+  let stopped = false;
+  let config = initialConfig;
+
+  function getOrCreateState(workflowName: string): WorkflowState {
+    let state = states.get(workflowName);
+    if (state === undefined) {
+      state = { active: new Set<string>(), queue: [] };
+      states.set(workflowName, state);
+    }
+    return state;
+  }
+
+  function workflowConfig(workflowName: string): WorkflowConfig | null {
+    return config.workflows?.[workflowName] ?? null;
+  }
+
+  function toWorkflowRunStatus(eventType: string): WorkflowRunStatus | null {
+    const map: Record<string, WorkflowRunStatus> = {
+      started: "started",
+      queued: "queued",
+      completed: "completed",
+      failed: "failed",
+      crashed: "crashed",
+      dropped: "dropped",
+      interrupted: "interrupted",
+    };
+    return map[eventType] ?? null;
+  }
+
+  function logWorkflowEvent(
+    workflowName: string,
+    runId: string,
+    eventType: string,
+    payload?: unknown,
+  ): void {
+    const ts = Date.now();
+    const serialised = payload !== undefined ? JSON.stringify(payload) : null;
+    const status = toWorkflowRunStatus(eventType);
+
+    if (status !== null) {
+      logStore.upsertWorkflowRun(
+        { source: "workflow", type: eventType, refId: runId, payload: serialised, ts },
+        { runId, workflow: workflowName, status, ts },
+      );
+    } else {
+      logStore.append({
+        source: "workflow",
+        type: eventType,
+        refId: runId,
+        payload: serialised,
+        ts,
+      });
+    }
+  }
+
+  function dispatchThread(workflowName: string, runId: string, payload: unknown): void {
+    const state = getOrCreateState(workflowName);
+    state.active.add(runId);
+
+    const worker = getOrSpawnWorker(workflowName);
+    const msg: StartThreadMessage = {
+      type: "start-thread",
+      runId,
+      workflow: workflowName,
+      triggerPayload: payload,
+    };
+    sendStartThread(worker.process, msg);
+    // Store triggerPayload in the log so it can be recovered after a crash
+    logWorkflowEvent(workflowName, runId, "started", { triggerPayload: payload });
+  }
+
+  function dequeueNext(workflowName: string): void {
+    const state = states.get(workflowName);
+    if (state === undefined || state.queue.length === 0) return;
+
+    const wfConfig = workflowConfig(workflowName);
+    const concurrency = wfConfig?.concurrency ?? 1;
+
+    if (state.active.size < concurrency) {
+      const next = state.queue.shift();
+      if (next !== undefined) {
+        dispatchThread(workflowName, next.runId, next.payload);
+      }
+    }
+  }
+
+  function handleThreadEvent(workflowName: string, msg: ThreadEventMessage): void {
+    const state = states.get(workflowName);
+    if (state === undefined) return;
+
+    if (msg.eventType === "completed" || msg.eventType === "failed") {
+      state.active.delete(msg.runId);
+      dequeueNext(workflowName);
+    }
+
+    if (msg.eventType === "completed" || msg.eventType === "failed") {
+      logWorkflowEvent(workflowName, msg.runId, msg.eventType, msg.payload);
+    }
+  }
+
+  function recoverQueuedRun(workflowName: string, runId: string, state: WorkflowState): void {
+    if (state.queue.some((q) => q.runId === runId)) return;
+    const triggerPayload = logStore.getTriggerPayload(runId);
+    state.queue.push({ runId, payload: triggerPayload });
+    process.stderr.write(
+      `[workflow-manager] crash-recovery: re-queued thread "${runId}" for "${workflowName}"\n`,
+    );
+  }
+
+  function recoverStartedRun(
+    workflowName: string,
+    runId: string,
+    state: WorkflowState,
+    worker: WorkerEntry,
+  ): void {
+    if (state.active.has(runId)) return;
+    const events = logStore.getThreadEvents(runId);
+    const triggerPayload = logStore.getTriggerPayload(runId);
+    state.active.add(runId);
+    const msg: ResumeThreadMessage = {
+      type: "resume-thread",
+      runId,
+      events,
+      triggerPayload,
+    };
+    sendResumeThread(worker.process, msg);
+    process.stderr.write(
+      `[workflow-manager] crash-recovery: resuming thread "${runId}" for "${workflowName}" (${events.length} events)\n`,
+    );
+  }
+
+  function recoverThreadsForWorker(workflowName: string, worker: WorkerEntry): void {
+    const activeRuns = logStore.getActiveWorkflowRuns(workflowName);
+    const state = getOrCreateState(workflowName);
+
+    for (const run of activeRuns) {
+      if (run.status === "queued") {
+        recoverQueuedRun(workflowName, run.runId, state);
+      } else if (run.status === "started") {
+        recoverStartedRun(workflowName, run.runId, state, worker);
+      }
+    }
+  }
+
+  function recordCrashAndCheckLimit(workflowName: string): boolean {
+    const now = Date.now();
+    const timestamps = (crashTimestamps.get(workflowName) ?? []).filter(
+      (t) => now - t < CRASH_WINDOW_MS,
+    );
+    timestamps.push(now);
+    crashTimestamps.set(workflowName, timestamps);
+    return timestamps.length > MAX_CRASHES_IN_WINDOW;
+  }
+
+  function handleWorkerCrash(workflowName: string): void {
+    const state = states.get(workflowName);
+    if (state === undefined) return;
+
+    const crashedCount = state.active.size;
+    if (crashedCount > 0) {
+      process.stderr.write(
+        `[workflow-manager] worker for "${workflowName}" crashed with ${crashedCount} active thread(s)\n`,
+      );
+      for (const runId of state.active) {
+        logWorkflowEvent(workflowName, runId, "crashed");
+      }
+    }
+
+    state.active.clear();
+    workers.delete(workflowName);
+
+    if (stopped || workflowConfig(workflowName) === null) return;
+
+    if (recordCrashAndCheckLimit(workflowName)) {
+      const count = crashTimestamps.get(workflowName)?.length ?? 0;
+      process.stderr.write(
+        `[workflow-manager] worker for "${workflowName}" crashed ${count} times in ${CRASH_WINDOW_MS}ms — stopping respawn\n`,
+      );
+      return;
+    }
+
+    process.stderr.write(
+      `[workflow-manager] respawning worker for "${workflowName}" after crash\n`,
+    );
+    const newWorker = getOrSpawnWorker(workflowName);
+    setImmediate(() => {
+      recoverThreadsForWorker(workflowName, newWorker);
+    });
+  }
+
+  function handleWorkerMessage(workflowName: string, raw: unknown): void {
+    const result = parseWorkerMessage(raw);
+    if (!result.ok) {
+      process.stderr.write(
+        `[workflow-manager] invalid message from "${workflowName}" worker: ${result.error.message}\n`,
+      );
+      return;
+    }
+    const msg = result.value;
+
+    if (msg.type === "thread-event") {
+      handleThreadEvent(workflowName, msg);
+      return;
+    }
+
+    if (msg.type === "thread-command-event") {
+      logStore.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: msg.runId,
+        payload: JSON.stringify(msg.event),
+        ts: Date.now(),
+      });
+      return;
+    }
+
+    if (msg.type === "workflow-error") {
+      process.stderr.write(
+        `[workflow-manager] workflow-error for runId "${msg.runId}" in "${workflowName}": ${msg.error}\n`,
+      );
+      const state = states.get(workflowName);
+      if (state !== undefined) {
+        state.active.delete(msg.runId);
+        dequeueNext(workflowName);
+      }
+      logWorkflowEvent(workflowName, msg.runId, "failed", { error: msg.error });
+      return;
+    }
+
+    if (msg.type === "error") {
+      process.stderr.write(
+        `[workflow-manager] error from "${workflowName}" worker: ${msg.error}\n`,
+      );
+    }
+  }
+
+  function markActiveRunsInterrupted(workflowName: string): void {
+    const state = states.get(workflowName);
+    if (state === undefined) return;
+    for (const runId of state.active) {
+      logWorkflowEvent(workflowName, runId, "interrupted");
+    }
+    state.active.clear();
+  }
+
+  function handleWorkerExit(workflowName: string, code: number | null): void {
+    const entry = workers.get(workflowName);
+    if (entry?.draining) {
+      workers.delete(workflowName);
+      markActiveRunsInterrupted(workflowName);
+      if (!stopped && workflowConfig(workflowName) !== null) {
+        process.stderr.write(
+          `[workflow-manager] worker for "${workflowName}" drained, respawning\n`,
+        );
+        getOrSpawnWorker(workflowName);
+      }
+      return;
+    }
+    if (entry?.stopping) {
+      workers.delete(workflowName);
+      const state = states.get(workflowName);
+      if (state !== undefined) {
+        state.active.clear();
+      }
+      return;
+    }
+    process.stderr.write(
+      `[workflow-manager] worker for "${workflowName}" exited with code ${code ?? "null"}\n`,
+    );
+    handleWorkerCrash(workflowName);
+  }
+
+  function getOrSpawnWorker(workflowName: string): WorkerEntry {
+    const existing = workers.get(workflowName);
+    if (existing !== undefined && existing.process.exitCode === null) {
+      return existing;
+    }
+
+    const child = spawnWorkflowWorker(nerveRoot, workflowName, workerScript);
+
+    child.on("message", (raw: unknown) => {
+      handleWorkerMessage(workflowName, raw);
+    });
+
+    child.on("exit", (code) => {
+      handleWorkerExit(workflowName, code);
+    });
+
+    const entry: WorkerEntry = { workflowName, process: child, stopping: false, draining: false };
+    workers.set(workflowName, entry);
+    return entry;
+  }
+
+  function startWorkflow(workflowName: string, payload: unknown): void {
+    if (stopped) return;
+
+    const wfConfig = workflowConfig(workflowName);
+    if (wfConfig === null) {
+      process.stderr.write(
+        `[workflow-manager] startWorkflow: unknown workflow "${workflowName}"\n`,
+      );
+      return;
+    }
+
+    const state = getOrCreateState(workflowName);
+    const runId = crypto.randomUUID();
+
+    if (state.active.size < wfConfig.concurrency) {
+      dispatchThread(workflowName, runId, payload);
+      return;
+    }
+
+    // At concurrency limit — apply overflow policy
+    if (wfConfig.overflow === "drop") {
+      logWorkflowEvent(workflowName, runId, "dropped");
+      process.stderr.write(
+        `[workflow-manager] dropped thread for "${workflowName}" (overflow=drop)\n`,
+      );
+      return;
+    }
+
+    // overflow === "queue"
+    const maxQueue = (wfConfig as { maxQueue?: number }).maxQueue ?? DEFAULT_MAX_QUEUE;
+    if (state.queue.length >= maxQueue) {
+      const oldest = state.queue.shift();
+      if (oldest !== undefined) {
+        process.stderr.write(
+          `[workflow-manager] queue overflow for "${workflowName}", dropping oldest runId "${oldest.runId}"\n`,
+        );
+        logWorkflowEvent(workflowName, oldest.runId, "dropped");
+      }
+    }
+
+    state.queue.push({ runId, payload });
+    logWorkflowEvent(workflowName, runId, "queued");
+    process.stderr.write(
+      `[workflow-manager] queued thread for "${workflowName}" runId "${runId}" (queue length: ${state.queue.length})\n`,
+    );
+  }
+
+  function activeCount(workflowName: string): number {
+    return states.get(workflowName)?.active.size ?? 0;
+  }
+
+  function queueLength(workflowName: string): number {
+    return states.get(workflowName)?.queue.length ?? 0;
+  }
+
+  function totalActiveCount(): number {
+    let total = 0;
+    for (const state of states.values()) {
+      total += state.active.size;
+    }
+    return total;
+  }
+
+  function updateConfig(newConfig: NerveConfig): void {
+    config = newConfig;
+  }
+
+  /**
+   * Default drain timeout must be at least WORKER_SHUTDOWN_TIMEOUT_MS so the worker
+   * has enough time to finish in-flight threads before the parent force-kills it.
+   */
+  const DEFAULT_DRAIN_TIMEOUT_MS = Math.max(30_000, WORKER_SHUTDOWN_TIMEOUT_MS + 5_000);
+
+  async function drainAndRespawn(
+    workflowName: string,
+    drainTimeoutMs: number = DEFAULT_DRAIN_TIMEOUT_MS,
+  ): Promise<void> {
+    const entry = workers.get(workflowName);
+    if (entry === undefined) {
+      // No active worker — nothing to drain
+      return;
+    }
+
+    entry.draining = true;
+    // Send shutdown without setting stopping=true (so the exit handler uses the draining branch)
+    if (entry.process.connected) {
+      const msg: ShutdownMessage = { type: "shutdown" };
+      try {
+        entry.process.send(msg);
+      } catch {
+        // IPC closed
+      }
+    }
+    await waitForExit(entry.process, drainTimeoutMs);
+    // The exit handler (draining branch) will respawn the worker automatically
+  }
+
+  async function stop(): Promise<void> {
+    stopped = true;
+    const exitPromises: Promise<void>[] = [];
+    for (const entry of workers.values()) {
+      sendShutdown(entry.process, entry);
+      exitPromises.push(waitForExit(entry.process, 5000));
+    }
+    await Promise.all(exitPromises);
+    workers.clear();
+  }
+
+  return {
+    startWorkflow,
+    activeCount,
+    queueLength,
+    totalActiveCount,
+    updateConfig,
+    drainAndRespawn,
+    stop,
+  };
+}

packages/daemon/src/workflow-worker.ts

@@ -0,0 +1,341 @@
+/**
+ * Workflow Worker runtime bootstrap.
+ *
+ * Entry point for a forked workflow worker process.
+ * Receives the workflow name and nerve root via CLI args, dynamically imports
+ * the user's WorkflowDefinition, then signals ready and enters the IPC event loop.
+ *
+ * Layout assumptions (nerve user config at `<nerveRoot>/`):
+ *   workflows/<name>/index.ts  (or .js)  ← user workflow definition
+ */
+
+import { existsSync } from "node:fs";
+import { join, resolve } from "node:path";
+
+import type {
+  CommandEvent,
+  ThreadState,
+  WorkflowContext,
+  WorkflowDefinition,
+} from "@uncaged/nerve-core";
+
+import type { ThreadCommandEventMessage, ThreadEventType, WorkerToParentMessage } from "./ipc.js";
+import { parseParentMessage } from "./ipc.js";
+
+// ---------------------------------------------------------------------------
+// IPC helpers
+// ---------------------------------------------------------------------------
+
+function send(msg: WorkerToParentMessage): void {
+  if (process.send) {
+    process.send(msg);
+  }
+}
+
+function sendReady(): void {
+  send({ type: "ready" });
+}
+
+function sendThreadEvent(runId: string, eventType: ThreadEventType, payload: unknown): void {
+  send({ type: "thread-event", runId, eventType, payload });
+}
+
+function sendWorkflowError(runId: string, error: string): void {
+  send({ type: "workflow-error", runId, error });
+}
+
+function sendCommandEvent(runId: string, event: CommandEvent): void {
+  const msg: ThreadCommandEventMessage = {
+    type: "thread-command-event",
+    runId,
+    event: event as { type: string; [key: string]: unknown },
+  };
+  send(msg);
+}
+
+// ---------------------------------------------------------------------------
+// Thread loop (RFC-002 §5.4)
+// ---------------------------------------------------------------------------
+
+/**
+ * Replay persisted events through moderate() to reconstruct ThreadState,
+ * then execute the next role and return the resulting CommandEvent.
+ * Returns null if the thread is already complete (moderate returned null).
+ */
+async function replayAndResume(
+  def: WorkflowDefinition,
+  runId: string,
+  ctx: WorkflowContext,
+  state: ThreadState,
+  resumeEvents: CommandEvent[],
+): Promise<CommandEvent | null> {
+  let lastNext: ReturnType<typeof def.moderate> = null;
+  for (const ev of resumeEvents) {
+    state.events.push(ev);
+    lastNext = def.moderate(state, ev);
+    if (lastNext === null) {
+      sendThreadEvent(runId, "completed", null);
+      return null;
+    }
+  }
+
+  const next = lastNext;
+  if (next === null) {
+    sendThreadEvent(runId, "completed", null);
+    return null;
+  }
+
+  const role = def.roles[next.role];
+  if (!role) {
+    sendWorkflowError(runId, `Unknown role: ${next.role}`);
+    return null;
+  }
+
+  try {
+    const event = await role.execute(next.prompt, ctx);
+    sendCommandEvent(runId, event);
+    return event;
+  } catch (e: unknown) {
+    const errMsg = e instanceof Error ? e.message : String(e);
+    sendThreadEvent(runId, "failed", { error: errMsg });
+    return null;
+  }
+}
+
+async function runThread(
+  def: WorkflowDefinition,
+  workflowName: string,
+  runId: string,
+  triggerPayload: unknown,
+  /** Pre-existing event history for crash-recovery resume. Empty for a fresh thread. */
+  resumeEvents: CommandEvent[] = [],
+): Promise<void> {
+  const state: ThreadState = { runId, events: [] };
+  const ctx: WorkflowContext = {
+    runId,
+    workflowName,
+    log: (msg) => sendThreadEvent(runId, "step_complete", { message: msg }),
+  };
+
+  const initialEvent: CommandEvent = {
+    type: "thread_start",
+    triggerPayload:
+      triggerPayload != null && typeof triggerPayload === "object" ? triggerPayload : {},
+  };
+
+  // On resume: replay persisted events, run the next un-executed role, then continue.
+  if (resumeEvents.length > 0) {
+    const nextEvent = await replayAndResume(def, runId, ctx, state, resumeEvents);
+    if (nextEvent === null) return;
+    await continueThread(def, runId, ctx, state, nextEvent);
+    return;
+  }
+
+  // Fresh thread — send the initial command event and enter the loop.
+  sendCommandEvent(runId, initialEvent);
+  await continueThread(def, runId, ctx, state, initialEvent);
+}
+
+async function continueThread(
+  def: WorkflowDefinition,
+  runId: string,
+  ctx: WorkflowContext,
+  state: ThreadState,
+  firstEvent: CommandEvent,
+): Promise<void> {
+  let event = firstEvent;
+
+  const MAX_STEPS = 1000;
+  let step = 0;
+  while (step < MAX_STEPS) {
+    step++;
+    state.events.push(event);
+    const next = def.moderate(state, event);
+
+    if (next === null) {
+      sendThreadEvent(runId, "completed", null);
+      return;
+    }
+
+    const role = def.roles[next.role];
+    if (!role) {
+      sendWorkflowError(runId, `Unknown role: ${next.role}`);
+      return;
+    }
+
+    try {
+      event = await role.execute(next.prompt, ctx);
+    } catch (e: unknown) {
+      const errMsg = e instanceof Error ? e.message : String(e);
+      sendThreadEvent(runId, "failed", { error: errMsg });
+      return;
+    }
+    sendCommandEvent(runId, event);
+  }
+  if (step >= MAX_STEPS) {
+    sendWorkflowError(runId, `Thread exceeded maximum steps (${MAX_STEPS})`);
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Workflow definition loader
+// ---------------------------------------------------------------------------
+
+async function loadWorkflowDefinition(
+  nerveRoot: string,
+  workflowName: string,
+): Promise<WorkflowDefinition> {
+  const candidates = [
+    resolve(join(nerveRoot, "workflows", workflowName, "index.ts")),
+    resolve(join(nerveRoot, "workflows", workflowName, "index.js")),
+  ];
+
+  const indexPath = candidates.find((p) => existsSync(p));
+  if (!indexPath) {
+    throw new Error(
+      `Workflow definition not found for "${workflowName}". Tried:\n${candidates.map((p) => `  ${p}`).join("\n")}`,
+    );
+  }
+
+  const mod = await import(indexPath);
+  const def: unknown = mod.default ?? mod;
+
+  if (
+    def === null ||
+    typeof def !== "object" ||
+    typeof (def as WorkflowDefinition).moderate !== "function" ||
+    typeof (def as WorkflowDefinition).roles !== "object"
+  ) {
+    throw new Error(
+      `Workflow "${workflowName}" must export a WorkflowDefinition with "roles" and "moderate".`,
+    );
+  }
+
+  return def as WorkflowDefinition;
+}
+
+// ---------------------------------------------------------------------------
+// IPC message dispatch
+// ---------------------------------------------------------------------------
+
+function handleMessage(
+  raw: unknown,
+  def: WorkflowDefinition,
+  workflowName: string,
+  inFlight: Map<string, Promise<void>>,
+  shuttingDown: { value: boolean },
+): void {
+  const parseResult = parseParentMessage(raw);
+  if (!parseResult.ok) {
+    process.stderr.write(`[workflow-worker] Invalid IPC message: ${parseResult.error.message}\n`);
+    return;
+  }
+
+  const msg = parseResult.value;
+
+  if (msg.type === "shutdown") {
+    shuttingDown.value = true;
+    const timeout = new Promise<void>((r) => setTimeout(r, 10_000));
+    Promise.race([Promise.all(inFlight.values()), timeout])
+      .then(() => process.exit(0))
+      .catch(() => process.exit(1));
+    return;
+  }
+
+  if (msg.type === "start-thread") {
+    if (shuttingDown.value) return;
+    const { runId, triggerPayload } = msg;
+
+    const previous = inFlight.get(runId) ?? Promise.resolve();
+    const next = previous
+      .then(() => runThread(def, workflowName, runId, triggerPayload))
+      .catch((e: unknown) => {
+        const errMsg = e instanceof Error ? e.message : String(e);
+        sendWorkflowError(runId, errMsg);
+      })
+      .finally(() => {
+        inFlight.delete(runId);
+      });
+
+    inFlight.set(runId, next);
+    return;
+  }
+
+  if (msg.type === "resume-thread") {
+    if (shuttingDown.value) return;
+    const { runId, events, triggerPayload } = msg;
+
+    const previous = inFlight.get(runId) ?? Promise.resolve();
+    const next = previous
+      .then(() => runThread(def, workflowName, runId, triggerPayload, events as CommandEvent[]))
+      .catch((e: unknown) => {
+        const errMsg = e instanceof Error ? e.message : String(e);
+        sendWorkflowError(runId, errMsg);
+      })
+      .finally(() => {
+        inFlight.delete(runId);
+      });
+
+    inFlight.set(runId, next);
+    return;
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Bootstrap
+// ---------------------------------------------------------------------------
+
+async function bootstrap(nerveRoot: string, workflowName: string): Promise<void> {
+  let def: WorkflowDefinition;
+  try {
+    def = await loadWorkflowDefinition(nerveRoot, workflowName);
+  } catch (e: unknown) {
+    const msg = e instanceof Error ? e.message : String(e);
+    process.stderr.write(`[workflow-worker] Failed to load workflow "${workflowName}": ${msg}\n`);
+    process.exit(1);
+  }
+
+  const inFlight = new Map<string, Promise<void>>();
+  const shuttingDown = { value: false };
+
+  sendReady();
+
+  process.on("message", (raw: unknown) => {
+    handleMessage(raw, def, workflowName, inFlight, shuttingDown);
+  });
+}
+
+// ---------------------------------------------------------------------------
+// CLI entrypoint
+// ---------------------------------------------------------------------------
+
+function parseArgs(): { nerveRoot: string; workflow: string } | null {
+  const args = process.argv.slice(2);
+  let workflow: string | null = null;
+  let nerveRoot: string | null = null;
+
+  for (let i = 0; i < args.length; i++) {
+    if (args[i] === "--workflow" && i + 1 < args.length) {
+      workflow = args[i + 1];
+      i++;
+    } else if (args[i] === "--root" && i + 1 < args.length) {
+      nerveRoot = args[i + 1];
+      i++;
+    }
+  }
+
+  if (!workflow || !nerveRoot) return null;
+  return { nerveRoot, workflow };
+}
+
+const parsed = parseArgs();
+if (!parsed) {
+  process.stderr.write("Usage: workflow-worker --workflow <name> --root <nerve-root-dir>\n");
+  process.exit(1);
+}
+
+bootstrap(parsed.nerveRoot, parsed.workflow).catch((e) => {
+  const msg = e instanceof Error ? e.message : String(e);
+  process.stderr.write(`[workflow-worker] Unhandled bootstrap error: ${msg}\n`);
+  process.exit(1);
+});