feat(workflow-utils): dryRun support for spawnSafe, cursorAgent, llmExtract

When dryRun=true, each function logs its parameters and returns a stub
result without executing any subprocess or network call. Log output is
captured by log-store for analysis.

- spawnSafe: returns { exitCode: 0, stdout: '[dryRun] skipped' }
- cursorAgent: short-circuits before spawnSafe, returns ok('[dryRun] skipped')
- llmExtract: skips fetch, returns ok({} as T)
- Tests added for spawnSafe and llmExtract dryRun paths

Fixes #104

.cursor/rules/global.mdc

@@ -0,0 +1,189 @@
+---
+description: Nerve project coding conventions — style, patterns, and toolchain
+globs: packages/*/src/**/*.ts
+alwaysApply: true
+---
+
+# Nerve Coding Conventions
+
+## Core Concepts
+
+```
+External World → Sense → Signal → Reflex → Workflow → Log
+                  ↑                                     ↑
+              "what to observe"                  "what to do"
+```
+
+**Nerve** is a lightweight observation engine daemon for autonomous agents. It continuously observes external state, reacts to changes via declarative rules, and orchestrates multi-step workflows.
+
+### Key Terms
+
+| Concept | What it is |
+|---------|-----------|
+| **Sense** | A `compute()` function that samples or derives data. Returns `T \| null` — non-null emits a Signal, null is silent. Each Sense has its own SQLite database. |
+| **Signal** | A notification emitted when a Sense returns non-null. Pure fact, no intent. Distributed via an in-memory Signal Bus. Not persisted. |
+| **Reflex** | A declarative trigger (YAML) connecting Senses to actions. Trigger types: `interval` (periodic), `on` (react to Signals). Action types: trigger a Sense, or start a Workflow. |
+| **Workflow** | A stateful multi-step execution. Contains **Roles** (actors with side effects) and a **Moderator** (pure router). Each instance is a **Thread** with a unique `runId`. |
+| **Log** | Immutable audit trail. Records executions, state transitions, errors. **Cannot trigger Reflexes** — prevents feedback loops. |
+| **Engine** | The kernel orchestrating everything. Holds Signal Bus, Reflex Scheduler, Process Manager, Workflow Manager. Never loads user code directly — all user code runs in isolated Workers. |
+| **Daemon** | The `nerve-daemon` package — engine runtime. Runs as a background process. |
+
+### Architecture Rules
+
+- **Three orthogonal extension points**: Sense (what to compute), Reflex (when to compute), Workflow (what to do)
+- **Process isolation**: One worker per Sense group (long-lived), one per Workflow type (on-demand). Workers never talk to each other.
+- **Causality is one-directional**: External world → Sense → Signal → Reflex → Action + Log. Logs are the end of the chain.
+
+
+
+## Language & Paradigm
+
+### Functional-first
+
+Use `function` + `type`, not `class` + `interface`.
+
+```typescript
+// ✅ Good
+type Signal = {
+  senseId: string;
+  value: unknown;
+  ts: number;
+};
+
+function createSignal(senseId: string, value: unknown): Signal {
+  return { senseId, value, ts: Date.now() };
+}
+
+// ❌ Bad — no class, no interface
+class Signal implements ISignal { ... }
+```
+
+### Rules
+
+| Rule | Description |
+|------|-------------|
+| `type` over `interface` | All type definitions use `type` |
+| `function` over `class` | Pure functions + closures, no class |
+| No `this` | Functions must not depend on `this` context |
+| No inheritance | No `extends`, `implements`, `abstract` |
+| Composition over inheritance | Use function composition |
+| Immutability first | Use `Readonly<T>`, `as const`, avoid mutation |
+| No optional properties | Use `T \| null` instead of `?:` — see below |
+
+### Exceptions
+
+Classes are allowed when:
+- Required by a third-party library (e.g. Drizzle's `sqliteTable`)
+- Error subclasses (`class NerveError extends Error`)
+
+### No Optional Properties
+
+Never use `?:`. All nullable fields must be explicit `T | null`.
+
+```typescript
+// ✅ Good
+type SenseConfig = {
+  group: string;
+  throttle: string | null;
+  timeout: string | null;
+};
+
+// ❌ Bad
+type SenseConfig = {
+  group: string;
+  throttle?: string;
+  timeout?: string;
+};
+```
+
+For mutually exclusive fields, use discriminated unions:
+
+```typescript
+// ✅ Good
+type ReflexConfig =
+  | { kind: "sense"; sense: string; interval: string | null; on: string[] | null }
+  | { kind: "workflow"; workflow: string; on: string[] | null };
+```
+
+## Modules & Exports
+
+- Always named exports, never default exports
+- One module = one responsibility, filename = purpose
+
+```typescript
+// ✅ Named exports only
+export function startEngine(config: EngineConfig): Engine { ... }
+export type EngineConfig = { ... };
+
+// ❌ No default exports
+export default function startEngine() { ... }
+```
+
+## Naming
+
+| Type | Style | Example |
+|------|-------|---------|
+| Files | kebab-case | `signal-bus.ts` |
+| Types | PascalCase | `SignalBus` |
+| Functions/variables | camelCase | `createSignalBus` |
+| Constants | UPPER_SNAKE | `MAX_RETRY_COUNT` |
+| Generics | Single letter or descriptive | `T`, `TValue` |
+
+## Error Handling
+
+- Use `Result` type for expected failures
+- `throw` only for unrecoverable bugs (programmer errors)
+- No try-catch for flow control
+
+```typescript
+type Result<T, E = Error> = { ok: true; value: T } | { ok: false; error: E };
+
+function parseSenseConfig(raw: unknown): Result<SenseConfig> { ... }
+```
+
+## Async
+
+- Always `async/await`, never `.then()` chains
+
+## Toolchain
+
+| Tool | Purpose |
+|------|---------|
+| **pnpm** | Package manager |
+| **TypeScript** | Type checking (strict mode) |
+| **Biome** | Lint + format (replaces ESLint + Prettier) |
+| **tsup** | Bundling |
+
+### Commands
+
+```bash
+pnpm run check      # biome check (lint + format)
+pnpm run format     # biome format --write
+pnpm run build      # full build
+pnpm test           # run tests
+```
+
+## Monorepo Structure
+
+```
+nerve/
+  packages/
+    core/           # @nerve/core — shared types and utils
+    cli/            # @nerve/cli — CLI entry point
+    daemon/         # @nerve/daemon — engine runtime
+  docs/             # RFCs, conventions
+  biome.json        # root Biome config
+  tsconfig.json     # root TypeScript config (composite project references)
+```
+
+- `core` is the shared layer; `cli` and `daemon` both depend on it
+- `cli` and `daemon` must NOT depend on each other
+
+## Commit Convention
+
+```
+<type>(<scope>): <description>
+
+type: feat | fix | refactor | docs | chore | test
+scope: core | cli | daemon | rfc-001 | ...
+```

.github/copilot-instructions.md

@@ -0,0 +1,180 @@
+# Nerve Coding Conventions
+
+## Core Concepts
+
+```
+External World → Sense → Signal → Reflex → Workflow → Log
+                  ↑                                     ↑
+              "what to observe"                  "what to do"
+```
+
+**Nerve** is a lightweight observation engine daemon for autonomous agents. It continuously observes external state, reacts to changes via declarative rules, and orchestrates multi-step workflows.
+
+### Key Terms
+
+| Concept | What it is |
+|---------|-----------|
+| **Sense** | A `compute()` function that samples or derives data. Returns `T \| null` — non-null emits a Signal, null is silent. Each Sense has its own SQLite database. |
+| **Signal** | A notification emitted when a Sense returns non-null. Pure fact, no intent. Distributed via an in-memory Signal Bus. Not persisted. |
+| **Reflex** | A declarative trigger (YAML) connecting Senses to actions. Trigger types: `interval` (periodic), `on` (react to Signals). Action types: trigger a Sense, or start a Workflow. |
+| **Workflow** | A stateful multi-step execution. Contains **Roles** (actors with side effects) and a **Moderator** (pure router). Each instance is a **Thread** with a unique `runId`. |
+| **Log** | Immutable audit trail. Records executions, state transitions, errors. **Cannot trigger Reflexes** — prevents feedback loops. |
+| **Engine** | The kernel orchestrating everything. Holds Signal Bus, Reflex Scheduler, Process Manager, Workflow Manager. Never loads user code directly — all user code runs in isolated Workers. |
+| **Daemon** | The `nerve-daemon` package — engine runtime. Runs as a background process. |
+
+### Architecture Rules
+
+- **Three orthogonal extension points**: Sense (what to compute), Reflex (when to compute), Workflow (what to do)
+- **Process isolation**: One worker per Sense group (long-lived), one per Workflow type (on-demand). Workers never talk to each other.
+- **Causality is one-directional**: External world → Sense → Signal → Reflex → Action + Log. Logs are the end of the chain.
+
+
+
+## Language & Paradigm
+
+### Functional-first
+
+Use `function` + `type`, not `class` + `interface`.
+
+```typescript
+// ✅ Good
+type Signal = {
+  senseId: string;
+  value: unknown;
+  ts: number;
+};
+
+function createSignal(senseId: string, value: unknown): Signal {
+  return { senseId, value, ts: Date.now() };
+}
+
+// ❌ Bad — no class, no interface
+class Signal implements ISignal { ... }
+```
+
+### Rules
+
+| Rule | Description |
+|------|-------------|
+| `type` over `interface` | All type definitions use `type` |
+| `function` over `class` | Pure functions + closures, no class |
+| No `this` | Functions must not depend on `this` context |
+| No inheritance | No `extends`, `implements`, `abstract` |
+| Composition over inheritance | Use function composition |
+| Immutability first | Use `Readonly<T>`, `as const`, avoid mutation |
+| No optional properties | Use `T \| null` instead of `?:` — see below |
+
+### Exceptions
+
+Classes are allowed when:
+- Required by a third-party library (e.g. Drizzle's `sqliteTable`)
+- Error subclasses (`class NerveError extends Error`)
+
+### No Optional Properties
+
+Never use `?:`. All nullable fields must be explicit `T | null`.
+
+```typescript
+// ✅ Good
+type SenseConfig = {
+  group: string;
+  throttle: string | null;
+  timeout: string | null;
+};
+
+// ❌ Bad
+type SenseConfig = {
+  group: string;
+  throttle?: string;
+  timeout?: string;
+};
+```
+
+For mutually exclusive fields, use discriminated unions:
+
+```typescript
+// ✅ Good
+type ReflexConfig =
+  | { kind: "sense"; sense: string; interval: string | null; on: string[] | null }
+  | { kind: "workflow"; workflow: string; on: string[] | null };
+```
+
+## Modules & Exports
+
+- Always named exports, never default exports
+- One module = one responsibility, filename = purpose
+
+## Naming
+
+| Type | Style | Example |
+|------|-------|---------|
+| Files | kebab-case | `signal-bus.ts` |
+| Types | PascalCase | `SignalBus` |
+| Functions/variables | camelCase | `createSignalBus` |
+| Constants | UPPER_SNAKE | `MAX_RETRY_COUNT` |
+| Generics | Single letter or descriptive | `T`, `TValue` |
+
+## Error Handling
+
+- Use `Result` type for expected failures
+- `throw` only for unrecoverable bugs (programmer errors)
+- No try-catch for flow control
+
+```typescript
+type Result<T, E = Error> = { ok: true; value: T } | { ok: false; error: E };
+```
+
+## Async
+
+- Always `async/await`, never `.then()` chains
+
+## No Dynamic Import
+
+Do NOT use `await import()` in production code. Always use static top-level `import`.
+
+Exceptions (must include a comment):
+1. `sense-runtime.ts` — user module paths known only at runtime
+2. `workflow-worker.ts` — user module paths known only at runtime
+
+Test files (`__tests__/**`) are exempt.
+
+## Toolchain
+
+| Tool | Purpose |
+|------|---------|
+| **pnpm** | Package manager |
+| **TypeScript** | Type checking (strict mode) |
+| **Biome** | Lint + format (replaces ESLint + Prettier) |
+| **tsup** | Bundling |
+
+### Commands
+
+```bash
+pnpm run check      # biome check (lint + format)
+pnpm run format     # biome format --write
+pnpm run build      # full build
+pnpm test           # run tests
+```
+
+## Monorepo Structure
+
+```
+nerve/
+  packages/
+    core/           # @nerve/core — shared types and utils
+    cli/            # @nerve/cli — CLI entry point
+    daemon/         # @nerve/daemon — engine runtime
+  docs/             # RFCs, conventions
+```
+
+- `core` is the shared layer; `cli` and `daemon` both depend on it
+- `cli` and `daemon` must NOT depend on each other
+
+## Commit Convention
+
+```
+<type>(<scope>): <description>
+
+type: feat | fix | refactor | docs | chore | test
+scope: core | cli | daemon | rfc-001 | ...
+```

.gitignore

@@ -2,3 +2,4 @@ node_modules
 dist
 .turbo
 *.tsbuildinfo
+*.tgz

.husky/pre-push

@@ -0,0 +1,3 @@
+#!/bin/sh
+pnpm check
+pnpm -r test

CLAUDE.md

@@ -0,0 +1,180 @@
+# Nerve Coding Conventions
+
+## Core Concepts
+
+```
+External World → Sense → Signal → Reflex → Workflow → Log
+                  ↑                                     ↑
+              "what to observe"                  "what to do"
+```
+
+**Nerve** is a lightweight observation engine daemon for autonomous agents. It continuously observes external state, reacts to changes via declarative rules, and orchestrates multi-step workflows.
+
+### Key Terms
+
+| Concept | What it is |
+|---------|-----------|
+| **Sense** | A `compute()` function that samples or derives data. Returns `T \| null` — non-null emits a Signal, null is silent. Each Sense has its own SQLite database. |
+| **Signal** | A notification emitted when a Sense returns non-null. Pure fact, no intent. Distributed via an in-memory Signal Bus. Not persisted. |
+| **Reflex** | A declarative trigger (YAML) connecting Senses to actions. Trigger types: `interval` (periodic), `on` (react to Signals). Action types: trigger a Sense, or start a Workflow. |
+| **Workflow** | A stateful multi-step execution. Contains **Roles** (actors with side effects) and a **Moderator** (pure router). Each instance is a **Thread** with a unique `runId`. |
+| **Log** | Immutable audit trail. Records executions, state transitions, errors. **Cannot trigger Reflexes** — prevents feedback loops. |
+| **Engine** | The kernel orchestrating everything. Holds Signal Bus, Reflex Scheduler, Process Manager, Workflow Manager. Never loads user code directly — all user code runs in isolated Workers. |
+| **Daemon** | The `nerve-daemon` package — engine runtime. Runs as a background process. |
+
+### Architecture Rules
+
+- **Three orthogonal extension points**: Sense (what to compute), Reflex (when to compute), Workflow (what to do)
+- **Process isolation**: One worker per Sense group (long-lived), one per Workflow type (on-demand). Workers never talk to each other.
+- **Causality is one-directional**: External world → Sense → Signal → Reflex → Action + Log. Logs are the end of the chain.
+
+
+
+## Language & Paradigm
+
+### Functional-first
+
+Use `function` + `type`, not `class` + `interface`.
+
+```typescript
+// ✅ Good
+type Signal = {
+  senseId: string;
+  value: unknown;
+  ts: number;
+};
+
+function createSignal(senseId: string, value: unknown): Signal {
+  return { senseId, value, ts: Date.now() };
+}
+
+// ❌ Bad — no class, no interface
+class Signal implements ISignal { ... }
+```
+
+### Rules
+
+| Rule | Description |
+|------|-------------|
+| `type` over `interface` | All type definitions use `type` |
+| `function` over `class` | Pure functions + closures, no class |
+| No `this` | Functions must not depend on `this` context |
+| No inheritance | No `extends`, `implements`, `abstract` |
+| Composition over inheritance | Use function composition |
+| Immutability first | Use `Readonly<T>`, `as const`, avoid mutation |
+| No optional properties | Use `T \| null` instead of `?:` — see below |
+
+### Exceptions
+
+Classes are allowed when:
+- Required by a third-party library (e.g. Drizzle's `sqliteTable`)
+- Error subclasses (`class NerveError extends Error`)
+
+### No Optional Properties
+
+Never use `?:`. All nullable fields must be explicit `T | null`.
+
+```typescript
+// ✅ Good
+type SenseConfig = {
+  group: string;
+  throttle: string | null;
+  timeout: string | null;
+};
+
+// ❌ Bad
+type SenseConfig = {
+  group: string;
+  throttle?: string;
+  timeout?: string;
+};
+```
+
+For mutually exclusive fields, use discriminated unions:
+
+```typescript
+// ✅ Good
+type ReflexConfig =
+  | { kind: "sense"; sense: string; interval: string | null; on: string[] | null }
+  | { kind: "workflow"; workflow: string; on: string[] | null };
+```
+
+## Modules & Exports
+
+- Always named exports, never default exports
+- One module = one responsibility, filename = purpose
+
+## Naming
+
+| Type | Style | Example |
+|------|-------|---------|
+| Files | kebab-case | `signal-bus.ts` |
+| Types | PascalCase | `SignalBus` |
+| Functions/variables | camelCase | `createSignalBus` |
+| Constants | UPPER_SNAKE | `MAX_RETRY_COUNT` |
+| Generics | Single letter or descriptive | `T`, `TValue` |
+
+## Error Handling
+
+- Use `Result` type for expected failures
+- `throw` only for unrecoverable bugs (programmer errors)
+- No try-catch for flow control
+
+```typescript
+type Result<T, E = Error> = { ok: true; value: T } | { ok: false; error: E };
+```
+
+## Async
+
+- Always `async/await`, never `.then()` chains
+
+## No Dynamic Import
+
+Do NOT use `await import()` in production code. Always use static top-level `import`.
+
+Exceptions (must include a comment):
+1. `sense-runtime.ts` — user module paths known only at runtime
+2. `workflow-worker.ts` — user module paths known only at runtime
+
+Test files (`__tests__/**`) are exempt.
+
+## Toolchain
+
+| Tool | Purpose |
+|------|---------|
+| **pnpm** | Package manager |
+| **TypeScript** | Type checking (strict mode) |
+| **Biome** | Lint + format (replaces ESLint + Prettier) |
+| **tsup** | Bundling |
+
+### Commands
+
+```bash
+pnpm run check      # biome check (lint + format)
+pnpm run format     # biome format --write
+pnpm run build      # full build
+pnpm test           # run tests
+```
+
+## Monorepo Structure
+
+```
+nerve/
+  packages/
+    core/           # @nerve/core — shared types and utils
+    cli/            # @nerve/cli — CLI entry point
+    daemon/         # @nerve/daemon — engine runtime
+  docs/             # RFCs, conventions
+```
+
+- `core` is the shared layer; `cli` and `daemon` both depend on it
+- `cli` and `daemon` must NOT depend on each other
+
+## Commit Convention
+
+```
+<type>(<scope>): <description>
+
+type: feat | fix | refactor | docs | chore | test
+scope: core | cli | daemon | rfc-001 | ...
+```

README.md

@@ -1,3 +1,189 @@
 # nerve
 
-Observation engine — Sense, Reflex, Workflow
+**Observation engine for autonomous agents** — sense the world, react to changes, run workflows.
+
+Nerve is a lightweight daemon that continuously observes external state through **Senses**, reacts via declarative **Reflexes**, and orchestrates multi-step **Workflows**. Built for the [Uncaged](https://github.com/uncaged) agent framework.
+
+## Core Concepts
+
+```
+External World → Sense ─┬→ Signal → Reflex → Sense (scheduled compute)
+                        │
+                        └→ Workflow (Sense return with workflow directive) → Log
+```
+
+| Concept | Metaphor | Role |
+|---------|----------|------|
+| **Sense** | 👁️ Perception | A `compute()` function that samples or derives data. Each sense has its own SQLite database. |
+| **Reflex** | ⚡ Reaction | Declarative rules that **only schedule Sense computes** (interval and/or `on` signal names). Reflex YAML cannot reference workflows. |
+| **Signal** | 📡 Notification | Emitted when a sense returns a non-null value that is routed as a normal signal (see Sense → Workflow below). Other reflexes can listen via `on`. |
+| **Workflow** | 🔧 Action | Stateful multi-step execution with Roles and a Moderator. Started from a Sense return value or from CLI/daemon IPC—not from reflex YAML. |
+| **Log** | 📝 Record | Immutable audit trail. Queryable by senses, but **cannot** trigger reflexes (prevents feedback loops). |
+
+**Sense → Workflow:** if `compute()` returns a plain object with a string field `workflow` in the form `name|maxRounds|prompt` (only the first two `|` delimit name and rounds; the rest is the prompt), the engine starts that workflow and **does not** emit a Signal for that return. `workflow: null` or `""` means “emit a signal” and strip the key from the payload. Invalid `workflow` strings are treated like a normal signal (directive stripped). See `@uncaged/nerve-core` `routeSenseComputeOutput` / `parseSenseWorkflowDirective`.
+
+Three extension points for **what / when / multi-step action** — reflexes never replace Sense-driven workflow launches.
+
+## Packages
+
+| Package | Description |
+|---------|-------------|
+| [`@uncaged/nerve-core`](./packages/core) | Shared types, config parser, Sense→workflow routing, daemon IPC protocol |
+| [`@uncaged/nerve-store`](./packages/store) | Append-only log SQLite, JSONL archive, CAS blob store, workflow run rows |
+| [`@uncaged/nerve-daemon`](./packages/daemon) | Kernel, workers, signal bus, reflex scheduler, workflow manager, file watcher, IPC |
+| [`@uncaged/nerve-cli`](./packages/cli) | CLI (`nerve`) — init, validate, daemon, dev, logs, sense, store, workflow |
+
+## Quick Start
+
+```bash
+# Requirements: Node.js ≥ 22.5, pnpm
+pnpm add -g @uncaged/nerve-cli
+
+# Initialize a workspace
+mkdir my-agent && cd my-agent
+nerve init
+
+# Write a sense
+cat > senses/cpu-usage/compute.ts << 'EOF'
+export async function compute() {
+  const [load] = (await import("node:os")).loadavg();
+  return load > 2.0 ? { load } : null;  // signal only when load is high
+}
+EOF
+
+# Configure reflexes in nerve.yaml
+cat > nerve.yaml << 'EOF'
+senses:
+  cpu-usage:
+    group: system
+    throttle: 10s
+
+reflexes:
+  - kind: sense
+    sense: cpu-usage
+    interval: 30s
+EOF
+
+# Run
+nerve dev          # foreground (development)
+nerve daemon start # background (production)
+nerve status       # check health
+nerve logs         # view logs
+```
+
+## Configuration
+
+`nerve.yaml` declares senses, reflexes (sense-only), optional workflows (concurrency), and optional engine `max_rounds`:
+
+```yaml
+max_rounds: 100              # default moderator cap (e.g. CLI workflow trigger)
+
+senses:
+  cpu-usage:
+    group: system          # senses in the same group share a worker process
+    throttle: 10s          # min interval between computes
+    timeout: 30s           # max compute duration
+    grace_period: 5s       # wait before first compute after startup
+
+reflexes:
+  - kind: sense
+    sense: cpu-usage
+    interval: 30s          # periodic trigger
+    on: [disk-pressure]    # also trigger on signals from other senses
+
+workflows:
+  cleanup:
+    concurrency: 1
+    overflow: drop         # discard if already running
+  code-review:
+    concurrency: 3
+    overflow: queue
+    max_queue: 20
+```
+
+YAML must **not** include `workflow:` under `reflexes` — the parser rejects it. Declare workflows under `workflows:` and start them from Sense `compute()` or `nerve workflow trigger`.
+
+**Example — Sense starts a workflow** (`senses/disk-pressure/compute.ts`):
+
+```typescript
+export async function compute() {
+  const full = await diskNearlyFull();
+  if (!full) return null;
+  return {
+    path: "/data",
+    workflow: "cleanup|10|Disk partition nearly full", // name|maxRounds|prompt
+  };
+}
+```
+
+## Architecture
+
+```
+┌────────────────────────────────────────────────────────────────────────┐
+│  Kernel                                                                 │
+│                                                                         │
+│  ┌──────────────┐   watches nerve.yaml / senses / workflows            │
+│  │ File Watcher ├──────────────────────────────────────────┐           │
+│  └──────────────┘                                          │           │
+│  ┌──────────────┐   CLI ↔ newline JSON (trigger-workflow, │           │
+│  │ Daemon IPC   │   trigger-sense, list-senses)            │           │
+│  └──────┬───────┘                                          ▼           │
+│         │            ┌──────────┐   ┌──────────┐   ┌──────────┐        │
+│         │            │ Worker   │   │ Worker   │   │ Worker   │ (1 per│
+│         │            │ (group A)│   │ (group B)│   │ (group C)│ group) │
+│         │            │ sense-1  │   │ sense-3  │   │ sense-5  │        │
+│         │            │ sense-2  │   │ sense-4  │   │          │        │
+│         │            └────┬─────┘   └────┬─────┘   └────┬─────┘        │
+│         │                 │              │              │            │
+│         │                 └──────────────┼──────────────┘            │
+│         │                                ▼                            │
+│         │                        ┌──────────────┐                     │
+│         │                        │  Signal Bus  │                     │
+│         │                        └──────┬───────┘                     │
+│         │                               ▼                             │
+│         │                     ┌──────────────────┐                      │
+│         │                     │ Reflex Scheduler│                      │
+│         │                     └────────┬─────────┘                     │
+│         │                              ▼                              │
+│         │                    ┌───────────────────┐                    │
+│         └───────────────────►│ Workflow Manager  │──→ @uncaged/nerve-store │
+│                              └───────────────────┘    (logs.db, …)   │
+└────────────────────────────────────────────────────────────────────────┘
+```
+
+- **Worker pool** — one child process per sense group; isolation between groups.
+- **Signal Bus** — in-memory pub/sub for signal distribution.
+- **Reflex Scheduler** — interval timers + signal subscriptions, with throttle/coalesce.
+- **Workflow Manager** — concurrency (drop/queue), per-workflow workers, crash recovery.
+- **File watcher** — hot reload for config, sense modules, and workflow modules.
+- **Daemon IPC** — Unix domain socket; used by the CLI when the daemon is running.
+- **Log / blob storage** — implemented in `@uncaged/nerve-store` (WAL SQLite, JSONL archive, CAS blobs).
+
+## Tech Stack
+
+- **Zero native addons** — uses Node.js built-in `node:sqlite` (DatabaseSync)
+- **Drizzle ORM** v1.0 for sense databases
+- **rslib** (rspack) for building
+- **Biome** for formatting/linting
+- **Vitest** for testing
+- **pnpm** workspaces for monorepo management
+
+## Development
+
+```bash
+git clone https://git.shazhou.work/uncaged/nerve.git
+cd nerve
+pnpm install
+pnpm build
+pnpm -r test     # run all tests
+```
+
+## Design Documents
+
+- [RFC-001: Observation Engine](./docs/rfc-001-observation-engine.md) — Sense, Signal, Reflex model
+- [RFC-002: Workflow Engine](./docs/rfc-002-workflow-engine.md) — Stateful workflow execution
+- [Coding Conventions](./docs/coding-conventions.md)
+
+## License
+
+MIT

biome.json

@@ -19,7 +19,7 @@
   },
   "overrides": [
     {
-      "include": ["tsup.config.ts"],
+      "include": ["tsup.config.ts", "*/rslib.config.ts"],
       "linter": {
         "rules": {
           "style": {
@@ -27,6 +27,19 @@
           }
         }
       }
+    },
+    {
+      "include": ["**/__tests__/**"],
+      "linter": {
+        "rules": {
+          "suspicious": {
+            "noExplicitAny": "off"
+          },
+          "style": {
+            "noNonNullAssertion": "off"
+          }
+        }
+      }
     }
   ],
   "linter": {

package.json

@@ -1,14 +1,19 @@
 {
   "name": "nerve",
   "private": true,
+  "engines": {
+    "node": ">=22.5.0"
+  },
   "scripts": {
+    "prepare": "husky",
     "build": "pnpm -r run build",
     "check": "biome check .",
     "format": "biome format --write ."
   },
   "devDependencies": {
     "@biomejs/biome": "^1.9.0",
-    "tsup": "^8.0.0",
+    "@rslib/core": "^0.21.3",
+    "husky": "^9.1.7",
     "typescript": "^5.5.0"
   }
 }

packages/cli/README.md

@@ -0,0 +1,87 @@
+# @uncaged/nerve-cli
+
+Command-line interface for the [nerve](../../README.md) observation engine.
+
+## Install
+
+```bash
+pnpm add -g @uncaged/nerve-cli
+# or
+npx @uncaged/nerve-cli
+```
+
+Requires Node.js ≥ 22.5.
+
+## Commands
+
+### Workspace
+
+```bash
+nerve init                  # Initialize a nerve workspace (installs deps, scaffolds config)
+nerve validate              # Validate nerve.yaml configuration
+```
+
+### Daemon management
+
+```bash
+nerve daemon start          # Start the daemon (background)
+nerve daemon stop           # Stop the daemon
+nerve daemon status         # Show pid, uptime, sense names from nerve.yaml (process must exist)
+nerve daemon restart        # Stop then start
+nerve daemon logs           # Tail daemon process logs (file under workspace logs/)
+```
+
+### Development
+
+```bash
+nerve dev                   # Foreground kernel with file watcher + IPC (Ctrl+C stops)
+```
+
+### Querying & status
+
+```bash
+nerve logs                  # Tail or page the daemon text log file (path in footer; default ~/.uncaged-nerve/logs/nerve.log)
+nerve status                # Short daemon health summary (aliases daemon status)
+```
+
+Structured rows in `data/logs.db` are surfaced via **`nerve workflow inspect`** / **`nerve workflow list`** (and `LogStore` in code), not via `nerve logs`.
+
+### Sense
+
+```bash
+nerve sense list            # List senses (live fields from daemon IPC when running)
+nerve sense trigger <name>  # IPC trigger-sense — queue a compute for that sense
+nerve sense query <name>    # Read-only SQL on data/senses/<name>.db (optional SQL args)
+nerve sense schema <name>   # Print CREATE TABLE statements for that sense DB
+```
+
+### Store maintenance
+
+```bash
+nerve store archive         # Move old log rows to JSONL under data/archive/logs/… (optional --vacuum)
+```
+
+### Workflows
+
+```bash
+nerve workflow list         # Queued/started runs (add --all for terminal states; --workflow, --limit, --offset)
+nerve workflow inspect <runId>   # Run metadata + paginated workflow log lines
+nerve workflow thread <runId>    # Role rounds from persisted messages (--before, --budget)
+nerve workflow trigger <name>    # IPC trigger-workflow (daemon must be running)
+                                   # Optional JSON: --payload '{"prompt":"…","maxRounds":50}'
+```
+
+`nerve workflow trigger` sends a `trigger-workflow` line on the daemon Unix socket (same protocol as `@uncaged/nerve-core` / `parseDaemonIpcRequest`). It does not read `nerve.yaml` workflow definitions beyond what the running daemon already loaded.
+
+### Top-level aliases
+
+```bash
+nerve start    → nerve daemon start
+nerve stop     → nerve daemon stop
+nerve status   → nerve daemon status
+nerve logs     → nerve daemon logs
+```
+
+## License
+
+MIT

packages/cli/package.json

@@ -1,31 +1,33 @@
 {
   "name": "@uncaged/nerve-cli",
-  "version": "0.1.8",
+  "engines": {
+    "node": ">=22.5.0"
+  },
+  "version": "0.4.0",
   "type": "module",
   "bin": {
     "nerve": "dist/cli.js"
   },
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
-  "files": [
+  "files": ["dist"],
-    "dist"
-  ],
   "publishConfig": {
     "access": "public"
   },
   "scripts": {
     "prepublishOnly": "bash ../../scripts/prepublish-check.sh",
-    "build": "tsup",
+    "build": "rslib build",
     "test": "vitest run"
   },
   "dependencies": {
     "@uncaged/nerve-core": "workspace:*",
-    "citty": "^0.1.6"
+    "@uncaged/nerve-store": "workspace:*",
+    "citty": "^0.1.6",
+    "yaml": "^2.8.3"
   },
   "devDependencies": {
-    "@types/better-sqlite3": "^7.6.13",
+    "@rslib/core": "^0.21.3",
     "@types/node": "^22.0.0",
-    "@uncaged/nerve-daemon": "workspace:*",
     "vitest": "^4.1.5"
   }
 }

packages/cli/rslib.config.ts

@@ -0,0 +1,25 @@
+import { defineConfig } from "@rslib/core";
+
+export default defineConfig({
+  lib: [
+    {
+      format: "esm",
+      dts: true,
+      banner: {
+        js: "#!/usr/bin/env node",
+      },
+    },
+  ],
+  source: {
+    entry: {
+      index: "src/index.ts",
+      cli: "src/cli.ts",
+      "daemon-bootstrap": "src/daemon-bootstrap.ts",
+    },
+  },
+  output: {
+    target: "node",
+    cleanDistPath: true,
+    externals: ["@uncaged/nerve-daemon", "@uncaged/nerve-store"],
+  },
+});

packages/cli/src/__tests__/daemon-types.test.ts

@@ -1,80 +0,0 @@
-/**
- * Compile-time check: daemon-types.ts stays in sync with @uncaged/nerve-daemon exports.
- * If the daemon package changes its public API, this file will fail to compile.
- */
-
-import type { SenseInfo } from "@uncaged/nerve-core";
-import type {
-  ArchiveLogsDayResult as DaemonArchiveLogsDayResult,
-  ArchiveLogsOptions as DaemonArchiveLogsOptions,
-  ArchiveLogsResult as DaemonArchiveLogsResult,
-  LogEntry as DaemonLogEntry,
-  LogQuery as DaemonLogQuery,
-  LogStore as DaemonLogStore,
-  SenseInfo as DaemonSenseInfo,
-  WorkflowRun as DaemonWorkflowRun,
-  WorkflowRunStatus as DaemonWorkflowRunStatus,
-} from "@uncaged/nerve-daemon";
-import { describe, expectTypeOf, it } from "vitest";
-
-import type {
-  ArchiveLogsDayResult,
-  ArchiveLogsOptions,
-  ArchiveLogsResult,
-  LogEntry,
-  LogQuery,
-  LogStore,
-  WorkflowRun,
-  WorkflowRunStatus,
-} from "../daemon-types.js";
-
-describe("daemon-types drift guard", () => {
-  it("SenseInfo matches daemon package export (list-senses IPC)", () => {
-    expectTypeOf<SenseInfo>().toMatchTypeOf<DaemonSenseInfo>();
-    expectTypeOf<DaemonSenseInfo>().toMatchTypeOf<SenseInfo>();
-  });
-
-  it("WorkflowRunStatus is assignable both ways", () => {
-    expectTypeOf<WorkflowRunStatus>().toMatchTypeOf<DaemonWorkflowRunStatus>();
-    expectTypeOf<DaemonWorkflowRunStatus>().toMatchTypeOf<WorkflowRunStatus>();
-  });
-
-  it("WorkflowRun is assignable both ways", () => {
-    expectTypeOf<WorkflowRun>().toMatchTypeOf<DaemonWorkflowRun>();
-    expectTypeOf<DaemonWorkflowRun>().toMatchTypeOf<WorkflowRun>();
-  });
-
-  it("LogEntry is assignable both ways", () => {
-    expectTypeOf<LogEntry>().toMatchTypeOf<DaemonLogEntry>();
-    expectTypeOf<DaemonLogEntry>().toMatchTypeOf<LogEntry>();
-  });
-
-  it("LogQuery is assignable both ways", () => {
-    expectTypeOf<LogQuery>().toMatchTypeOf<DaemonLogQuery>();
-    expectTypeOf<DaemonLogQuery>().toMatchTypeOf<LogQuery>();
-  });
-
-  it("LogStore has all required methods", () => {
-    expectTypeOf<LogStore>().toMatchTypeOf<
-      Pick<
-        DaemonLogStore,
-        | "query"
-        | "getWorkflowRun"
-        | "getActiveWorkflowRuns"
-        | "getAllWorkflowRuns"
-        | "upsertWorkflowRun"
-        | "archiveLogs"
-        | "close"
-      >
-    >();
-  });
-
-  it("ArchiveLogs types match daemon", () => {
-    expectTypeOf<ArchiveLogsOptions>().toMatchTypeOf<DaemonArchiveLogsOptions>();
-    expectTypeOf<DaemonArchiveLogsOptions>().toMatchTypeOf<ArchiveLogsOptions>();
-    expectTypeOf<ArchiveLogsResult>().toMatchTypeOf<DaemonArchiveLogsResult>();
-    expectTypeOf<DaemonArchiveLogsResult>().toMatchTypeOf<ArchiveLogsResult>();
-    expectTypeOf<ArchiveLogsDayResult>().toMatchTypeOf<DaemonArchiveLogsDayResult>();
-    expectTypeOf<DaemonArchiveLogsDayResult>().toMatchTypeOf<ArchiveLogsDayResult>();
-  });
-});

packages/cli/src/__tests__/logs.test.ts

@@ -234,7 +234,11 @@ describe("logsCommand negative offset", () => {
 
   it("exits with code 1 and writes to stderr when offset is negative", async () => {
     await expect(
-      logsCommand.run!({ args: { n: "50", offset: "-5", follow: false }, rawArgs: [], cmd: logsCommand as never }),
+      logsCommand.run?.({
+        args: { n: "50", offset: "-5", follow: false },
+        rawArgs: [],
+        cmd: logsCommand as never,
+      }),
     ).rejects.toThrow("process.exit(1)");
     expect(exitCode).toBe(1);
     expect(stderrOutput).toContain("--offset must be a non-negative integer");
@@ -243,7 +247,11 @@ describe("logsCommand negative offset", () => {
 
   it("exits with code 1 for offset=-1", async () => {
     await expect(
-      logsCommand.run!({ args: { n: "10", offset: "-1", follow: false }, rawArgs: [], cmd: logsCommand as never }),
+      logsCommand.run?.({
+        args: { n: "10", offset: "-1", follow: false },
+        rawArgs: [],
+        cmd: logsCommand as never,
+      }),
     ).rejects.toThrow("process.exit(1)");
     expect(exitCode).toBe(1);
   });

packages/cli/src/__tests__/workflow.test.ts

@@ -12,19 +12,23 @@ import { createServer } from "node:net";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
-import { createLogStore } from "@uncaged/nerve-daemon";
+import { createLogStore } from "@uncaged/nerve-store";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 
+import type { LogStore, ThreadRoundRow, WorkflowRun } from "@uncaged/nerve-store";
 import {
+  DEFAULT_THREAD_BUDGET_CHARS,
   buildInspectOutput,
   buildListOutput,
+  buildThreadCommandOutput,
+  formatThreadRoundBlock,
   formatTs,
   getAllWorkflowRuns,
   parseIntArg,
+  partitionWorkflowMessage,
   statusIcon,
 } from "../commands/workflow.js";
 import { triggerWorkflowViaDaemon } from "../daemon-client.js";
-import type { LogStore, WorkflowRun } from "../daemon-types.js";
 
 // ---------------------------------------------------------------------------
 // Test helpers
@@ -322,6 +326,98 @@ describe("workflow list — integration with real store", () => {
   });
 });
 
+// ---------------------------------------------------------------------------
+// nerve workflow thread — formatting helpers
+// ---------------------------------------------------------------------------
+
+describe("partitionWorkflowMessage", () => {
+  it("extracts role, content, and meta", () => {
+    const p = partitionWorkflowMessage({
+      role: "scanner",
+      content: "ok",
+      meta: { items: [1, 2] },
+    });
+    expect(p.roleStr).toBe("scanner");
+    expect(p.contentBody).toBe("ok");
+    expect(p.meta).toEqual({ items: [1, 2] });
+  });
+
+  it("passes through role and content as-is", () => {
+    const p = partitionWorkflowMessage({
+      role: "unknown",
+      content: '{"n":1}',
+      meta: null,
+      timestamp: 0,
+    });
+    expect(p.roleStr).toBe("unknown");
+    expect(p.contentBody).toBe('{"n":1}');
+  });
+});
+
+describe("formatThreadRoundBlock", () => {
+  const row: ThreadRoundRow = {
+    round: 2,
+    logId: 99,
+    ts: new Date("2026-01-02T03:04:05.006Z").getTime(),
+    message: { role: "bot", content: "hi", meta: { score: 0.5 }, timestamp: 1735783445006 },
+  };
+
+  it("includes header, YAML frontmatter for meta, and body", () => {
+    const text = formatThreadRoundBlock(row);
+    expect(text).toContain("[#2 bot]");
+    expect(text).toContain("---\n");
+    expect(text).toContain("score: 0.5");
+    expect(text).toContain("hi");
+  });
+});
+
+describe("buildThreadCommandOutput", () => {
+  function row(n: number, content: string): ThreadRoundRow {
+    return {
+      round: n,
+      logId: 10 + n,
+      ts: 1000 + n,
+      message: { role: "r", content, meta: { extra: n }, timestamp: 1000 + n },
+    };
+  }
+
+  it("orders rounds chronologically (oldest first in output)", () => {
+    const desc = [row(3, "ccc"), row(2, "bbb"), row(1, "aaa")];
+    const prefix = ["HEADER\n"];
+    const { lines, paginationHint } = buildThreadCommandOutput(prefix, desc, 50_000, "run-x");
+    const text = lines.join("");
+    const idxA = text.indexOf("\naaa\n");
+    const idxB = text.indexOf("\nbbb\n");
+    const idxC = text.indexOf("\nccc\n");
+    expect(idxA).toBeGreaterThan(-1);
+    expect(idxB).toBeGreaterThan(idxA);
+    expect(idxC).toBeGreaterThan(idxB);
+    expect(paginationHint).toBeNull();
+  });
+
+  it("emits pagination hint with --before when oldest shown round is still > 1", () => {
+    const desc = [row(4, "d"), row(3, "c")];
+    const { paginationHint } = buildThreadCommandOutput([], desc, 50_000, "run-y");
+    expect(paginationHint).toContain("--before 3");
+    expect(paginationHint).toContain("run-y");
+  });
+
+  it("respects budget and hints with non-default --budget in command", () => {
+    const big = "y".repeat(500);
+    const desc = [row(2, big), row(1, "a")];
+    const { lines, paginationHint } = buildThreadCommandOutput([], desc, 400, "run-z");
+    const text = lines.join("");
+    expect(text).toContain("[#2");
+    expect(text).not.toContain("[#1");
+    expect(paginationHint).toContain("--before 2");
+    expect(paginationHint).toContain("--budget 400");
+  });
+
+  it("default budget constant matches workflow command default", () => {
+    expect(DEFAULT_THREAD_BUDGET_CHARS).toBe(8000);
+  });
+});
+
 // ---------------------------------------------------------------------------
 // parseIntArg
 // ---------------------------------------------------------------------------
@@ -418,7 +514,7 @@ describe("triggerWorkflowViaDaemon", () => {
     await new Promise<void>((r) => server.listen(sockPath, r));
 
     try {
-      const result = await triggerWorkflowViaDaemon(sockPath, "my-workflow", {});
+      const result = await triggerWorkflowViaDaemon(sockPath, "my-workflow", "", 100);
       expect(result).toEqual({ ok: true });
     } finally {
       await new Promise<void>((r) => server.close(() => r()));
@@ -434,7 +530,7 @@ describe("triggerWorkflowViaDaemon", () => {
     await new Promise<void>((r) => server.listen(sockPath, r));
 
     try {
-      const result = await triggerWorkflowViaDaemon(sockPath, "missing", {});
+      const result = await triggerWorkflowViaDaemon(sockPath, "missing", "", 100);
       expect(result).toEqual({ ok: false, error: "unknown workflow" });
     } finally {
       await new Promise<void>((r) => server.close(() => r()));
@@ -442,7 +538,7 @@ describe("triggerWorkflowViaDaemon", () => {
   });
 
   it("rejects when no daemon is listening on the socket", async () => {
-    await expect(triggerWorkflowViaDaemon(sockPath, "my-workflow", {})).rejects.toThrow(
+    await expect(triggerWorkflowViaDaemon(sockPath, "my-workflow", "", 100)).rejects.toThrow(
       /Cannot connect to daemon/,
     );
   });

packages/cli/src/cli.ts

@@ -12,6 +12,26 @@ import { storeCommand } from "./commands/store.js";
 import { validateCommand } from "./commands/validate.js";
 import { workflowCommand } from "./commands/workflow.js";
 
+/**
+ * Citty picks the first non-flag token as a subcommand name. Rewrite
+ * `nerve init --from <url>` so the URL is not mistaken for `workflow`/`workspace`.
+ */
+function normalizeNerveArgv(argv: string[]): string[] {
+  const initIdx = argv.indexOf("init");
+  if (initIdx === -1) return argv;
+  const tail = argv.slice(initIdx + 1);
+  const fromAt = tail.indexOf("--from");
+  if (fromAt === -1) return argv;
+  const beforeFrom = tail.slice(0, fromAt);
+  if (beforeFrom.some((a) => !a.startsWith("-"))) return argv;
+  const next = tail[fromAt + 1];
+  if (next === undefined || next.startsWith("-")) return argv;
+  const reserved = new Set(["workflow", "workspace"]);
+  if (reserved.has(next)) return argv;
+  const mergedTail = [...tail.slice(0, fromAt), `--from=${next}`, ...tail.slice(fromAt + 2)];
+  return [...argv.slice(0, initIdx + 1), ...mergedTail];
+}
+
 const main = defineCommand({
   meta: {
     name: "nerve",
@@ -32,4 +52,4 @@ const main = defineCommand({
   },
 });
 
-runMain(main);
+runMain(main, { rawArgs: normalizeNerveArgv(process.argv.slice(2)) });

packages/cli/src/commands/init.ts

@@ -1,5 +1,5 @@
-import { spawn, execFile } from "node:child_process";
+import { execFile, spawn } from "node:child_process";
-import { existsSync, mkdirSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readdirSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { promisify } from "node:util";
 
@@ -35,7 +35,7 @@ const PACKAGE_JSON = `{
     "drizzle-kit": "latest"
   },
   "pnpm": {
-    "onlyBuiltDependencies": ["better-sqlite3", "esbuild"]
+    "onlyBuiltDependencies": ["esbuild"]
   }
 }
 `;
@@ -218,20 +218,94 @@ const initWorkspaceCommand = defineCommand({
   },
 });
 
-async function tryRequireSqlite(nerveRoot: string): Promise<boolean> {
+/** Verify built-in `node:sqlite` (Node.js ≥22.5) loads in a child process. */
+async function verifyNodeSqlite(): Promise<boolean> {
   try {
-    const modulePath = join(nerveRoot, "node_modules", "better-sqlite3");
+    await execFileAsync(
-    // Use a child process to test if the native module loads
+      "node",
-    await execFileAsync("node", ["-e", `require(${JSON.stringify(modulePath)})`], {
+      [
-      cwd: nerveRoot,
+        "--input-type=module",
-      timeout: 10_000,
+        "-e",
-    });
+        "import { DatabaseSync } from 'node:sqlite'; new DatabaseSync(':memory:').exec('SELECT 1');",
+      ],
+      { timeout: 10_000 },
+    );
     return true;
   } catch {
     return false;
   }
 }
 
+function isNerveRootNonEmpty(nerveRoot: string): boolean {
+  if (!existsSync(nerveRoot)) return false;
+  return readdirSync(nerveRoot).length > 0;
+}
+
+async function runInitFromGit(url: string): Promise<void> {
+  const trimmed = url.trim();
+  if (trimmed.length === 0) {
+    process.stderr.write("❌ --from requires a non-empty git URL.\n");
+    process.exit(1);
+  }
+
+  const nerveRoot = getNerveRoot();
+  if (isNerveRootNonEmpty(nerveRoot)) {
+    process.stderr.write(
+      `❌ ${nerveRoot} already exists and is not empty. Remove it (or empty it) before using --from.\n`,
+    );
+    process.exit(1);
+  }
+
+  try {
+    await execFileAsync("git", ["--version"]);
+  } catch {
+    process.stderr.write("❌ git is not available. Install git and retry.\n");
+    process.exit(1);
+  }
+
+  try {
+    await execFileAsync("pnpm", ["--version"]);
+  } catch {
+    process.stderr.write("❌ pnpm is not available. Install pnpm and retry.\n");
+    process.exit(1);
+  }
+
+  process.stdout.write(`Cloning ${trimmed} → ${nerveRoot} …\n`);
+  try {
+    await runCommand("git", ["clone", trimmed, nerveRoot], process.cwd());
+  } catch {
+    process.stderr.write("❌ git clone failed.\n");
+    process.exit(1);
+  }
+
+  if (!existsSync(join(nerveRoot, "nerve.yaml"))) {
+    process.stdout.write(`⚠️ ${join(nerveRoot, "nerve.yaml")} not found after clone.\n`);
+  }
+  if (!existsSync(join(nerveRoot, "package.json"))) {
+    process.stdout.write(`⚠️ ${join(nerveRoot, "package.json")} not found after clone.\n`);
+  }
+
+  process.stdout.write("Installing dependencies with pnpm …\n");
+  try {
+    await runCommand("pnpm", ["install", "--no-cache"], nerveRoot);
+  } catch {
+    process.stdout.write(
+      `⚠️ pnpm install failed. Try manually:\n   cd ${nerveRoot} && pnpm install --no-cache\n`,
+    );
+  }
+
+  if (!(await verifyNodeSqlite())) {
+    process.stdout.write(
+      "⚠️ Built-in SQLite (node:sqlite) is not available in this Node.js build. " +
+        "The daemon requires Node.js 22.5 or newer with SQLite enabled.\n",
+    );
+  }
+
+  process.stdout.write(
+    `✅ Workspace cloned to ${nerveRoot}\n\n💡 Next steps:\n   1. Review nerve.yaml and install any missing tooling.\n   2. Run \`nerve start\` to launch the daemon.\n`,
+  );
+}
+
 async function runInitWorkspace(force: boolean): Promise<void> {
   const nerveRoot = getNerveRoot();
 
@@ -264,27 +338,11 @@ async function runInitWorkspace(force: boolean): Promise<void> {
     );
   }
 
-  // Verify better-sqlite3 native module — rebuild up to 2 times if broken
+  if (!(await verifyNodeSqlite())) {
-  const sqlitePath = join(nerveRoot, "node_modules", "better-sqlite3");
+    process.stdout.write(
-  if (existsSync(sqlitePath)) {
+      "⚠️ Built-in SQLite (node:sqlite) is not available in this Node.js build. " +
-    for (let attempt = 1; attempt <= 2; attempt++) {
+        "The daemon requires Node.js 22.5 or newer with SQLite enabled.\n",
-      if (await tryRequireSqlite(nerveRoot)) break;
+    );
-      process.stdout.write(
-        `${attempt === 1 ? "Building" : "Retrying build of"} native module better-sqlite3 (attempt ${attempt}/2)…\n`,
-      );
-      try {
-        await runCommand(cmd, ["rebuild", "better-sqlite3"], nerveRoot);
-      } catch {
-        // will be caught by the verify below
-      }
-    }
-    if (!(await tryRequireSqlite(nerveRoot))) {
-      process.stdout.write(
-        `⚠️ better-sqlite3 native module is not working. The daemon will fail to start.\n` +
-          `   Fix: cd ${nerveRoot} && ${cmd} rebuild better-sqlite3\n` +
-          `   Or:  npm install --build-from-source better-sqlite3\n`,
-      );
-    }
   }
 
   if (!existsSync(join(nerveRoot, ".git"))) {
@@ -306,7 +364,7 @@ export const initCommand = defineCommand({
   meta: {
     name: "init",
     description:
-      "Initialize workspace (nerve init) or scaffold templates (nerve init workflow <name>)",
+      "Initialize workspace (nerve init), clone from git (nerve init --from <url>), or scaffold templates (nerve init workflow <name>)",
   },
   args: {
     force: {
@@ -314,12 +372,21 @@ export const initCommand = defineCommand({
       description: "Reinitialize even if workspace already exists (preserves data/)",
       default: false,
     },
+    from: {
+      type: "string",
+      description: "Clone an existing git repo into ~/.uncaged-nerve instead of scaffolding",
+      required: false,
+    },
   },
   subCommands: {
     workflow: initWorkflowCommand,
     workspace: initWorkspaceCommand,
   },
   async run({ args }) {
+    if (args.from !== undefined) {
+      await runInitFromGit(String(args.from));
+      return;
+    }
     await runInitWorkspace(args.force);
   },
 });

packages/cli/src/commands/logs.ts

@@ -85,7 +85,7 @@ export function buildLogFooter(slice: LogSlice, nArg: number, logPath: string):
   let footer = `\n📄 ${rangeStr} | ${logPath}\n`;
 
   if (slice.nextOffset !== null) {
-    footer += `⏩ Earlier lines available. Fetch previous page:\n`;
+    footer += "⏩ Earlier lines available. Fetch previous page:\n";
     footer += `   nerve logs --offset ${slice.nextOffset} -n ${nArg}\n`;
   }
 

packages/cli/src/commands/sense.ts

@@ -1,13 +1,12 @@
 import { readFileSync } from "node:fs";
 import { join } from "node:path";
-import { DatabaseSync } from "node:sqlite";
+import type { DatabaseSync } from "node:sqlite";
 
-import { type SenseInfo, parseNerveConfig } from "@uncaged/nerve-core";
+import { type SenseInfo, isPlainRecord, parseNerveConfig } from "@uncaged/nerve-core";
 import { defineCommand } from "citty";
 
 import { listSensesViaDaemon, triggerSenseViaDaemon } from "../daemon-client.js";
 import {
-  assertSenseDbExists,
   defaultPreviewSql,
   formatRowsAsAlignedTable,
   listTableSqlStatements,
@@ -240,7 +239,8 @@ const senseQueryCommand = defineCommand({
         }
       }
 
-      const rows = db.prepare(sql).all() as Record<string, unknown>[];
+      const rawRows: unknown[] = db.prepare(sql).all();
+      const rows: Record<string, unknown>[] = rawRows.filter(isPlainRecord);
 
       if (args.json) {
         process.stdout.write(`${JSON.stringify(rows, null, 2)}\n`);

packages/cli/src/commands/start.ts

@@ -74,9 +74,11 @@ async function runDaemon(nerveRoot: string): Promise<void> {
 
   const bootstrapPath = daemonBootstrapScript();
 
+  // After `open`, file-backed WriteStream has a numeric OS fd for spawn stdio; `@types/node` omits `fd` on this WriteStream alias.
+  const logFd = (logStream as unknown as { fd: number }).fd;
   const child = spawn(process.execPath, [bootstrapPath], {
     detached: true,
-    stdio: ["ignore", logStream.fd, logStream.fd],
+    stdio: ["ignore", logFd, logFd],
     env: { ...process.env, NERVE_ROOT: nerveRoot },
     cwd: nerveRoot,
   });

packages/cli/src/commands/status.ts

@@ -47,7 +47,11 @@ export const statusCommand = defineCommand({
       return;
     }
 
-    const pid = readPidFile() as number;
+    const pid = readPidFile();
+    if (pid === null) {
+      process.stdout.write("😴 Nerve daemon is not running.\n");
+      return;
+    }
 
     const configPath = join(getNerveRoot(), "nerve.yaml");
     let senseList: string[] = [];

packages/cli/src/commands/workflow.ts

@@ -1,15 +1,24 @@
 import { existsSync } from "node:fs";
 import { join } from "node:path";
 
+import type { DaemonIpcTriggerResponse } from "@uncaged/nerve-core";
+import { DEFAULT_ENGINE_MAX_ROUNDS, isPlainRecord } from "@uncaged/nerve-core";
 import { defineCommand } from "citty";
+import { stringify } from "yaml";
 
+import type { LogStore, ThreadRoundRow, WorkflowRun } from "@uncaged/nerve-store";
 import { triggerWorkflowViaDaemon } from "../daemon-client.js";
-import type { LogStore, WorkflowRun } from "../daemon-types.js";
 import { loadDaemonModule } from "../workspace-daemon.js";
 import { getNerveRoot, getSocketPath, isRunning } from "../workspace.js";
 
 export const DEFAULT_PAGE_SIZE = 20;
 
+/** Default max characters for `nerve workflow thread` output (including run header). */
+export const DEFAULT_THREAD_BUDGET_CHARS = 8000;
+
+/** Max role-round rows read from SQLite per invocation (DESC by round). */
+export const THREAD_ROUNDS_FETCH_LIMIT = 8192;
+
 export function parseIntArg(raw: string, fallback: number): number {
   const v = Number.parseInt(raw, 10);
   return Number.isNaN(v) ? fallback : v;
@@ -172,6 +181,119 @@ export function buildInspectOutput(
   return { header, eventLines, paginationHint };
 }
 
+// ---------------------------------------------------------------------------
+// nerve workflow thread <runId> — agent-oriented role rounds
+// ---------------------------------------------------------------------------
+
+export type PartitionedMessage = {
+  roleStr: string;
+  contentBody: string;
+  meta: Record<string, unknown>;
+};
+
+/**
+ * Extract display fields from a WorkflowMessage-shaped object.
+ * `role` and `content` are used for header/body; `meta` is serialized as YAML frontmatter.
+ */
+export function partitionWorkflowMessage(msg: {
+  role: string;
+  content: string;
+  meta: unknown;
+  timestamp: number;
+}): PartitionedMessage {
+  const roleStr = msg.role;
+  const contentBody = msg.content;
+  const meta: Record<string, unknown> =
+    msg.meta !== null && msg.meta !== undefined && typeof msg.meta === "object"
+      ? isPlainRecord(msg.meta)
+        ? msg.meta
+        : (msg.meta as Record<string, unknown>)
+      : {};
+  return { roleStr, contentBody, meta };
+}
+
+/**
+ * One role round as plain text: header line, YAML frontmatter (meta only), body (content).
+ */
+export function formatThreadRoundBlock(row: ThreadRoundRow): string {
+  const { roleStr, contentBody, meta } = partitionWorkflowMessage(row.message);
+  const yamlBlock =
+    Object.keys(meta).length === 0 ? "{}\n" : `${stringify(meta, { lineWidth: 100 })}\n`;
+  return `[#${row.round} ${roleStr}] ${formatTs(row.ts)}\n---\n${yamlBlock}---\n${contentBody}\n\n`;
+}
+
+export type ThreadCommandOutput = {
+  lines: string[];
+  paginationHint: string | null;
+};
+
+function buildTruncatedSingleRound(
+  row: ThreadRoundRow,
+  remaining: number,
+  prefixLines: string[],
+  runId: string,
+  budgetFlag: string,
+): ThreadCommandOutput {
+  const { roleStr, contentBody, meta } = partitionWorkflowMessage(row.message);
+  const yamlBlock =
+    Object.keys(meta).length === 0 ? "{}\n" : `${stringify(meta, { lineWidth: 100 })}\n`;
+  const header = `[#${row.round} ${roleStr}] ${formatTs(row.ts)}\n---\n${yamlBlock}---\n`;
+  const maxBody = Math.max(0, remaining - header.length - "[truncated]\n".length);
+  const truncated =
+    maxBody > 0 && contentBody.length > maxBody
+      ? `${contentBody.slice(0, maxBody)}\n[truncated]\n`
+      : `${contentBody}\n[truncated]\n`;
+  const single = `${header + truncated}\n`;
+  const hintRound = row.round;
+  return {
+    lines: [...prefixLines, single],
+    paginationHint:
+      hintRound > 1
+        ? `\n⏩ Older rounds exist. Fetch with:\n   nerve workflow thread ${runId} --before ${String(hintRound)}${budgetFlag}\n`
+        : null,
+  };
+}
+
+/**
+ * Build stdout lines for `nerve workflow thread`: newest-first selection from
+ * `descRows` until `budgetChars` (including `prefixLines`), then chronological order.
+ */
+export function buildThreadCommandOutput(
+  prefixLines: string[],
+  descRows: ThreadRoundRow[],
+  budgetChars: number,
+  runId: string,
+): ThreadCommandOutput {
+  const prefixText = prefixLines.join("");
+  let remaining = Math.max(0, budgetChars - prefixText.length);
+  const picked: ThreadRoundRow[] = [];
+
+  const budgetFlag =
+    budgetChars === DEFAULT_THREAD_BUDGET_CHARS ? "" : ` --budget ${String(budgetChars)}`;
+
+  for (const row of descRows) {
+    const block = formatThreadRoundBlock(row);
+    if (block.length <= remaining) {
+      picked.push(row);
+      remaining -= block.length;
+      continue;
+    }
+    if (picked.length === 0) {
+      return buildTruncatedSingleRound(row, remaining, prefixLines, runId, budgetFlag);
+    }
+    break;
+  }
+
+  const blocksAsc = picked.map(formatThreadRoundBlock).reverse();
+  const shownMinRound = picked.length === 0 ? null : Math.min(...picked.map((r) => r.round));
+  let paginationHint: string | null = null;
+  if (shownMinRound !== null && shownMinRound > 1) {
+    paginationHint = `\n⏩ Older rounds not shown. Fetch with:\n   nerve workflow thread ${runId} --before ${String(shownMinRound)}${budgetFlag}\n`;
+  }
+
+  return { lines: [...prefixLines, ...blocksAsc], paginationHint };
+}
+
 // ---------------------------------------------------------------------------
 // nerve workflow list
 // ---------------------------------------------------------------------------
@@ -293,6 +415,89 @@ const workflowInspectCommand = defineCommand({
   },
 });
 
+// ---------------------------------------------------------------------------
+// nerve workflow thread <runId>
+// ---------------------------------------------------------------------------
+
+const workflowThreadCommand = defineCommand({
+  meta: {
+    name: "thread",
+    description: "Print role rounds for a workflow run (agent-oriented, budget-limited)",
+  },
+  args: {
+    runId: {
+      type: "positional",
+      description: "The run ID to dump role rounds for",
+    },
+    before: {
+      type: "string",
+      description:
+        "Exclusive upper bound on 1-based round index (use with hint from prior output to load older rounds)",
+      default: "0",
+    },
+    budget: {
+      type: "string",
+      description: `Max output characters including header (default: ${String(DEFAULT_THREAD_BUDGET_CHARS)})`,
+      default: String(DEFAULT_THREAD_BUDGET_CHARS),
+    },
+  },
+  async run({ args }) {
+    const store = await openStore();
+
+    try {
+      const before = Math.max(0, parseIntArg(args.before, 0));
+      const budgetChars = Math.max(1, parseIntArg(args.budget, DEFAULT_THREAD_BUDGET_CHARS));
+
+      const run = store.getWorkflowRun(args.runId);
+      if (run === null) {
+        process.stderr.write(`❌ No workflow run found with runId: ${args.runId}\n`);
+        process.exit(1);
+      }
+
+      const totalRoleRounds = store.getThreadRoundCount(args.runId);
+      if (totalRoleRounds === 0) {
+        process.stdout.write(
+          `🧵 Workflow thread: ${run.runId}\n   workflow: ${run.workflow}\n   status:   ${run.status}\n\n📭 No role rounds recorded for this run.\n`,
+        );
+        return;
+      }
+
+      const descRows = store.getThreadRounds(args.runId, {
+        before,
+        limit: THREAD_ROUNDS_FETCH_LIMIT,
+      });
+
+      const prefixLines = [
+        "🧵 Role rounds (workflow thread)\n",
+        `   runId:    ${run.runId}\n`,
+        `   workflow: ${run.workflow}\n`,
+        `   status:   ${run.status}\n`,
+        `   rounds:   ${String(totalRoleRounds)} role event(s) total\n\n`,
+      ];
+
+      const { lines, paginationHint } = buildThreadCommandOutput(
+        prefixLines,
+        descRows,
+        budgetChars,
+        args.runId,
+      );
+
+      for (const line of lines) {
+        process.stdout.write(line);
+      }
+      if (paginationHint !== null) {
+        process.stdout.write(paginationHint);
+      }
+
+      if (descRows.length === 0 && before > 0) {
+        process.stdout.write(`\n📭 No rounds with index < ${String(before)}.\n`);
+      }
+    } finally {
+      store.close();
+    }
+  },
+});
+
 // ---------------------------------------------------------------------------
 // nerve workflow trigger <name>
 // ---------------------------------------------------------------------------
@@ -309,7 +514,8 @@ const workflowTriggerCommand = defineCommand({
     },
     payload: {
       type: "string",
-      description: "JSON payload to pass as trigger payload (default: {})",
+      description:
+        'JSON with optional "prompt" (string), "maxRounds" (number), and "dryRun" (boolean) for the workflow run (default: {})',
       default: "{}",
     },
   },
@@ -322,15 +528,25 @@ const workflowTriggerCommand = defineCommand({
       process.exit(1);
     }
 
+    let prompt = "";
+    let maxRounds = DEFAULT_ENGINE_MAX_ROUNDS;
+    let dryRun = false;
+    if (isPlainRecord(triggerPayload)) {
+      const p = triggerPayload;
+      if (typeof p.prompt === "string") prompt = p.prompt;
+      if (typeof p.maxRounds === "number") maxRounds = p.maxRounds;
+      if (typeof p.dryRun === "boolean") dryRun = p.dryRun;
+    }
+
     if (!isRunning()) {
       process.stderr.write("❌ Nerve daemon is not running. Start it with `nerve start`.\n");
       process.exit(1);
     }
 
     const socketPath = getSocketPath();
-    let response: { ok: true } | { ok: false; error: string };
+    let response: DaemonIpcTriggerResponse;
     try {
-      response = await triggerWorkflowViaDaemon(socketPath, args.name, triggerPayload);
+      response = await triggerWorkflowViaDaemon(socketPath, args.name, prompt, maxRounds, dryRun);
     } catch (e) {
       const msg = e instanceof Error ? e.message : String(e);
       process.stderr.write(`❌ Failed to contact daemon: ${msg}\n`);
@@ -359,6 +575,7 @@ export const workflowCommand = defineCommand({
   subCommands: {
     list: workflowListCommand,
     inspect: workflowInspectCommand,
+    thread: workflowThreadCommand,
     trigger: workflowTriggerCommand,
   },
 });

packages/cli/src/daemon-client.ts

@@ -8,22 +8,35 @@
 import { connect } from "node:net";
 import type { Socket } from "node:net";
 
-import type { SenseInfo } from "@uncaged/nerve-core";
+import type {
+  DaemonIpcListSensesResponse,
+  DaemonIpcRequest,
+  DaemonIpcTriggerResponse,
+  SenseInfo,
+} from "@uncaged/nerve-core";
+import { isPlainRecord } from "@uncaged/nerve-core";
 
 const CONNECT_TIMEOUT_MS = 3_000;
 const RESPONSE_TIMEOUT_MS = 5_000;
 
 export type { SenseInfo };
 
-type TriggerResponse = { ok: true } | { ok: false; error: string };
+function isSenseInfo(value: unknown): value is SenseInfo {
+  if (!isPlainRecord(value)) return false;
+  return (
+    typeof value.name === "string" &&
+    typeof value.group === "string" &&
+    (value.throttle === null || typeof value.throttle === "number") &&
+    (value.timeout === null || typeof value.timeout === "number") &&
+    (value.lastSignalTs === null || typeof value.lastSignalTs === "number")
+  );
+}
 
-type ListSensesResponse = { ok: true; senses: SenseInfo[] } | { ok: false; error: string };
+function parseDaemonResponse(line: string): DaemonIpcTriggerResponse {
-
-function parseDaemonResponse(line: string): TriggerResponse {
   try {
-    const obj = JSON.parse(line) as unknown;
+    const obj: unknown = JSON.parse(line);
-    if (obj !== null && typeof obj === "object") {
+    if (isPlainRecord(obj)) {
-      const r = obj as Record<string, unknown>;
+      const r = obj;
       if (r.ok === true) return { ok: true };
       if (r.ok === false && typeof r.error === "string") return { ok: false, error: r.error };
     }
@@ -33,14 +46,15 @@ function parseDaemonResponse(line: string): TriggerResponse {
   return { ok: false, error: `Unexpected daemon response: ${line}` };
 }
 
-function parseListSensesResponse(line: string): ListSensesResponse {
+function parseListSensesResponse(line: string): DaemonIpcListSensesResponse {
   try {
-    const obj = JSON.parse(line) as unknown;
+    const obj: unknown = JSON.parse(line);
-    if (obj !== null && typeof obj === "object") {
+    if (isPlainRecord(obj)) {
-      const r = obj as Record<string, unknown>;
+      const r = obj;
       if (r.ok === false && typeof r.error === "string") return { ok: false, error: r.error };
-      if (r.ok === true && Array.isArray(r.senses))
+      if (r.ok === true && Array.isArray(r.senses) && r.senses.every(isSenseInfo)) {
-        return { ok: true, senses: r.senses as SenseInfo[] };
+        return { ok: true, senses: r.senses };
+      }
     }
   } catch {
     // fall through
@@ -54,7 +68,7 @@ function parseListSensesResponse(line: string): ListSensesResponse {
  */
 function sendAndReceive<T>(
   socketPath: string,
-  message: object,
+  message: DaemonIpcRequest,
   parseFirstLine: (trimmed: string) => T,
   responseTimeoutMs: number = RESPONSE_TIMEOUT_MS,
 ): Promise<T> {
@@ -119,27 +133,37 @@ function sendAndReceive<T>(
 export function triggerWorkflowViaDaemon(
   socketPath: string,
   workflow: string,
-  payload: unknown,
+  prompt: string,
-): Promise<TriggerResponse> {
+  maxRounds: number,
-  return sendAndReceive(
+  dryRun = false,
-    socketPath,
+): Promise<DaemonIpcTriggerResponse> {
-    { type: "trigger-workflow", workflow, payload },
+  const message: DaemonIpcRequest = {
-    parseDaemonResponse,
+    type: "trigger-workflow",
-  );
+    workflow,
+    prompt,
+    maxRounds,
+    dryRun,
+  };
+  return sendAndReceive(socketPath, message, parseDaemonResponse);
 }
 
 /**
  * Send a trigger-sense message to the running daemon via its Unix socket.
  * Resolves with the daemon's response or rejects on connection/timeout errors.
  */
-export function triggerSenseViaDaemon(socketPath: string, sense: string): Promise<TriggerResponse> {
+export function triggerSenseViaDaemon(
-  return sendAndReceive(socketPath, { type: "trigger-sense", sense }, parseDaemonResponse);
+  socketPath: string,
+  sense: string,
+): Promise<DaemonIpcTriggerResponse> {
+  const message: DaemonIpcRequest = { type: "trigger-sense", sense };
+  return sendAndReceive(socketPath, message, parseDaemonResponse);
 }
 
 /**
  * Send a list-senses message to the running daemon via its Unix socket.
  * Resolves with the list of registered senses or rejects on connection/timeout errors.
  */
-export function listSensesViaDaemon(socketPath: string): Promise<ListSensesResponse> {
+export function listSensesViaDaemon(socketPath: string): Promise<DaemonIpcListSensesResponse> {
-  return sendAndReceive(socketPath, { type: "list-senses" }, parseListSensesResponse);
+  const message: DaemonIpcRequest = { type: "list-senses" };
+  return sendAndReceive(socketPath, message, parseListSensesResponse);
 }

packages/cli/src/daemon-types.ts

@@ -1,70 +0,0 @@
-/**
- * Structural types for workflow CLI — mirrors @uncaged/nerve-daemon log-store
- * public API so the CLI runtime does not statically depend on the daemon package.
- *
- * ⚠️  Keep in sync with @uncaged/nerve-daemon exports.
- *     Run `pnpm --filter @uncaged/nerve-cli test` to catch drift via satisfies assertions.
- */
-
-export type WorkflowRunStatus =
-  | "queued"
-  | "started"
-  | "completed"
-  | "failed"
-  | "crashed"
-  | "dropped"
-  | "interrupted";
-
-export type WorkflowRun = {
-  runId: string;
-  workflow: string;
-  status: WorkflowRunStatus;
-  ts: number;
-};
-
-export type LogEntry = {
-  id?: number;
-  source: string;
-  type: string;
-  refId: string | null;
-  payload: string | null;
-  ts: number;
-};
-
-export type LogQuery = {
-  source?: string;
-  type?: string;
-  refId?: string;
-  since?: number;
-  until?: number;
-  limit?: number;
-};
-
-export type ArchiveLogsOptions = {
-  now?: number;
-  vacuum?: boolean;
-  maxDays?: number;
-  retentionMs?: number;
-};
-
-export type ArchiveLogsDayResult = {
-  day: string;
-  rowCount: number;
-  filePath: string;
-};
-
-export type ArchiveLogsResult = {
-  days: ArchiveLogsDayResult[];
-  vacuumed: boolean;
-};
-
-/** Subset of daemon LogStore used by the CLI workflow commands. */
-export type LogStore = {
-  query: (filter?: LogQuery) => LogEntry[];
-  getWorkflowRun: (runId: string) => WorkflowRun | null;
-  getActiveWorkflowRuns: (workflowName?: string) => WorkflowRun[];
-  getAllWorkflowRuns: (workflowName: string | null) => WorkflowRun[];
-  upsertWorkflowRun: (entry: Omit<LogEntry, "id">, run: WorkflowRun) => LogEntry;
-  archiveLogs: (options?: ArchiveLogsOptions) => ArchiveLogsResult;
-  close: () => void;
-};

packages/cli/src/workspace-daemon.ts

@@ -5,7 +5,7 @@ import { pathToFileURL } from "node:url";
 
 import type { NerveConfig } from "@uncaged/nerve-core";
 
-import type { LogStore } from "./daemon-types.js";
+import type { LogStore } from "@uncaged/nerve-store";
 
 export function getDaemonEntryPath(nerveRoot: string): string | undefined {
   const pkgPath = join(nerveRoot, "node_modules", "@uncaged", "nerve-daemon", "package.json");
@@ -29,7 +29,7 @@ export function assertWorkspaceDaemonInstalled(nerveRoot: string): string {
   return entry;
 }
 
-/** Loaded from ~/.uncaged-nerve/node_modules at runtime — keep types structural only. */
+/** Loaded from ~/.uncaged-nerve/node_modules at runtime. */
 export type DaemonModule = {
   createKernel: (
     config: NerveConfig,
@@ -46,5 +46,6 @@ export type DaemonModule = {
 export async function loadDaemonModule(nerveRoot: string): Promise<DaemonModule> {
   const entry = assertWorkspaceDaemonInstalled(nerveRoot);
   const url = pathToFileURL(entry).href;
+  // Dynamic import return type is module-specific; narrow at this workspace boundary.
   return import(url) as Promise<DaemonModule>;
 }

packages/cli/tsconfig.json

@@ -6,5 +6,6 @@
     "composite": false,
     "types": ["node"]
   },
-  "include": ["src"]
+  "include": ["src"],
+  "exclude": ["src/__tests__"]
 }

packages/cli/tsup.config.ts

@@ -1,13 +0,0 @@
-import { defineConfig } from "tsup";
-
-export default defineConfig({
-  entry: ["src/index.ts", "src/cli.ts", "src/daemon-bootstrap.ts"],
-  format: ["esm"],
-  dts: true,
-  clean: true,
-  banner: {
-    js: "#!/usr/bin/env node",
-  },
-  /** Daemon is loaded from workspace node_modules at runtime — never bundle it. */
-  external: ["@uncaged/nerve-daemon"],
-});

packages/core/README.md

@@ -0,0 +1,65 @@
+# @uncaged/nerve-core
+
+Shared types and configuration parser for the [nerve](../../README.md) observation engine.
+
+## What's Inside
+
+- **Type definitions** — `Signal`, `SenseConfig`, `SenseInfo`, `SenseReflexConfig`, `ReflexConfig` (sense-only), `WorkflowConfig`, `NerveConfig`, and related types
+- **Config parser** — `parseNerveConfig(yaml)` validates and parses `nerve.yaml` into `NerveConfig` (rejects reflex entries that declare a `workflow` key; reflexes only schedule senses)
+- **Sense → workflow routing** — `parseSenseWorkflowDirective`, `routeSenseComputeOutput`, and types `ParsedSenseWorkflowDirective`, `SenseComputeRoute`
+- **Daemon IPC protocol** — request/response types (`DaemonIpcRequest`, `DaemonIpcResponse`, …) and `parseDaemonIpcRequest` for newline-delimited JSON on the CLI ↔ daemon socket
+- **Workflow automaton types** — `START` / `END` sentinel constants, `WorkflowMessage`, `StartSignal`, `RoleSignal`, `Moderator`, `WorkflowDefinition`, `Role`, `SenseResult`, plus `DEFAULT_ENGINE_MAX_ROUNDS`
+- **Result type** — `Result<T>` with `ok()` / `err()` helpers for explicit error handling (no thrown exceptions for parse paths)
+
+## Usage
+
+```typescript
+import { parseNerveConfig, ok, err } from "@uncaged/nerve-core";
+import type { NerveConfig, Signal, Result } from "@uncaged/nerve-core";
+
+const result: Result<NerveConfig> = parseNerveConfig(yamlString);
+if (result.ok) {
+  console.log(result.value.senses);
+}
+```
+
+### Sense return → signal vs workflow
+
+```typescript
+import { parseSenseWorkflowDirective, routeSenseComputeOutput } from "@uncaged/nerve-core";
+
+const directive = parseSenseWorkflowDirective("my-workflow|8|Hello from sense");
+if (directive.ok) {
+  console.log(directive.value.workflowName, directive.value.maxRounds, directive.value.prompt);
+}
+
+const route = routeSenseComputeOutput({
+  metric: 42,
+  workflow: "my-workflow|8|Run now",
+});
+if (route.kind === "launch") {
+  // engine starts workflow; no Signal to the bus for this return
+  console.log(route.launch);
+} else {
+  // normal signal with payload
+  console.log(route.payload);
+}
+```
+
+## Duration Format
+
+Config fields like `throttle`, `timeout`, and `interval` accept human-readable durations:
+
+- `5s` — 5 seconds
+- `10m` — 10 minutes
+- `1h` — 1 hour
+
+## Install
+
+```bash
+pnpm add @uncaged/nerve-core
+```
+
+## License
+
+MIT

packages/core/package.json

@@ -1,18 +1,23 @@
 {
   "name": "@uncaged/nerve-core",
-  "version": "0.1.4",
+  "version": "0.4.0",
   "type": "module",
   "main": "dist/index.js",
+  "files": ["dist"],
+  "publishConfig": {
+    "access": "public"
+  },
   "types": "dist/index.d.ts",
   "scripts": {
     "prepublishOnly": "bash ../../scripts/prepublish-check.sh",
-    "build": "tsup",
+    "build": "rslib build",
     "test": "vitest run"
   },
   "dependencies": {
     "yaml": "^2.8.3"
   },
   "devDependencies": {
+    "@rslib/core": "^0.21.3",
     "vitest": "^4.1.5"
   }
 }

packages/core/rslib.config.ts

@@ -0,0 +1,19 @@
+import { defineConfig } from "@rslib/core";
+
+export default defineConfig({
+  lib: [
+    {
+      format: "esm",
+      dts: true,
+    },
+  ],
+  source: {
+    entry: {
+      index: "src/index.ts",
+    },
+  },
+  output: {
+    target: "node",
+    cleanDistPath: true,
+  },
+});

packages/core/src/__tests__/config.test.ts

@@ -18,9 +18,6 @@ reflexes:
   - sense: memory
     on:
       - high_usage
-  - workflow: alert
-    on:
-      - cpu
 
 workflows:
   alert:
@@ -48,7 +45,7 @@ describe("parseNerveConfig", () => {
         timeout: 10_000,
         gracePeriod: 3000,
       });
-      expect(result.value.reflexes).toHaveLength(3);
+      expect(result.value.reflexes).toHaveLength(2);
       expect(result.value.reflexes[0]).toEqual({
         kind: "sense",
         sense: "cpu",
@@ -61,11 +58,6 @@ describe("parseNerveConfig", () => {
         interval: null,
         on: ["high_usage"],
       });
-      expect(result.value.reflexes[2]).toEqual({
-        kind: "workflow",
-        workflow: "alert",
-        on: ["cpu"],
-      });
       expect(result.value.workflows?.alert).toEqual({
         concurrency: 2,
         overflow: "queue",
@@ -201,7 +193,7 @@ reflexes:
       expect(result.error.message).toMatch(/disk.*not found in senses/);
     });
 
-    it("returns error when workflow reflex references a non-existent workflow", () => {
+    it("returns error when reflex uses unsupported workflow field", () => {
       const yaml = `
 senses:
   cpu:
@@ -214,10 +206,10 @@ reflexes:
       const result = parseNerveConfig(yaml);
       expect(result.ok).toBe(false);
       if (result.ok) return;
-      expect(result.error.message).toMatch(/missing_wf.*not found in workflows/);
+      expect(result.error.message).toMatch(/workflow.*not supported/);
     });
 
-    it("returns error when workflow reflex references non-existent workflow (with workflows defined)", () => {
+    it("returns error when reflex uses unsupported workflow field (with workflows defined)", () => {
       const yaml = `
 senses:
   cpu:
@@ -234,7 +226,7 @@ workflows:
       const result = parseNerveConfig(yaml);
       expect(result.ok).toBe(false);
       if (result.ok) return;
-      expect(result.error.message).toMatch(/unknown.*not found in workflows/);
+      expect(result.error.message).toMatch(/workflow.*not supported/);
     });
 
     it("returns error for invalid throttle format", () => {
@@ -362,7 +354,7 @@ reflexes:
       const result = parseNerveConfig(yaml);
       expect(result.ok).toBe(false);
       if (result.ok) return;
-      expect(result.error.message).toMatch(/cannot have both/);
+      expect(result.error.message).toMatch(/workflow.*not supported/);
     });
 
     it("returns error when reflex has neither sense nor workflow", () => {
@@ -376,7 +368,7 @@ reflexes:
       const result = parseNerveConfig(yaml);
       expect(result.ok).toBe(false);
       if (result.ok) return;
-      expect(result.error.message).toMatch(/must have either/);
+      expect(result.error.message).toMatch(/must include "sense"/);
     });
   });
 });

packages/core/src/__tests__/daemon-ipc-protocol.test.ts

@@ -0,0 +1,72 @@
+import { describe, expect, it } from "vitest";
+
+import { parseDaemonIpcRequest } from "../daemon-ipc-protocol.js";
+
+describe("parseDaemonIpcRequest", () => {
+  it("parses trigger-workflow", () => {
+    expect(
+      parseDaemonIpcRequest(
+        JSON.stringify({
+          type: "trigger-workflow",
+          workflow: "wf",
+          prompt: "go",
+          maxRounds: 3,
+        }),
+      ),
+    ).toEqual({
+      type: "trigger-workflow",
+      workflow: "wf",
+      prompt: "go",
+      maxRounds: 3,
+      dryRun: false,
+    });
+  });
+
+  it("parses trigger-workflow with dryRun true", () => {
+    expect(
+      parseDaemonIpcRequest(
+        JSON.stringify({
+          type: "trigger-workflow",
+          workflow: "wf",
+          prompt: "go",
+          maxRounds: 3,
+          dryRun: true,
+        }),
+      ),
+    ).toEqual({
+      type: "trigger-workflow",
+      workflow: "wf",
+      prompt: "go",
+      maxRounds: 3,
+      dryRun: true,
+    });
+  });
+
+  it("rejects trigger-workflow with empty workflow", () => {
+    expect(
+      parseDaemonIpcRequest(
+        JSON.stringify({
+          type: "trigger-workflow",
+          workflow: "",
+          prompt: "",
+          maxRounds: 1,
+        }),
+      ),
+    ).toBeNull();
+  });
+
+  it("parses trigger-sense and list-senses", () => {
+    expect(parseDaemonIpcRequest(JSON.stringify({ type: "trigger-sense", sense: "x" }))).toEqual({
+      type: "trigger-sense",
+      sense: "x",
+    });
+    expect(parseDaemonIpcRequest(JSON.stringify({ type: "list-senses" }))).toEqual({
+      type: "list-senses",
+    });
+  });
+
+  it("returns null for invalid JSON or unknown type", () => {
+    expect(parseDaemonIpcRequest("not json")).toBeNull();
+    expect(parseDaemonIpcRequest(JSON.stringify({ type: "nope" }))).toBeNull();
+  });
+});

packages/core/src/config.ts

@@ -1,8 +1,10 @@
 import { parse } from "yaml";
 
+import { isPlainRecord } from "./is-plain-record.js";
 import type { Result } from "./result.js";
 import { err, ok } from "./result.js";
 import type { NerveConfig, ReflexConfig, SenseConfig, WorkflowConfig } from "./types.js";
+import { DEFAULT_ENGINE_MAX_ROUNDS } from "./types.js";
 
 const DURATION_RE = /^(\d+)([smh])$/;
 
@@ -39,11 +41,11 @@ function parseDurationField(field: unknown, label: string): Result<number | null
 }
 
 function validateSenseConfig(name: string, raw: unknown): Result<SenseConfig> {
-  if (raw === null || typeof raw !== "object" || Array.isArray(raw)) {
+  if (!isPlainRecord(raw)) {
     return err(new Error(`senses.${name}: must be an object`));
   }
 
-  const obj = raw as Record<string, unknown>;
+  const obj = raw;
 
   if (typeof obj.group !== "string" || obj.group.trim() === "") {
     return err(new Error(`senses.${name}.group: required string`));
@@ -76,10 +78,10 @@ function validateSenseConfig(name: string, raw: unknown): Result<SenseConfig> {
 
 function parseOnField(index: number, obj: Record<string, unknown>): Result<string[] | null> {
   if (obj.on === undefined || obj.on === null) return ok(null);
-  if (!Array.isArray(obj.on) || !obj.on.every((item) => typeof item === "string")) {
+  if (!Array.isArray(obj.on) || !obj.on.every((item): item is string => typeof item === "string")) {
     return err(new Error(`reflexes[${index}].on: must be an array of strings`));
   }
-  return ok(obj.on as string[]);
+  return ok(obj.on);
 }
 
 function parseSenseReflex(
@@ -112,61 +114,56 @@ function parseSenseReflex(
   });
 }
 
-function parseWorkflowReflex(
-  index: number,
-  obj: Record<string, unknown>,
-  on: string[] | null,
-): Result<ReflexConfig> {
-  if (typeof obj.workflow !== "string") {
-    return err(new Error(`reflexes[${index}].workflow: must be a string`));
-  }
-  if (obj.interval !== undefined) {
-    return err(
-      new Error(`reflexes[${index}]: workflow reflex does not support "interval" (use "on")`),
-    );
-  }
-  return ok({
-    kind: "workflow" as const,
-    workflow: obj.workflow,
-    on,
-  });
-}
-
 function validateReflexConfig(
   index: number,
   raw: unknown,
   senseNames: Set<string>,
 ): Result<ReflexConfig> {
-  if (raw === null || typeof raw !== "object" || Array.isArray(raw)) {
+  if (!isPlainRecord(raw)) {
     return err(new Error(`reflexes[${index}]: must be an object`));
   }
 
-  const obj = raw as Record<string, unknown>;
+  const obj = raw;
   const hasSense = obj.sense !== undefined;
-  const hasWorkflow = obj.workflow !== undefined;
+  const hasWorkflowKey = Object.hasOwn(obj, "workflow");
 
-  if (hasSense && hasWorkflow) {
+  if (hasWorkflowKey) {
-    return err(new Error(`reflexes[${index}]: cannot have both "sense" and "workflow"`));
+    return err(
+      new Error(
+        `reflexes[${index}]: YAML "workflow" entries are not supported — start workflows from a Sense compute return value using a "workflow" string field (format: name|maxRounds|prompt)`,
+      ),
+    );
   }
-  if (!hasSense && !hasWorkflow) {
+  if (!hasSense) {
-    return err(new Error(`reflexes[${index}]: must have either "sense" or "workflow"`));
+    return err(new Error(`reflexes[${index}]: must include "sense"`));
   }
 
   const onResult = parseOnField(index, obj);
   if (!onResult.ok) return onResult;
 
-  if (hasSense) {
+  return parseSenseReflex(index, obj, senseNames, onResult.value);
-    return parseSenseReflex(index, obj, senseNames, onResult.value);
+}
+
+function parseEngineMaxRounds(obj: Record<string, unknown>): Result<number> {
+  if (obj.max_rounds === undefined || obj.max_rounds === null) {
+    return ok(DEFAULT_ENGINE_MAX_ROUNDS);
   }
-  return parseWorkflowReflex(index, obj, onResult.value);
+  if (
+    typeof obj.max_rounds !== "number" ||
+    !Number.isInteger(obj.max_rounds) ||
+    obj.max_rounds < 1
+  ) {
+    return err(new Error("max_rounds: must be a positive integer"));
+  }
+  return ok(obj.max_rounds);
 }
 
 function validateWorkflowConfig(name: string, raw: unknown): Result<WorkflowConfig> {
-  if (raw === null || typeof raw !== "object" || Array.isArray(raw)) {
+  if (!isPlainRecord(raw)) {
     return err(new Error(`workflows.${name}: must be an object`));
   }
 
-  const obj = raw as Record<string, unknown>;
+  const obj = raw;
 
   if (
     typeof obj.concurrency !== "number" ||
@@ -213,11 +210,11 @@ function validateWorkflowConfig(name: string, raw: unknown): Result<WorkflowConf
 function parseSenses(
   obj: Record<string, unknown>,
 ): Result<{ senses: Record<string, SenseConfig>; senseNames: Set<string> }> {
-  if (obj.senses === null || typeof obj.senses !== "object" || Array.isArray(obj.senses)) {
+  if (!isPlainRecord(obj.senses)) {
     return err(new Error("senses: required object"));
   }
 
-  const sensesRaw = obj.senses as Record<string, unknown>;
+  const sensesRaw = obj.senses;
   const senses: Record<string, SenseConfig> = {};
   const senseNames = new Set(Object.keys(sensesRaw));
 
@@ -253,11 +250,11 @@ function parseWorkflows(
 ): Result<Record<string, WorkflowConfig> | null> {
   if (obj.workflows === undefined || obj.workflows === null) return ok(null);
 
-  if (typeof obj.workflows !== "object" || Array.isArray(obj.workflows)) {
+  if (!isPlainRecord(obj.workflows)) {
     return err(new Error("workflows: must be an object if provided"));
   }
 
-  const workflowsRaw = obj.workflows as Record<string, unknown>;
+  const workflowsRaw = obj.workflows;
   const workflows: Record<string, WorkflowConfig> = {};
 
   for (const [name, wfRaw] of Object.entries(workflowsRaw)) {
@@ -279,11 +276,11 @@ export function parseNerveConfig(raw: string): Result<NerveConfig> {
     return err(new Error(`YAML parse error: ${message}`));
   }
 
-  if (parsed === null || typeof parsed !== "object" || Array.isArray(parsed)) {
+  if (!isPlainRecord(parsed)) {
     return err(new Error("Config must be a YAML object"));
   }
 
-  const obj = parsed as Record<string, unknown>;
+  const obj = parsed;
 
   const sensesResult = parseSenses(obj);
   if (!sensesResult.ok) return sensesResult;
@@ -295,16 +292,11 @@ export function parseNerveConfig(raw: string): Result<NerveConfig> {
   const workflowsResult = parseWorkflows(obj);
   if (!workflowsResult.ok) return workflowsResult;
 
-  // Cross-validate: workflow reflexes must reference defined workflows
+  const maxRoundsResult = parseEngineMaxRounds(obj);
-  const workflowNames = new Set(workflowsResult.value ? Object.keys(workflowsResult.value) : []);
+  if (!maxRoundsResult.ok) return maxRoundsResult;
-  for (let i = 0; i < reflexesResult.value.length; i++) {
-    const reflex = reflexesResult.value[i];
-    if (reflex.kind === "workflow" && !workflowNames.has(reflex.workflow)) {
-      return err(new Error(`reflexes[${i}].workflow: "${reflex.workflow}" not found in workflows`));
-    }
-  }
 
   return ok({
+    maxRounds: maxRoundsResult.value,
     senses,
     reflexes: reflexesResult.value,
     workflows: workflowsResult.value,

packages/core/src/daemon-ipc-protocol.ts

@@ -0,0 +1,94 @@
+/**
+ * Daemon Unix-socket IPC protocol (CLI → daemon).
+ * Newline-delimited JSON: one request object per line from the client,
+ * one response object per line from the daemon.
+ */
+
+import { isPlainRecord } from "./is-plain-record.js";
+import type { SenseInfo } from "./types.js";
+
+/** Client → daemon: start a workflow run. */
+export type DaemonIpcTriggerWorkflowRequest = {
+  type: "trigger-workflow";
+  workflow: string;
+  prompt: string;
+  maxRounds: number;
+  dryRun: boolean;
+};
+
+/** Client → daemon: run a sense compute on demand. */
+export type DaemonIpcTriggerSenseRequest = {
+  type: "trigger-sense";
+  sense: string;
+};
+
+/** Client → daemon: list registered senses. */
+export type DaemonIpcListSensesRequest = {
+  type: "list-senses";
+};
+
+/** Union of all JSON requests the daemon IPC server accepts. */
+export type DaemonIpcRequest =
+  | DaemonIpcTriggerWorkflowRequest
+  | DaemonIpcTriggerSenseRequest
+  | DaemonIpcListSensesRequest;
+
+/** Successful trigger / trigger-sense reply (no body). */
+export type DaemonIpcTriggerOkResponse = { ok: true };
+
+export type DaemonIpcErrorResponse = { ok: false; error: string };
+
+/** Replies for trigger-workflow and trigger-sense. */
+export type DaemonIpcTriggerResponse = DaemonIpcTriggerOkResponse | DaemonIpcErrorResponse;
+
+/** Reply for list-senses. */
+export type DaemonIpcListSensesResponse =
+  | { ok: true; senses: SenseInfo[] }
+  | DaemonIpcErrorResponse;
+
+/** Any JSON response the daemon may write on the IPC socket. */
+export type DaemonIpcResponse =
+  | DaemonIpcTriggerOkResponse
+  | DaemonIpcErrorResponse
+  | { ok: true; senses: SenseInfo[] };
+
+function parseTriggerWorkflowFields(
+  req: Record<string, unknown>,
+): DaemonIpcTriggerWorkflowRequest | null {
+  if (typeof req.workflow !== "string" || req.workflow.length === 0) return null;
+  if (typeof req.prompt !== "string") return null;
+  if (typeof req.maxRounds !== "number") return null;
+  const dryRun = typeof req.dryRun === "boolean" ? req.dryRun : false;
+  return {
+    type: "trigger-workflow",
+    workflow: req.workflow,
+    prompt: req.prompt,
+    maxRounds: req.maxRounds,
+    dryRun,
+  };
+}
+
+/**
+ * Parse a single line of JSON into a {@link DaemonIpcRequest}, or null if invalid.
+ * Kept in core with the request types so CLI and daemon stay aligned at compile time.
+ */
+export function parseDaemonIpcRequest(line: string): DaemonIpcRequest | null {
+  try {
+    const obj: unknown = JSON.parse(line);
+    if (!isPlainRecord(obj)) return null;
+    const req = obj;
+    if (req.type === "trigger-workflow") {
+      return parseTriggerWorkflowFields(req);
+    }
+    if (req.type === "trigger-sense") {
+      if (typeof req.sense !== "string" || req.sense.length === 0) return null;
+      return { type: "trigger-sense", sense: req.sense };
+    }
+    if (req.type === "list-senses") {
+      return { type: "list-senses" };
+    }
+    return null;
+  } catch {
+    return null;
+  }
+}

packages/core/src/index.ts

@@ -3,21 +3,46 @@ export type {
   SenseConfig,
   SenseInfo,
   SenseReflexConfig,
-  WorkflowReflexConfig,
   ReflexConfig,
   DropOverflowConfig,
   QueueOverflowConfig,
   WorkflowConfig,
   NerveConfig,
-  CommandEvent,
+  WorkflowMessage,
-  ThreadState,
+  RoleResult,
-  ModerateResult,
-  WorkflowContext,
-  RoleExecuteFn,
   Role,
-  ModerateFn,
+  RoleMeta,
+  StartSignal,
+  RoleSignal,
+  ModeratorContext,
+  Moderator,
   WorkflowDefinition,
+  SenseResult,
 } from "./types.js";
+export { START, END, DEFAULT_ENGINE_MAX_ROUNDS } from "./types.js";
 export type { Result } from "./result.js";
 export { ok, err } from "./result.js";
 export { parseNerveConfig } from "./config.js";
+export { isPlainRecord } from "./is-plain-record.js";
+
+export type {
+  ParsedSenseWorkflowDirective,
+  SenseComputeRoute,
+} from "./sense-workflow-directive.js";
+export {
+  parseSenseWorkflowDirective,
+  routeSenseComputeOutput,
+} from "./sense-workflow-directive.js";
+
+export type {
+  DaemonIpcTriggerWorkflowRequest,
+  DaemonIpcTriggerSenseRequest,
+  DaemonIpcListSensesRequest,
+  DaemonIpcRequest,
+  DaemonIpcTriggerOkResponse,
+  DaemonIpcErrorResponse,
+  DaemonIpcTriggerResponse,
+  DaemonIpcListSensesResponse,
+  DaemonIpcResponse,
+} from "./daemon-ipc-protocol.js";
+export { parseDaemonIpcRequest } from "./daemon-ipc-protocol.js";

packages/core/src/is-plain-record.ts

@@ -0,0 +1,7 @@
+/**
+ * Narrows `unknown` to a plain JSON-style object (not null, not array).
+ * Use after `JSON.parse` / YAML / IPC when validating structure field-by-field.
+ */
+export function isPlainRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}

packages/core/src/sense-workflow-directive.ts

@@ -0,0 +1,77 @@
+import { isPlainRecord } from "./is-plain-record.js";
+import type { Result } from "./result.js";
+import { err, ok } from "./result.js";
+
+/** Parsed `workflow-name|maxRounds|prompt` from a Sense compute return value. */
+export type ParsedSenseWorkflowDirective = {
+  workflowName: string;
+  maxRounds: number;
+  prompt: string;
+};
+
+/**
+ * Parses the pipe-separated `workflow` field from a Sense compute result.
+ * `prompt` may contain `|` — only the first two pipes delimit name and rounds.
+ */
+export function parseSenseWorkflowDirective(field: string): Result<ParsedSenseWorkflowDirective> {
+  const trimmed = field.trim();
+  if (trimmed.length === 0) {
+    return err(new Error("workflow directive is empty"));
+  }
+  const parts = trimmed.split("|");
+  if (parts.length < 3) {
+    return err(
+      new Error(
+        `workflow directive must be "name|maxRounds|prompt" (got ${String(parts.length)} segment(s))`,
+      ),
+    );
+  }
+  const workflowName = (parts[0] ?? "").trim();
+  if (workflowName.length === 0) {
+    return err(new Error("workflow directive: empty workflow name"));
+  }
+  const roundsRaw = (parts[1] ?? "").trim();
+  const maxRounds = Number.parseInt(roundsRaw, 10);
+  if (!Number.isInteger(maxRounds) || maxRounds < 1) {
+    return err(new Error(`workflow directive: invalid maxRounds "${roundsRaw}"`));
+  }
+  const prompt = parts.slice(2).join("|");
+  return ok({ workflowName, maxRounds, prompt });
+}
+
+export type SenseComputeRoute =
+  | { kind: "launch"; launch: ParsedSenseWorkflowDirective }
+  | { kind: "signal"; payload: unknown };
+
+function stripWorkflowKey(payload: Record<string, unknown>): Record<string, unknown> {
+  const { workflow: _drop, ...rest } = payload;
+  return rest;
+}
+
+/**
+ * Interprets a Sense compute non-null return value for the engine:
+ * - `workflow` missing → normal signal with full payload
+ * - `workflow: null` or `""` → normal signal; `workflow` key stripped from emitted payload
+ * - `workflow: "name|n|prompt"` → launch workflow; no Signal is emitted to the bus
+ */
+export function routeSenseComputeOutput(payload: unknown): SenseComputeRoute {
+  if (!isPlainRecord(payload)) {
+    return { kind: "signal", payload };
+  }
+  const obj = payload;
+  if (!Object.hasOwn(obj, "workflow")) {
+    return { kind: "signal", payload };
+  }
+  const w = obj.workflow;
+  if (w === null || w === "") {
+    return { kind: "signal", payload: stripWorkflowKey(obj) };
+  }
+  if (typeof w !== "string") {
+    return { kind: "signal", payload };
+  }
+  const parsed = parseSenseWorkflowDirective(w);
+  if (!parsed.ok) {
+    return { kind: "signal", payload: stripWorkflowKey(obj) };
+  }
+  return { kind: "launch", launch: parsed.value };
+}

packages/core/src/types.ts

@@ -28,13 +28,8 @@ export type SenseReflexConfig = {
   on: string[] | null;
 };
 
-export type WorkflowReflexConfig = {
+/** Reflexes only schedule Senses; workflow launches come from Sense return values. */
-  kind: "workflow";
+export type ReflexConfig = SenseReflexConfig;
-  workflow: string;
-  on: string[] | null;
-};
-
-export type ReflexConfig = SenseReflexConfig | WorkflowReflexConfig;
 
 export type DropOverflowConfig = {
   concurrency: number;
@@ -50,62 +45,90 @@ export type QueueOverflowConfig = {
 export type WorkflowConfig = DropOverflowConfig | QueueOverflowConfig;
 
 export type NerveConfig = {
+  /** Engine-wide default max moderator rounds (e.g. CLI workflow trigger when omitted). */
+  maxRounds: number;
   senses: Record<string, SenseConfig>;
   reflexes: ReflexConfig[];
   workflows: Record<string, WorkflowConfig> | null;
 };
 
 // ---------------------------------------------------------------------------
-// Workflow Engine types (RFC-002)
+// Workflow Automaton types (issue #80)
 // ---------------------------------------------------------------------------
 
-/** A single event in the command event stream that drives a workflow thread. */
+export const START = "__start__" as const;
-export type CommandEvent = {
+export const END = "__end__" as const;
-  type: string;
+export type START = typeof START;
-  [key: string]: unknown;
+export type END = typeof END;
-};
 
-/** Accumulated state of a running thread — the event history for moderate(). */
+/** Engine-wide fallback for max moderator rounds when not specified in config. */
-export type ThreadState = {
+export const DEFAULT_ENGINE_MAX_ROUNDS = 100;
-  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. */
+/** A single message in the workflow conversation chain (runtime, type-erased). */
-export type ModerateResult = {
+export type WorkflowMessage = {
   role: string;
-  prompt: unknown;
+  content: string;
+  meta: unknown;
+  timestamp: number;
 };
 
-/** Context injected into every role execute() call. */
+/** The typed output of a Role execution. */
-export type WorkflowContext = {
+export type RoleResult<Meta> = { content: string; meta: Meta };
-  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.).
+ * A Role is a pure async function: receives the engine start frame plus prior
- * Returns a CommandEvent that is fed back into moderate().
+ * role messages only (the start frame is not included in `messages`).
+ * Returns typed content + meta. Implementation can be an agent, LLM call,
+ * script, HTTP request, etc.
  */
-export type RoleExecuteFn = (prompt: unknown, ctx: WorkflowContext) => Promise<CommandEvent>;
+export type Role<Meta> = (
+  start: StartSignal,
+  messages: WorkflowMessage[],
+) => Promise<RoleResult<Meta>>;
 
-/** A role in a workflow — a named unit of execution with side effects. */
+/** Maps role names to their meta types — the single generic that drives all inference. */
-export type Role = {
+export type RoleMeta = Record<string, Record<string, unknown>>;
-  execute: RoleExecuteFn;
+
+/** Engine start frame: prompt, max rounds cap, dry-run flag, and timestamps for the thread. */
+export type StartSignal = {
+  role: START;
+  content: string;
+  meta: { maxRounds: number; dryRun: boolean };
+  timestamp: number;
 };
 
+/** A discriminated union of signals from each role, derived from the meta map. */
+export type RoleSignal<M extends RoleMeta> = {
+  [K in keyof M & string]: { role: K; meta: M[K] };
+}[keyof M & string];
+
 /**
- * The moderator function — pure, no side effects.
+ * Moderator input: either the initial start frame or a role signal after a step.
- * Decides which role to pass control to next.
+ * Lets implementations branch on `context.kind` with full typing for each arm.
- * Returns null to signal thread completion.
  */
-export type ModerateFn = (thread: ThreadState, event: CommandEvent) => ModerateResult | null;
+export type ModeratorContext<M extends RoleMeta> =
+  | { kind: "start"; start: StartSignal }
+  | { kind: "step"; signal: RoleSignal<M> };
+
+/**
+ * The moderator — a pure routing function. Receives start vs step context,
+ * current round, and maxRounds. Returns the next role name or END.
+ */
+export type Moderator<M extends RoleMeta> = (
+  context: ModeratorContext<M>,
+  round: number,
+  maxRounds: number,
+) => (keyof M & string) | END;
 
 /** The complete definition of a workflow, as authored by users. */
-export type WorkflowDefinition = {
+export type WorkflowDefinition<M extends RoleMeta> = {
-  roles: Record<string, Role>;
+  name: string;
-  moderate: ModerateFn;
+  roles: { [K in keyof M & string]: Role<M[K]> };
+  moderator: Moderator<M>;
+};
+
+/** The result of a Sense compute — payload plus optional workflow directive. */
+export type SenseResult = {
+  payload: unknown;
+  workflow: string | null;
 };

packages/core/tsup.config.ts

@@ -1,8 +0,0 @@
-import { defineConfig } from "tsup";
-
-export default defineConfig({
-  entry: ["src/index.ts"],
-  format: ["esm"],
-  dts: true,
-  clean: true,
-});

packages/daemon/README.md

@@ -0,0 +1,92 @@
+# @uncaged/nerve-daemon
+
+The observation engine runtime for [nerve](../../README.md) — runs senses, routes signals, schedules reflexes, and manages workflows.
+
+## Architecture
+
+| Module | Source (indicative) | Responsibility |
+|--------|---------------------|----------------|
+| **Kernel** | `kernel.ts` | Orchestrator — worker pool, signal bus, reflex scheduler, workflow manager, optional file watcher and daemon IPC, config reload hooks |
+| **Worker pool** | `worker-pool.ts` | Fork and supervise one child process per sense group; restart/shutdown; crash cleanup hooks for scheduler state |
+| **Kernel sense groups** | `kernel-sense-groups.ts` | Derive sense groups from config; list senses per group for scheduling |
+| **Sense runtime** | sense worker + Drizzle | Per-sense SQLite (`node:sqlite`), migrations, peer DB reads |
+| **Sense worker** | `sense-worker.ts` (fork target) | Child process entry — runs `compute()` per sense in a group |
+| **Signal bus** | `signal-bus.ts` | In-memory pub/sub for sense signals |
+| **Reflex scheduler** | `reflex-scheduler.ts` | Interval + `on` subscriptions, throttle/coalesce |
+| **Workflow manager** | `workflow-manager.ts` | One worker per workflow name, concurrency (drop/queue), queue caps |
+| **Workflow worker** | `workflow-worker.ts` | Child process — runs RFC-002 threads (`start-thread`, `resume-thread` IPC) |
+| **IPC (parent ↔ workers)** | `ipc.ts` | Typed messages for sense and workflow workers (includes `resume-thread` for recovery) |
+| **Log / workflow persistence** | via `@uncaged/nerve-store` | Structured logs, `workflow_runs`, thread messages (used for recovery) |
+| **Blob store** | `@uncaged/nerve-store` | CAS under `data/blobs/` — sense workers construct `createBlobStore(join(nerveRoot, "data", "blobs"))` for artifact writes |
+| **File watcher** | `file-watcher.ts` | Watches workspace paths for config / sense / workflow file changes |
+| **Kernel file watch** | `kernel-file-watch.ts` | Maps watcher events to `reloadConfig`, sense group restart, workflow `drainAndRespawn` |
+| **Daemon IPC** | `daemon-ipc.ts` | Unix socket server — parses `@uncaged/nerve-core` `DaemonIpcRequest`, dispatches trigger-workflow / trigger-sense / list-senses |
+
+## Crash recovery (workflow workers)
+
+If a workflow worker exits unexpectedly while threads are active:
+
+- In-flight runs are marked **`crashed`** in the log store; the manager respawns a fresh worker.
+- Runs still in **`started`** state can be **`resume-thread`**’d: the manager rebuilds the message chain from persisted workflow log rows and sends `resume-thread` to the new worker.
+- **Crash-loop backoff:** repeated crashes for the same workflow name are counted in a sliding window (`60s`); after **`5`** crashes in that window, the manager **stops respawning** that worker and logs the condition (avoids tight crash loops).
+
+Hot reload (`drainAndRespawn`) uses a controlled drain: in-flight runs may be marked **`interrupted`** when the old worker is torn down after a timeout — that path is distinct from unexpected crash recovery.
+
+## Key Design Decisions
+
+- **One worker process per sense group** — isolation between groups, shared compute within a group
+- **`node:sqlite` (DatabaseSync)** — zero native addons, WAL mode, built into Node.js ≥ 22.5
+- **Throttle + coalesce** — if compute is in-flight, at most one pending trigger is queued (no unbounded accumulation)
+- **Log ≠ Signal** — logs are queryable data assets but cannot trigger reflexes (prevents feedback loops)
+
+## Usage
+
+The daemon is typically started via the CLI (`nerve daemon start` / `nerve dev`), but you can embed the kernel:
+
+```typescript
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+
+import { parseNerveConfig } from "@uncaged/nerve-core";
+import { createKernel } from "@uncaged/nerve-daemon";
+
+const nerveRoot = "/path/to/workspace";
+const yamlPath = join(nerveRoot, "nerve.yaml");
+const parsed = parseNerveConfig(readFileSync(yamlPath, "utf8"));
+if (!parsed.ok) {
+  throw parsed.error;
+}
+
+const kernel = createKernel(parsed.value, nerveRoot, {
+  enableFileWatcher: true,
+  ipcSocketPath: join(nerveRoot, "nerve.sock"),
+});
+
+await kernel.ready;
+
+kernel.triggerSense("cpu-usage");
+const health = kernel.getHealth();
+
+await kernel.stop();
+```
+
+`createKernel(config, nerveRoot, options?)` — `config` is a parsed `NerveConfig`; `nerveRoot` is the workspace root (contains `nerve.yaml`, `data/`, etc.). Optional `KernelOptions`:
+
+| Field | Meaning |
+|-------|---------|
+| `workerScript` | Override path to the sense worker entry script (defaults to the package’s resolved worker) |
+| `enableFileWatcher` | Watch config / senses / workflows for hot reload |
+| `logStore` | Inject a `LogStore` instance (defaults to `createLogStore(join(nerveRoot, "data", "logs.db"))`) |
+| `ipcSocketPath` | When non-null, listen for daemon IPC on this Unix socket path |
+
+## Install
+
+```bash
+pnpm add @uncaged/nerve-daemon
+```
+
+Requires Node.js ≥ 22.5 (for `node:sqlite`).
+
+## License
+
+MIT

packages/daemon/package.json

@@ -1,28 +1,27 @@
 {
   "name": "@uncaged/nerve-daemon",
-  "version": "0.1.5",
+  "version": "0.4.0",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
-  "files": [
+  "files": ["dist"],
-    "dist"
-  ],
   "publishConfig": {
     "access": "public"
   },
   "scripts": {
     "prepublishOnly": "bash ../../scripts/prepublish-check.sh",
-    "build": "tsup",
+    "build": "rslib build",
     "test": "vitest run"
   },
   "dependencies": {
     "@uncaged/nerve-core": "workspace:*",
-    "better-sqlite3": "^11.10.0",
+    "@uncaged/nerve-store": "workspace:*",
-    "drizzle-orm": "^0.43.1",
+    "drizzle-orm": "1.0.0-beta.23-c10d10c",
     "yaml": "^2.8.3"
   },
   "devDependencies": {
-    "@types/better-sqlite3": "^7.6.13",
+    "@rslib/core": "^0.21.3",
+    "@types/node": "^22.0.0",
     "vitest": "^4.1.5"
   }
 }

packages/daemon/rslib.config.ts

@@ -0,0 +1,21 @@
+import { defineConfig } from "@rslib/core";
+
+export default defineConfig({
+  lib: [
+    {
+      format: "esm",
+      dts: true,
+    },
+  ],
+  source: {
+    entry: {
+      index: "src/index.ts",
+      "sense-worker": "src/sense-worker.ts",
+      "workflow-worker": "src/workflow-worker.ts",
+    },
+  },
+  output: {
+    target: "node",
+    cleanDistPath: true,
+  },
+});

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

@@ -64,6 +64,7 @@ function makeConfig(workflows: Record<string, WorkflowConfig> = {}): NerveConfig
     senses: {},
     reflexes: [],
     workflows,
+    maxRounds: 10,
   };
 }
 
@@ -89,10 +90,20 @@ function makeLogStore(
       }
       return activeRuns;
     }),
-    getTriggerPayload: vi.fn(() => ({ value: 42 })),
+    getTriggerPayload: vi.fn((): unknown => ({ value: 42 })),
-    getThreadEvents: vi.fn(() => [{ type: "thread_start", triggerPayload: {} }]),
+    getThreadEvents: vi.fn(
+      (): Array<{ type: string; [key: string]: unknown }> => [
+        { type: "thread_start", triggerPayload: {} },
+      ],
+    ),
+    getThreadMessages: vi.fn(
+      (): Array<{ role: string; content: string; meta: unknown; timestamp: number }> => [],
+    ),
+    getThreadRoundCount: vi.fn(() => 0),
+    getThreadRounds: vi.fn(() => []),
     archiveLogs: vi.fn(() => ({ days: [], vacuumed: false })),
     close: vi.fn(),
+    getAllWorkflowRuns: vi.fn(() => []),
   };
   return store;
 }
@@ -116,8 +127,8 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-wf", { n: 1 });
+      mgr.startWorkflow("my-wf", { prompt: "test 1", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("my-wf", { n: 2 });
+      mgr.startWorkflow("my-wf", { prompt: "test 2", maxRounds: 10, dryRun: false });
       expect(mgr.activeCount("my-wf")).toBe(2);
 
       // Simulate unexpected exit (not shutdown)
@@ -127,7 +138,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       child.emit("exit", 1, null);
 
       const crashedCalls = logStore.upsertWorkflowRun.mock.calls.filter(
-        ([entry]: [{ type: string }]) => entry.type === "crashed",
+        (args: any[]) => (args[0] as { type: string }).type === "crashed",
       );
       expect(crashedCalls).toHaveLength(2);
 
@@ -143,8 +154,8 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       expect(mgr.activeCount("my-wf")).toBe(2);
 
       const child = mockChildren[0];
@@ -168,7 +179,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       expect(mockChildren).toHaveLength(1);
 
       const child = mockChildren[0];
@@ -191,9 +202,8 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
         { runId: "run-started-1", workflow: "my-wf", status: "started" as const, ts: 1000 },
       ];
       const logStore = makeLogStore(activeRuns);
-      logStore.getThreadEvents.mockReturnValue([
+      logStore.getThreadMessages.mockReturnValue([
-        { type: "thread_start", triggerPayload: {} },
+        { role: "scanner", content: "done", meta: { items: ["a"] }, timestamp: 1000 },
-        { type: "scan_complete", items: ["a"] },
       ]);
       logStore.getTriggerPayload.mockReturnValue({ trigger: "initial" });
 
@@ -202,7 +212,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       const firstChild = mockChildren[0];
       firstChild.exitCode = 1;
       firstChild.connected = false;
@@ -216,18 +226,17 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
 
       // resume-thread should have been sent
       const resumeCalls = (secondChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
-        ([msg]: [unknown]) =>
+        (args: any[]) =>
-          msg !== null &&
+          args[0] !== null &&
-          typeof msg === "object" &&
+          typeof args[0] === "object" &&
-          (msg as Record<string, unknown>).type === "resume-thread",
+          (args[0] 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);
+      expect(Array.isArray((resumeCalls[0][0] as Record<string, unknown>).messages)).toBe(true);
 
       const stopPromise = mgr.stop();
       await vi.runAllTimersAsync();
@@ -247,7 +256,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       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", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       const firstChild = mockChildren[0];
       firstChild.exitCode = 1;
       firstChild.connected = false;
@@ -264,34 +273,33 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
     });
   });
 
-  describe("command events are persisted (for crash recovery replay)", () => {
+  describe("workflow messages are persisted (for crash recovery replay)", () => {
-    it("persists thread_command_event when worker sends thread-command-event IPC", async () => {
+    it("persists thread_workflow_message when worker sends thread-workflow-message 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 });
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
       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",
+        type: "thread-workflow-message",
         runId,
-        event: { type: "scan_complete", items: ["a", "b"] },
+        message: { role: "scanner", content: "done", meta: { items: ["a", "b"] }, timestamp: 1000 },
       });
 
       const appendCalls = logStore.append.mock.calls.filter(
-        ([entry]: [{ type: string }]) => entry.type === "thread_command_event",
+        (args: any[]) => (args[0] as { type: string }).type === "thread_workflow_message",
       );
       expect(appendCalls).toHaveLength(1);
       expect(appendCalls[0][0]).toMatchObject({
         source: "workflow",
-        type: "thread_command_event",
+        type: "thread_workflow_message",
         refId: runId,
       });
 
@@ -309,17 +317,21 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      const payload = { task: "build-docker", repo: "myrepo" };
+      const launch = { prompt: "build-docker for myrepo", maxRounds: 10, dryRun: false };
-      mgr.startWorkflow("my-wf", payload);
+      mgr.startWorkflow("my-wf", launch);
 
       const startedCall = logStore.upsertWorkflowRun.mock.calls.find(
-        ([entry]: [{ type: string }]) => entry.type === "started",
+        (args: any[]) => (args[0] as { type: string }).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);
+      expect(parsed).toMatchObject({
+        prompt: "build-docker for myrepo",
+        maxRounds: 10,
+        dryRun: false,
+      });
 
       const stopPromise = mgr.stop();
       await vi.runAllTimersAsync();
@@ -341,7 +353,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
       // Start one thread to fill the concurrency slot
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       const firstChild = mockChildren[0];
 
       // Crash once → respawn → crash again → second respawn
@@ -369,7 +381,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
         { 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.getThreadMessages.mockReturnValue([]);
       logStore.getTriggerPayload.mockReturnValue({ s: 1 });
 
       const config = makeConfig({
@@ -377,7 +389,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-wf", {});
+      mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
       const firstChild = mockChildren[0];
       firstChild.exitCode = 1;
       firstChild.connected = false;
@@ -407,7 +419,7 @@ describe("WorkflowManager — crash recovery (Phase 3)", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("crash-wf", {});
+      mgr.startWorkflow("crash-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
       // Crash the worker 6 times in rapid succession (within CRASH_WINDOW_MS = 60s)
       for (let i = 0; i < 6; i++) {

packages/daemon/src/__tests__/daemon-ipc.test.ts

@@ -2,7 +2,7 @@
  * Unit + integration tests for daemon-ipc.ts — trigger-sense request type.
  *
  * Tests cover:
- *   - parseRequest correctly accepts/rejects trigger-sense messages
+ *   - parseDaemonIpcRequest (core) / server correctly accept or reject trigger-sense messages
  *   - createDaemonIpcServer routes trigger-sense to opts.triggerSense
  *   - Error response when triggerSense throws (unknown sense)
  *   - Success response on valid sense trigger
@@ -63,7 +63,10 @@ function sendRaw(path: string, message: object): Promise<object> {
 }
 
 beforeEach(() => {
-  sockPath = join(tmpdir(), `nerve-ipc-test-${Date.now()}-${Math.random().toString(36).slice(2)}.sock`);
+  sockPath = join(
+    tmpdir(),
+    `nerve-ipc-test-${Date.now()}-${Math.random().toString(36).slice(2)}.sock`,
+  );
 });
 
 afterEach(async () => {
@@ -149,12 +152,18 @@ describe("daemon-ipc — trigger-sense", () => {
     const resp = await sendRaw(sockPath, {
       type: "trigger-workflow",
       workflow: "my-workflow",
-      payload: {},
+      prompt: "test prompt",
+      maxRounds: 10,
+      dryRun: false,
     });
 
     expect(resp).toEqual({ ok: true });
     expect(triggerSense).not.toHaveBeenCalled();
-    expect(wfManager.startWorkflow).toHaveBeenCalledWith("my-workflow", {});
+    expect(wfManager.startWorkflow).toHaveBeenCalledWith("my-workflow", {
+      prompt: "test prompt",
+      maxRounds: 10,
+      dryRun: false,
+    });
   });
 
   it("responds ok:false for completely unknown request type", async () => {

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

@@ -62,7 +62,7 @@ const { createWorkflowManager } = await import("../workflow-manager.js");
 const { createKernel } = await import("../kernel.js");
 
 function makeWfConfig(workflows: Record<string, WorkflowConfig> = {}): NerveConfig {
-  return { senses: {}, reflexes: [], workflows };
+  return { senses: {}, reflexes: [], workflows, maxRounds: 10 };
 }
 
 function makeLogStore() {
@@ -77,8 +77,12 @@ function makeLogStore() {
     getActiveWorkflowRuns: vi.fn(() => []),
     getTriggerPayload: vi.fn(() => null),
     getThreadEvents: vi.fn(() => []),
+    getThreadMessages: vi.fn(() => []),
+    getThreadRoundCount: vi.fn(() => 0),
+    getThreadRounds: vi.fn(() => []),
     archiveLogs: vi.fn(() => ({ days: [], vacuumed: false })),
     close: vi.fn(),
+    getAllWorkflowRuns: vi.fn(() => []),
   };
 }
 
@@ -98,7 +102,7 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
     const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-    mgr.startWorkflow("my-wf", {});
+    mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     expect(mockChildren).toHaveLength(1);
 
     // Remove workflow from config before drain completes
@@ -117,8 +121,8 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     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", { prompt: "test", maxRounds: 10, dryRun: false });
-    mgr.startWorkflow("my-wf", { n: 2 });
+    mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     expect(mgr.activeCount("my-wf")).toBe(2);
 
     const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
@@ -126,7 +130,7 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     await drainPromise;
 
     const interruptedCalls = logStore.upsertWorkflowRun.mock.calls.filter(
-      ([entry]: [{ type: string }]) => entry.type === "interrupted",
+      (args: any[]) => (args[0] as { type: string }).type === "interrupted",
     );
     expect(interruptedCalls).toHaveLength(2);
 
@@ -149,7 +153,7 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
     const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-    mgr.startWorkflow("my-wf", {});
+    mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     expect(mockChildren).toHaveLength(1);
 
     const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
@@ -165,7 +169,7 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
     const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-    mgr.startWorkflow("my-wf", {});
+    mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     expect(mockChildren).toHaveLength(1);
 
     const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
@@ -182,7 +186,7 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
     const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-    mgr.startWorkflow("my-wf", {});
+    mgr.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
     const drainPromise = mgr.drainAndRespawn("my-wf", 5000);
     await vi.runAllTimersAsync();
@@ -190,10 +194,10 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
 
     const newChild = mockChildren[1];
     const resumeCalls = (newChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
-      ([msg]: [unknown]) =>
+      (args: any[]) =>
-        msg !== null &&
+        args[0] !== null &&
-        typeof msg === "object" &&
+        typeof args[0] === "object" &&
-        (msg as Record<string, unknown>).type === "resume-thread",
+        (args[0] as Record<string, unknown>).type === "resume-thread",
     );
     expect(resumeCalls).toHaveLength(0);
 
@@ -207,21 +211,21 @@ describe("WorkflowManager — drainAndRespawn (Phase 3 hot reload)", () => {
     const config = makeWfConfig({ "my-wf": { concurrency: 1, overflow: "drop" } });
     const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-    mgr.startWorkflow("my-wf", { first: true });
+    mgr.startWorkflow("my-wf", { prompt: "first", maxRounds: 10, dryRun: false });
 
     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 });
+    mgr.startWorkflow("my-wf", { prompt: "second", maxRounds: 10, dryRun: false });
 
     const newChild = mockChildren[1];
     const startCalls = (newChild.send as ReturnType<typeof vi.fn>).mock.calls.filter(
-      ([msg]: [unknown]) =>
+      (args: any[]) =>
-        msg !== null &&
+        args[0] !== null &&
-        typeof msg === "object" &&
+        typeof args[0] === "object" &&
-        (msg as Record<string, unknown>).type === "start-thread",
+        (args[0] as Record<string, unknown>).type === "start-thread",
     );
     expect(startCalls).toHaveLength(1);
 
@@ -246,8 +250,9 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     const logStore = makeLogStore();
     const config: NerveConfig = {
       senses: {},
-      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null }],
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null } as any],
       workflows: { "my-wf": { concurrency: 1, overflow: "drop" } },
+      maxRounds: 10,
     };
 
     const kernel = createKernel(config, "/tmp/nerve-hot-reload-test", {
@@ -256,7 +261,7 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     });
 
     // Trigger a workflow thread so a worker is spawned
-    kernel.workflowManager.startWorkflow("my-wf", {});
+    kernel.workflowManager.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
     // Manually call drainAndRespawn (simulating what kernel does on workflow file change)
     const drainPromise = kernel.workflowManager.drainAndRespawn("my-wf", 1000);
@@ -266,7 +271,9 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     // 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");
+    const startCall = appendCalls.find(
+      (args: any[]) => (args[0] as { type: string }).type === "start",
+    );
     expect(startCall).toBeDefined();
 
     const stopPromise = kernel.stop();
@@ -278,8 +285,9 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     const logStore = makeLogStore();
     const initialConfig: NerveConfig = {
       senses: {},
-      reflexes: [{ kind: "workflow", workflow: "old-wf", on: null }],
+      reflexes: [{ kind: "workflow", workflow: "old-wf", on: null } as any],
       workflows: { "old-wf": { concurrency: 1, overflow: "drop" } },
+      maxRounds: 10,
     };
 
     const kernel = createKernel(initialConfig, "/tmp/nerve-hot-reload-test", {
@@ -288,7 +296,11 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     });
 
     // Spawn a worker for old-wf
-    kernel.workflowManager.startWorkflow("old-wf", {});
+    kernel.workflowManager.startWorkflow("old-wf", {
+      prompt: "test",
+      maxRounds: 10,
+      dryRun: false,
+    });
     expect(mockChildren).toHaveLength(1);
 
     // Reload config without old-wf
@@ -296,6 +308,7 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
       senses: {},
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
     kernel.reloadConfig(newConfig);
 
@@ -315,8 +328,9 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     const logStore = makeLogStore();
     const initialConfig: NerveConfig = {
       senses: {},
-      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null }],
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null } as any],
       workflows: { "my-wf": { concurrency: 1, overflow: "drop" } },
+      maxRounds: 10,
     };
 
     const kernel = createKernel(initialConfig, "/tmp/nerve-hot-reload-test", {
@@ -324,14 +338,15 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
       logStore,
     });
 
-    kernel.workflowManager.startWorkflow("my-wf", {});
+    kernel.workflowManager.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     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 }],
+      reflexes: [{ kind: "workflow", workflow: "my-wf", on: null } as any],
       workflows: { "my-wf": { concurrency: 5, overflow: "queue", maxQueue: 50 } },
+      maxRounds: 10,
     };
     kernel.reloadConfig(newConfig);
 
@@ -343,8 +358,8 @@ describe("Kernel — workflow hot reload via file-watcher (Phase 3)", () => {
     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", { prompt: "test", maxRounds: 10, dryRun: false });
-    kernel.workflowManager.startWorkflow("my-wf", { n: 3 });
+    kernel.workflowManager.startWorkflow("my-wf", { prompt: "test", maxRounds: 10, dryRun: false });
     expect(kernel.workflowManager.activeCount("my-wf")).toBe(3);
 
     const stopPromise = kernel.stop();

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

@@ -27,6 +27,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }

packages/daemon/src/__tests__/kernel-phase6.test.ts

@@ -74,6 +74,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -180,6 +181,7 @@ describe("kernel — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     });
 
     expect(kernel.groups.has("network")).toBe(true);
@@ -196,6 +198,7 @@ describe("kernel — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
     const kernel = createKernel(config, "/tmp/nerve-test");
 
@@ -210,6 +213,7 @@ describe("kernel — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     });
 
     expect(kernel.groups.has("network")).toBe(false);
@@ -232,6 +236,7 @@ describe("kernel — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     });
 
     expect(kernel.getHealth().activeSenses).toBe(2);

packages/daemon/src/__tests__/kernel-trigger-sense.test.ts

@@ -2,7 +2,7 @@
  * Unit tests for kernel.triggerSense() — IPC issue #36.
  *
  * These tests use a mock child_process and a mock LogStore so they do NOT
- * require better-sqlite3 to be present in the test environment.
+ * require a real LogStore (node:sqlite) in integration tests.
  */
 
 import { EventEmitter } from "node:events";
@@ -58,7 +58,7 @@ vi.mock("node:child_process", () => ({
 const { createKernel } = await import("../kernel.js");
 
 // ---------------------------------------------------------------------------
-// Mock LogStore factory (avoids better-sqlite3 dependency)
+// Mock LogStore factory (avoids SQLite I/O in this unit test)
 // ---------------------------------------------------------------------------
 
 function makeMockLogStore() {
@@ -74,6 +74,9 @@ function makeMockLogStore() {
     getAllWorkflowRuns: vi.fn(() => []),
     getTriggerPayload: vi.fn(() => null),
     getThreadEvents: vi.fn(() => []),
+    getThreadMessages: vi.fn(() => []),
+    getThreadRoundCount: vi.fn(() => 0),
+    getThreadRounds: vi.fn(() => []),
     archiveLogs: vi.fn(() => ({ days: [], vacuumed: false })),
     close: vi.fn(),
   };
@@ -90,6 +93,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -191,9 +195,7 @@ describe("kernel.triggerSense()", () => {
 
     // Should not throw even when the worker is disconnected
     expect(() => kernel.triggerSense("cpu-usage")).not.toThrow();
-    expect(worker.send).not.toHaveBeenCalledWith(
+    expect(worker.send).not.toHaveBeenCalledWith(expect.objectContaining({ type: "compute" }));
-      expect.objectContaining({ type: "compute" }),
-    );
 
     await kernel.stop();
   }, 10_000);

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

@@ -78,8 +78,12 @@ function makeLogStore() {
     appendWithWorkflowUpdate: vi.fn(),
     getWorkflowRun: vi.fn(() => null),
     getActiveWorkflowRuns: vi.fn(() => []),
+    getAllWorkflowRuns: vi.fn(() => []),
     getTriggerPayload: vi.fn(() => null),
     getThreadEvents: vi.fn(() => []),
+    getThreadMessages: vi.fn(() => []),
+    getThreadRoundCount: vi.fn(() => 0),
+    getThreadRounds: vi.fn(() => []),
     archiveLogs: vi.fn(() => ({ days: [], vacuumed: false })),
     close: vi.fn(),
   };
@@ -92,6 +96,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -111,14 +116,14 @@ describe("kernel + workflowManager integration", () => {
     vi.clearAllMocks();
   });
 
-  describe("sense signal triggers workflow via reflex", () => {
+  describe("sense compute triggers workflow via return value", () => {
-    it("calls workflowManager.startWorkflow when a sense signal fires on a workflow reflex", async () => {
+    it("calls workflowManager.startWorkflow when a sense compute returns a workflow launch", 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"] }],
+        reflexes: [],
         workflows: { "my-workflow": { concurrency: 2, overflow: "drop" } },
       });
 
@@ -127,20 +132,26 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      // Emit a signal from "cpu-usage" on the bus
+      // Simulate a sense worker sending a signal with workflow launch payload
-      const { createSignalBus } = await import("../signal-bus.js");
+      // The kernel's handleWorkerMessage processes "signal" type messages
-      void createSignalBus; // ensure import resolves
+      // and uses routeSenseComputeOutput to detect workflow launches
-      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: { value: 80 }, ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        // Simulate the worker sending a signal message with workflow field
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "my-workflow|10|run this workflow" },
+        });
+      }
 
-      // The workflow worker should be spawned (one for the sense group, one for workflow)
+      // A workflow worker should be spawned and a start-thread message sent
-      // 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]) =>
+          (args: unknown[]) =>
-            msg !== null &&
+            args[0] !== null &&
-            typeof msg === "object" &&
+            typeof args[0] === "object" &&
-            (msg as Record<string, unknown>).type === "start-thread",
+            (args[0] as Record<string, unknown>).type === "start-thread",
         ),
       );
       expect(workflowWorker).toBeDefined();
@@ -150,13 +161,13 @@ describe("kernel + workflowManager integration", () => {
       await stopPromise;
     });
 
-    it("passes the signal payload as triggerPayload to the workflow", async () => {
+    it("passes prompt and maxRounds from the workflow field 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"] }],
+        reflexes: [],
         workflows: { "alert-workflow": { concurrency: 1, overflow: "drop" } },
       });
 
@@ -165,8 +176,15 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      const payload = { level: "critical", value: 99 };
+      // Simulate sense worker returning a workflow launch
-      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload, ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "alert-workflow|5|handle critical alert" },
+        });
+      }
 
       // Find the start-thread call and verify triggerPayload
       const startThreadCall = mockChildren
@@ -182,7 +200,9 @@ describe("kernel + workflowManager integration", () => {
       expect(startThreadCall?.[0]).toMatchObject({
         type: "start-thread",
         workflow: "alert-workflow",
-        triggerPayload: payload,
+        prompt: "handle critical alert",
+        maxRounds: 5,
+        dryRun: false,
       });
 
       const stopPromise = kernel.stop();
@@ -197,7 +217,7 @@ describe("kernel + workflowManager integration", () => {
           "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"] }],
+        reflexes: [],
         workflows: { "my-workflow": { concurrency: 1, overflow: "drop" } },
       });
 
@@ -206,16 +226,23 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      // Emit signal from cpu-usage — NOT in the workflow's "on" list
+      // Emit a regular signal (no workflow field) — should NOT trigger any workflow
-      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: 50, ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: 50,
+        });
+      }
 
-      // No workflow worker should have been spawned (only the sense group worker)
+      // No workflow should have been started
       const workflowWorkerSpawned = mockChildren.some((c) =>
         (c.send as ReturnType<typeof vi.fn>).mock.calls.some(
-          ([msg]: [unknown]) =>
+          (args: unknown[]) =>
-            msg !== null &&
+            args[0] !== null &&
-            typeof msg === "object" &&
+            typeof args[0] === "object" &&
-            (msg as Record<string, unknown>).type === "start-thread",
+            (args[0] as Record<string, unknown>).type === "start-thread",
         ),
       );
       expect(workflowWorkerSpawned).toBe(false);
@@ -227,13 +254,13 @@ describe("kernel + workflowManager integration", () => {
   });
 
   describe("workflow events are logged", () => {
-    it("logs a 'started' event when workflow thread is triggered", async () => {
+    it("logs a 'started' event when workflow thread is triggered via sense compute", 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"] }],
+        reflexes: [],
         workflows: { "log-test-workflow": { concurrency: 2, overflow: "drop" } },
       });
 
@@ -242,7 +269,15 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: null, ts: Date.now() });
+      // Simulate sense compute returning a workflow launch
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "log-test-workflow|10|test prompt" },
+        });
+      }
 
       expect(logStore.upsertWorkflowRun).toHaveBeenCalledWith(
         expect.objectContaining({ source: "workflow", type: "started" }),
@@ -256,7 +291,7 @@ describe("kernel + workflowManager integration", () => {
   });
 
   describe("reloadConfig handles workflow changes", () => {
-    it("new workflow reflexes are active after reloadConfig", async () => {
+    it("new workflows are available after reloadConfig", async () => {
       const logStore = makeLogStore();
       const initialConfig = makeConfig({
         senses: {
@@ -264,6 +299,7 @@ describe("kernel + workflowManager integration", () => {
         },
         reflexes: [],
         workflows: null,
+        maxRounds: 10,
       });
 
       const kernel = createKernel(initialConfig, "/tmp/nerve-test", {
@@ -271,18 +307,26 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      // Reload with a workflow reflex added
+      // Reload with a workflow added
       const newConfig: NerveConfig = {
         senses: {
           "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
         },
-        reflexes: [{ kind: "workflow", workflow: "new-workflow", on: ["cpu-usage"] }],
+        reflexes: [],
         workflows: { "new-workflow": { concurrency: 1, overflow: "drop" } },
+        maxRounds: 10,
       };
       kernel.reloadConfig(newConfig);
 
-      // Now emit a signal — should trigger the new workflow
+      // Simulate sense compute returning a workflow launch for the new workflow
-      kernel.bus.emit({ id: 2, senseId: "cpu-usage", payload: "reload-test", ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "new-workflow|10|reload test" },
+        });
+      }
 
       const startThreadCall = mockChildren
         .flatMap((c) => (c.send as ReturnType<typeof vi.fn>).mock.calls as [unknown][])
@@ -301,13 +345,13 @@ describe("kernel + workflowManager integration", () => {
       await stopPromise;
     });
 
-    it("old workflow reflexes are removed after reloadConfig", async () => {
+    it("old workflows 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"] }],
+        reflexes: [],
         workflows: { "old-workflow": { concurrency: 1, overflow: "drop" } },
       });
 
@@ -316,13 +360,14 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      // Reload with the workflow reflex removed
+      // Reload with the workflow removed
       const newConfig: NerveConfig = {
         senses: {
           "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
         },
         reflexes: [],
         workflows: null,
+        maxRounds: 10,
       };
       kernel.reloadConfig(newConfig);
 
@@ -331,8 +376,15 @@ describe("kernel + workflowManager integration", () => {
         (c.send as ReturnType<typeof vi.fn>).mockClear();
       }
 
-      // Emit a signal — old-workflow should NOT be triggered
+      // Simulate sense compute trying to launch the old workflow — it should still not start
-      kernel.bus.emit({ id: 3, senseId: "cpu-usage", payload: "after-reload", ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "old-workflow|10|should not work" },
+        });
+      }
 
       const startThreadCall = mockChildren
         .flatMap((c) => (c.send as ReturnType<typeof vi.fn>).mock.calls as [unknown][])
@@ -358,7 +410,7 @@ describe("kernel + workflowManager integration", () => {
         senses: {
           "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
         },
-        reflexes: [{ kind: "workflow", workflow: "shutdown-test", on: ["cpu-usage"] }],
+        reflexes: [],
         workflows: { "shutdown-test": { concurrency: 1, overflow: "drop" } },
       });
 
@@ -367,8 +419,15 @@ describe("kernel + workflowManager integration", () => {
         logStore,
       });
 
-      // Trigger a workflow so a worker is spawned
+      // Trigger a workflow via sense compute return value
-      kernel.bus.emit({ id: 1, senseId: "cpu-usage", payload: null, ts: Date.now() });
+      const workerPool = mockChildren[0];
+      if (workerPool) {
+        workerPool.emit("message", {
+          type: "signal",
+          sense: "cpu-usage",
+          payload: { workflow: "shutdown-test|10|test" },
+        });
+      }
 
       const stopPromise = kernel.stop();
       await vi.runAllTimersAsync();
@@ -400,7 +459,7 @@ describe("kernel + workflowManager integration", () => {
         senses: {
           "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
         },
-        reflexes: [{ kind: "workflow", workflow: "health-wf", on: ["cpu-usage"] }],
+        reflexes: [],
         workflows: { "health-wf": { concurrency: 2, overflow: "drop" } },
       });
 

packages/daemon/src/__tests__/kernel.test.ts

@@ -47,7 +47,7 @@ vi.mock("node:child_process", () => ({
 
 // Import after mock is set up
 const { createKernel } = await import("../kernel.js");
-const { createLogStore } = await import("../log-store.js");
+const { createLogStore } = await import("@uncaged/nerve-store");
 
 // ---------------------------------------------------------------------------
 // Helpers
@@ -60,6 +60,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -200,6 +201,7 @@ describe("kernel — groupForSense mapping", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
     const kernel = createKernel(config, "/tmp/nerve-test");
 

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

@@ -4,8 +4,8 @@ import { join } from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 
 import type { NerveConfig, Signal } from "@uncaged/nerve-core";
-import { createLogStore } from "../log-store.js";
+import { createLogStore } from "@uncaged/nerve-store";
-import type { LogStore } from "../log-store.js";
+import type { LogStore } from "@uncaged/nerve-store";
 import { createReflexScheduler } from "../reflex-scheduler.js";
 import { createSignalBus } from "../signal-bus.js";
 
@@ -30,6 +30,7 @@ describe("LogStore + ReflexScheduler integration", () => {
       },
       reflexes: [{ kind: "sense", sense: "cpu-usage", interval: null, on: ["cpu-usage"] }],
       workflows: null,
+      maxRounds: 10,
     };
     const bus = createSignalBus();
     const triggered: string[] = [];
@@ -57,6 +58,7 @@ describe("LogStore + ReflexScheduler integration", () => {
       },
       reflexes: [{ kind: "sense", sense: "cpu-usage", interval: 1000, on: null }],
       workflows: null,
+      maxRounds: 10,
     };
     const bus = createSignalBus();
     const ref: { scheduler: ReturnType<typeof createReflexScheduler> | null } = { scheduler: null };
@@ -87,6 +89,7 @@ describe("LogStore + ReflexScheduler integration", () => {
       },
       reflexes: [{ kind: "sense", sense: "cpu-usage", interval: null, on: ["cpu-usage"] }],
       workflows: null,
+      maxRounds: 10,
     };
     const bus = createSignalBus();
     const triggered: string[] = [];

packages/daemon/src/__tests__/phase6-integration.test.ts

@@ -24,6 +24,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -136,6 +137,7 @@ describe("phase6 — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
 
     kernel.reloadConfig(newConfig);
@@ -155,6 +157,7 @@ describe("phase6 — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
     kernel = createKernel(config, "/tmp/nerve-phase6-test", {
       workerScript: MOCK_WORKER,
@@ -169,6 +172,7 @@ describe("phase6 — reloadConfig", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
 
     kernel.reloadConfig(newConfig);
@@ -199,6 +203,7 @@ describe("phase6 — error isolation", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
 
     kernel = createKernel(config, "/tmp/nerve-phase6-test", {
@@ -302,6 +307,7 @@ describe("phase6 — getHealth", () => {
       },
       reflexes: [],
       workflows: null,
+      maxRounds: 10,
     };
     kernel.reloadConfig(newConfig);
 

packages/daemon/src/__tests__/reflex-scheduler-throttle-pending.test.ts

@@ -11,6 +11,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }

packages/daemon/src/__tests__/reflex-scheduler.test.ts

@@ -17,6 +17,7 @@ function makeConfig(overrides: Partial<NerveConfig> = {}): NerveConfig {
     },
     reflexes: [],
     workflows: null,
+    maxRounds: 10,
     ...overrides,
   };
 }
@@ -290,10 +291,11 @@ describe("ReflexScheduler — workflow reflexes ignored", () => {
   it("does not set up any scheduling for workflow kind reflexes", () => {
     const triggered: string[] = [];
     const config: NerveConfig = {
+      maxRounds: 10,
       senses: {
         "cpu-usage": { group: "system", throttle: null, timeout: null, gracePeriod: null },
       },
-      reflexes: [{ kind: "workflow", workflow: "my-workflow", on: ["cpu-usage"] }],
+      reflexes: [{ kind: "workflow", workflow: "my-workflow", on: ["cpu-usage"] } as any],
       workflows: {
         "my-workflow": { concurrency: 1, overflow: "drop" },
       },

packages/daemon/src/__tests__/sense-runtime.test.ts

@@ -2,15 +2,15 @@ import { mkdirSync, mkdtempSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
-import Database from "better-sqlite3";
+import { DatabaseSync } from "node:sqlite";
-import { drizzle } from "drizzle-orm/better-sqlite3";
+import { drizzle } from "drizzle-orm/node-sqlite";
 import { integer, real, sqliteTable } from "drizzle-orm/sqlite-core";
 import { describe, expect, it } from "vitest";
 
-import { createBlobStore } from "../blob-store.js";
+import { createBlobStore } from "@uncaged/nerve-store";
 import { parseParentMessage } from "../ipc.js";
 import { executeCompute, openPeerDb, openSenseDb, runMigrations } from "../sense-runtime.js";
-import type { DrizzleDB, PeerMap, SenseRuntime } from "../sense-runtime.js";
+import type { ComputeFn, DrizzleDB, PeerMap, SenseRuntime } from "../sense-runtime.js";
 
 // ---------------------------------------------------------------------------
 // Helpers
@@ -49,7 +49,7 @@ const samples = sqliteTable("samples", {
 
 describe("runMigrations", () => {
   it("creates table via SQL migration file", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const migrationsDir = makeTempMigrationsDir(INIT_SQL);
     const result = runMigrations(sqlite, migrationsDir);
 
@@ -64,7 +64,7 @@ describe("runMigrations", () => {
   });
 
   it("runs multiple migrations in lexicographic order", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const dir = mkdtempSync(join(tmpdir(), "nerve-multi-"));
 
     writeFileSync(join(dir, "0001_init.sql"), INIT_SQL);
@@ -81,7 +81,7 @@ describe("runMigrations", () => {
   });
 
   it("returns ok when migrations directory is empty", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const dir = makeTempMigrationsDirEmpty();
     const result = runMigrations(sqlite, dir);
     expect(result.ok).toBe(true);
@@ -89,14 +89,14 @@ describe("runMigrations", () => {
   });
 
   it("returns err when migrations directory does not exist", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const result = runMigrations(sqlite, "/nonexistent/path/migrations");
     expect(result.ok).toBe(false);
     sqlite.close();
   });
 
   it("returns err when a migration SQL is invalid", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const dir = mkdtempSync(join(tmpdir(), "nerve-bad-sql-"));
     writeFileSync(join(dir, "0001_bad.sql"), "NOT VALID SQL !!!;");
     const result = runMigrations(sqlite, dir);
@@ -141,7 +141,7 @@ describe("openPeerDb", () => {
   it("opens an existing db in read-only mode", () => {
     // Create a writable db first
     const dbPath = makeTempDbPath();
-    const sqlite = new Database(dbPath);
+    const sqlite = new DatabaseSync(dbPath);
     sqlite.exec(INIT_SQL);
     sqlite.prepare("INSERT INTO samples (ts, value) VALUES (1, 42.0)").run();
     sqlite.close();
@@ -168,13 +168,13 @@ describe("openPeerDb", () => {
 // ---------------------------------------------------------------------------
 
 describe("executeCompute", () => {
-  function makeRuntime(computeFn: (db: DrizzleDB, peers: PeerMap) => Promise<unknown | null>): {
+  function makeRuntime(computeFn: ComputeFn): {
     runtime: SenseRuntime;
-    sqlite: Database.Database;
+    sqlite: DatabaseSync;
   } {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     sqlite.exec(INIT_SQL);
-    const db = drizzle(sqlite) as DrizzleDB;
+    const db = drizzle({ client: sqlite }) as DrizzleDB;
     return {
       runtime: { name: "test-sense", db, compute: computeFn },
       sqlite,
@@ -226,10 +226,10 @@ describe("executeCompute", () => {
 
   it("compute can read from peers", async () => {
     // Set up a peer db with data
-    const peerSqlite = new Database(":memory:");
+    const peerSqlite = new DatabaseSync(":memory:");
     peerSqlite.exec(INIT_SQL);
     peerSqlite.prepare("INSERT INTO samples (ts, value) VALUES (100, 3.14)").run();
-    const peerDb = drizzle(peerSqlite) as DrizzleDB;
+    const peerDb = drizzle({ client: peerSqlite }) as DrizzleDB;
 
     const peers: PeerMap = { "other-sense": peerDb };
 
@@ -248,9 +248,9 @@ describe("executeCompute", () => {
   });
 
   it("write to own db does not affect peer db (isolation)", async () => {
-    const peerSqlite = new Database(":memory:");
+    const peerSqlite = new DatabaseSync(":memory:");
     peerSqlite.exec(INIT_SQL);
-    const peerDb = drizzle(peerSqlite) as DrizzleDB;
+    const peerDb = drizzle({ client: peerSqlite }) as DrizzleDB;
     const peers: PeerMap = { "peer-sense": peerDb };
 
     const { runtime, sqlite } = makeRuntime(async (db) => {
@@ -403,7 +403,7 @@ describe("parseParentMessage", () => {
 
 describe("runMigrations journal", () => {
   it("does not re-run an already-applied migration", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const dir = mkdtempSync(join(tmpdir(), "nerve-journal-"));
     writeFileSync(join(dir, "0001_init.sql"), INIT_SQL);
 
@@ -430,7 +430,7 @@ describe("runMigrations journal", () => {
   });
 
   it("tracks migrations in _migrations table", () => {
-    const sqlite = new Database(":memory:");
+    const sqlite = new DatabaseSync(":memory:");
     const dir = mkdtempSync(join(tmpdir(), "nerve-journal2-"));
     writeFileSync(join(dir, "0001_init.sql"), INIT_SQL);
 

packages/daemon/src/__tests__/worker-pool.test.ts

@@ -0,0 +1,235 @@
+import { EventEmitter } from "node:events";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+const mockChildren: MockChild[] = [];
+
+type MockChild = EventEmitter & {
+  send: ReturnType<typeof vi.fn>;
+  kill: ReturnType<typeof vi.fn>;
+  pid: number;
+  connected: boolean;
+};
+
+function makeMockChild(pid = 1): MockChild {
+  const child = new EventEmitter() as MockChild;
+  child.connected = true;
+  child.send = vi.fn((msg: unknown) => {
+    if (
+      msg !== null &&
+      typeof msg === "object" &&
+      (msg as Record<string, unknown>).type === "shutdown"
+    ) {
+      child.connected = false;
+      setImmediate(() => child.emit("exit", 0, null));
+    }
+  });
+  child.kill = vi.fn((_signal?: string) => {
+    child.connected = false;
+    child.emit("exit", null, _signal ?? "SIGKILL");
+  });
+  child.pid = pid;
+  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 { createSenseWorkerPool } = await import("../worker-pool.js");
+
+async function flushSetImmediate(): Promise<void> {
+  await new Promise<void>((resolve) => setImmediate(resolve));
+}
+
+async function startWorkerWithReady(
+  pool: ReturnType<typeof createSenseWorkerPool>,
+  group: string,
+): Promise<void> {
+  const pr = pool.startWorker(group);
+  const child = mockChildren[mockChildren.length - 1];
+  child.emit("message", { type: "ready" });
+  await pr;
+}
+
+describe("createSenseWorkerPool", () => {
+  beforeEach(() => {
+    mockChildren.length = 0;
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+  });
+
+  it("forks one child per startWorker and routes IPC to onWorkerMessage", async () => {
+    const onWorkerMessage = vi.fn();
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage,
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "g1");
+    expect(mockChildren).toHaveLength(1);
+    const child = mockChildren[0];
+    child.emit("message", { type: "signal", sense: "s", payload: 1 });
+    expect(onWorkerMessage).toHaveBeenCalledWith({ type: "signal", sense: "s", payload: 1 });
+  });
+
+  it("sendCompute delivers to the worker for that group", async () => {
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "sys");
+    const child = mockChildren[0];
+    pool.sendCompute("sys", "cpu");
+    expect(child.send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "compute", sense: "cpu" }),
+    );
+  });
+
+  it("hasWorkerForGroup and getWorkerPid reflect running workers", async () => {
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    expect(pool.hasWorkerForGroup("a")).toBe(false);
+    expect(pool.getWorkerPid("a")).toBeNull();
+
+    await startWorkerWithReady(pool, "a");
+    expect(pool.hasWorkerForGroup("a")).toBe(true);
+    expect(pool.getWorkerPid("a")).toBe(1);
+    expect(pool.activeGroupCount()).toBe(1);
+  });
+
+  it("evictGroup sends shutdown and removes the entry without waiting", async () => {
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "x");
+    expect(pool.activeGroupCount()).toBe(1);
+    pool.evictGroup("x");
+    expect(pool.hasWorkerForGroup("x")).toBe(false);
+    expect(mockChildren[0].send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "shutdown" }),
+    );
+  });
+
+  it("restartGroup invokes onBeforeGroupRestart then respawns", async () => {
+    const onBeforeGroupRestart = vi.fn();
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => ["s1"],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart,
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "g");
+    expect(mockChildren).toHaveLength(1);
+
+    const p = pool.restartGroup("g");
+    expect(onBeforeGroupRestart).toHaveBeenCalledWith("g");
+    expect(mockChildren[0].send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "shutdown" }),
+    );
+
+    await flushSetImmediate();
+    expect(mockChildren).toHaveLength(2);
+    mockChildren[1].emit("message", { type: "ready" });
+    await p;
+    expect(pool.hasWorkerForGroup("g")).toBe(true);
+  });
+
+  it("onWorkerCrashed runs and schedules respawn after non-zero exit", async () => {
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+    const onWorkerCrashed = vi.fn();
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: (g) => (g === "g" ? ["a", "b"] : []),
+      onWorkerCrashed,
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "g");
+    expect(mockChildren).toHaveLength(1);
+    mockChildren[0].emit("exit", 1, null);
+    expect(onWorkerCrashed).toHaveBeenCalledWith("g");
+
+    await vi.advanceTimersByTimeAsync(1000);
+    expect(mockChildren).toHaveLength(2);
+  });
+
+  it("shutdownAll sends shutdown to every worker", async () => {
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => false,
+    });
+
+    await startWorkerWithReady(pool, "a");
+    await startWorkerWithReady(pool, "b");
+    await pool.shutdownAll();
+    expect(mockChildren[0].send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "shutdown" }),
+    );
+    expect(mockChildren[1].send).toHaveBeenCalledWith(
+      expect.objectContaining({ type: "shutdown" }),
+    );
+  });
+
+  it("does not respawn after crash when isStopped is true", async () => {
+    vi.useFakeTimers({ shouldAdvanceTime: true });
+    const pool = createSenseWorkerPool({
+      nerveRoot: "/tmp/n",
+      workerScript: "/fake/sense-worker.js",
+      onWorkerMessage: vi.fn(),
+      sensesForGroup: () => [],
+      onWorkerCrashed: vi.fn(),
+      onBeforeGroupRestart: vi.fn(),
+      isStopped: () => true,
+    });
+
+    await startWorkerWithReady(pool, "g");
+    const n = mockChildren.length;
+    mockChildren[0].emit("exit", 1, null);
+    await vi.advanceTimersByTimeAsync(1000);
+    expect(mockChildren.length).toBe(n);
+  });
+});

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

@@ -74,13 +74,18 @@ function makeLogStore() {
     getActiveWorkflowRuns: vi.fn(() => []),
     getTriggerPayload: vi.fn(() => null),
     getThreadEvents: vi.fn(() => []),
+    getThreadMessages: vi.fn(() => []),
+    getThreadRoundCount: vi.fn(() => 0),
+    getThreadRounds: vi.fn(() => []),
     archiveLogs: vi.fn(() => ({ days: [], vacuumed: false })),
     close: vi.fn(),
+    getAllWorkflowRuns: vi.fn(() => []),
   };
 }
 
 function makeConfig(overrides: Partial<NerveConfig["workflows"]> = {}): NerveConfig {
   return {
+    maxRounds: 10,
     senses: {},
     reflexes: [],
     workflows: overrides as NerveConfig["workflows"],
@@ -110,7 +115,7 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-workflow", { event: "test" });
+      mgr.startWorkflow("my-workflow", { prompt: "test", maxRounds: 10, dryRun: false });
 
       expect(mockChildren).toHaveLength(1);
       expect(mockChildren[0].send).toHaveBeenCalledWith(
@@ -126,8 +131,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-workflow", { n: 1 });
+      mgr.startWorkflow("my-workflow", { prompt: "test 1", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("my-workflow", { n: 2 });
+      mgr.startWorkflow("my-workflow", { prompt: "test 2", maxRounds: 10, dryRun: false });
 
       // Only one forked child — worker is reused
       expect(mockChildren).toHaveLength(1);
@@ -142,7 +147,7 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("my-workflow", { x: 1 });
+      mgr.startWorkflow("my-workflow", { prompt: "test", maxRounds: 10, dryRun: false });
 
       expect(logStore.upsertWorkflowRun).toHaveBeenCalledWith(
         expect.objectContaining({ source: "workflow", type: "started" }),
@@ -159,9 +164,9 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("drop-wf", { first: true });
+      mgr.startWorkflow("drop-wf", { prompt: "first", maxRounds: 10, dryRun: false });
       // now at limit — second call should be dropped
-      mgr.startWorkflow("drop-wf", { second: true });
+      mgr.startWorkflow("drop-wf", { prompt: "second", maxRounds: 10, dryRun: false });
 
       expect(mgr.activeCount("drop-wf")).toBe(1);
       expect(mgr.queueLength("drop-wf")).toBe(0);
@@ -176,8 +181,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("drop-wf", {});
+      mgr.startWorkflow("drop-wf", { prompt: "test", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("drop-wf", {});
+      mgr.startWorkflow("drop-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
       const droppedCall = logStore.upsertWorkflowRun.mock.calls.find(
         ([entry]) => entry.type === "dropped",
@@ -194,8 +199,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { prompt: "first", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("queue-wf", { second: true });
+      mgr.startWorkflow("queue-wf", { prompt: "second", maxRounds: 10, dryRun: false });
 
       expect(mgr.activeCount("queue-wf")).toBe(1);
       expect(mgr.queueLength("queue-wf")).toBe(1);
@@ -208,8 +213,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("queue-wf", {});
+      mgr.startWorkflow("queue-wf", { prompt: "test", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("queue-wf", {});
+      mgr.startWorkflow("queue-wf", { prompt: "test", maxRounds: 10, dryRun: false });
 
       const queuedCall = logStore.upsertWorkflowRun.mock.calls.find(
         ([entry]) => entry.type === "queued",
@@ -228,12 +233,12 @@ describe("WorkflowManager", () => {
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
       // Fill the concurrency slot
-      mgr.startWorkflow("queue-wf", { n: 0 });
+      mgr.startWorkflow("queue-wf", { prompt: "test 0", maxRounds: 10, dryRun: false });
       // Fill the queue to maxQueue
-      mgr.startWorkflow("queue-wf", { n: 1 });
+      mgr.startWorkflow("queue-wf", { prompt: "test 1", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("queue-wf", { n: 2 });
+      mgr.startWorkflow("queue-wf", { prompt: "test 2", maxRounds: 10, dryRun: false });
-      // This one should push out { n: 1 }
+      // This one should push out the oldest
-      mgr.startWorkflow("queue-wf", { n: 3 });
+      mgr.startWorkflow("queue-wf", { prompt: "test 3", maxRounds: 10, dryRun: false });
 
       // Queue should still be at maxQueue (2)
       expect(mgr.queueLength("queue-wf")).toBe(2);
@@ -254,8 +259,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { prompt: "first", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("queue-wf", { second: true });
+      mgr.startWorkflow("queue-wf", { prompt: "second", maxRounds: 10, dryRun: false });
 
       expect(mgr.activeCount("queue-wf")).toBe(1);
       expect(mgr.queueLength("queue-wf")).toBe(1);
@@ -289,8 +294,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("queue-wf", { first: true });
+      mgr.startWorkflow("queue-wf", { prompt: "first", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("queue-wf", { second: true });
+      mgr.startWorkflow("queue-wf", { prompt: "second", maxRounds: 10, dryRun: false });
 
       const child = mockChildren[0];
       const firstRunId = (child.send as ReturnType<typeof vi.fn>).mock.calls[0][0].runId as string;
@@ -316,8 +321,8 @@ describe("WorkflowManager", () => {
       });
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("wf-a", {});
+      mgr.startWorkflow("wf-a", { prompt: "test", maxRounds: 10, dryRun: false });
-      mgr.startWorkflow("wf-b", {});
+      mgr.startWorkflow("wf-b", { prompt: "test", maxRounds: 10, dryRun: false });
 
       // Two distinct workers should have been forked
       expect(mockChildren).toHaveLength(2);
@@ -343,7 +348,7 @@ describe("WorkflowManager", () => {
       await vi.runAllTimersAsync();
       await stopPromise;
 
-      mgr.startWorkflow("wf-a", {});
+      mgr.startWorkflow("wf-a", { prompt: "test", maxRounds: 10, dryRun: false });
 
       // No worker should have been spawned
       expect(mockChildren).toHaveLength(0);
@@ -356,7 +361,7 @@ describe("WorkflowManager", () => {
       const config = makeConfig({});
       const mgr = createWorkflowManager("/nerve-root", config, logStore);
 
-      mgr.startWorkflow("no-such-workflow", {});
+      mgr.startWorkflow("no-such-workflow", { prompt: "test", maxRounds: 10, dryRun: false });
 
       expect(mockChildren).toHaveLength(0);
       expect(logStore.upsertWorkflowRun).not.toHaveBeenCalled();

packages/daemon/src/daemon-ipc.ts

@@ -2,74 +2,24 @@
  * Daemon IPC server — listens on a Unix domain socket so that the CLI
  * can send commands (e.g. trigger-workflow, trigger-sense) to the running daemon process.
  *
- * Protocol: newline-delimited JSON messages.
+ * Protocol: newline-delimited JSON — request/response types and
- * Each request: { type: "trigger-workflow"; workflow: string; payload: unknown }
+ * `parseDaemonIpcRequest` live in `@uncaged/nerve-core`.
- *             | { type: "trigger-sense"; sense: string }
- *             | { type: "list-senses" }
- * Each response: { ok: true } | { ok: false; error: string }
- *              | { ok: true; senses: SenseInfo[] }  (for list-senses)
  */
 
 import { rmSync } from "node:fs";
 import { type Server, type Socket, createServer } from "node:net";
 
-import type { SenseInfo } from "@uncaged/nerve-core";
+import type { DaemonIpcResponse, SenseInfo } from "@uncaged/nerve-core";
+import { parseDaemonIpcRequest } from "@uncaged/nerve-core";
 
 import type { WorkflowManager } from "./workflow-manager.js";
 
 export type { SenseInfo };
 
-/** JSON message sent by the CLI to trigger a workflow. */
-export type TriggerWorkflowRequest = {
-  type: "trigger-workflow";
-  workflow: string;
-  payload: unknown;
-};
-
-/** JSON message sent by the CLI to trigger a sense compute on-demand. */
-export type TriggerSenseRequest = {
-  type: "trigger-sense";
-  sense: string;
-};
-
-/** JSON message sent by the CLI to list registered senses. */
-export type ListSensesRequest = {
-  type: "list-senses";
-};
-
-type DaemonRequest = TriggerWorkflowRequest | TriggerSenseRequest | ListSensesRequest;
-
-type DaemonResponse =
-  | { ok: true }
-  | { ok: false; error: string }
-  | { ok: true; senses: SenseInfo[] };
-
 export type DaemonIpcServer = {
   close: () => Promise<void>;
 };
 
-function parseRequest(line: string): DaemonRequest | null {
-  try {
-    const obj = JSON.parse(line) as unknown;
-    if (obj === null || typeof obj !== "object") return null;
-    const req = obj as Record<string, unknown>;
-    if (req.type === "trigger-workflow") {
-      if (typeof req.workflow !== "string" || req.workflow.length === 0) return null;
-      return { type: "trigger-workflow", workflow: req.workflow, payload: req.payload ?? {} };
-    }
-    if (req.type === "trigger-sense") {
-      if (typeof req.sense !== "string" || req.sense.length === 0) return null;
-      return { type: "trigger-sense", sense: req.sense };
-    }
-    if (req.type === "list-senses") {
-      return { type: "list-senses" };
-    }
-    return null;
-  } catch {
-    return null;
-  }
-}
-
 export type DaemonIpcServerOptions = {
   /** Called when a trigger-sense request arrives. Should throw if the sense is unknown. */
   triggerSense: (senseName: string) => void;
@@ -93,30 +43,37 @@ export function createDaemonIpcServer(
     const trimmed = line.trim();
     if (trimmed.length === 0) return;
 
-    const req = parseRequest(trimmed);
+    const req = parseDaemonIpcRequest(trimmed);
     if (req === null) {
-      const resp: DaemonResponse = { ok: false, error: "Invalid request" };
+      const resp: DaemonIpcResponse = { ok: false, error: "Invalid request" };
       socket.write(`${JSON.stringify(resp)}\n`);
       return;
     }
 
     try {
       if (req.type === "trigger-workflow") {
-        workflowManager.startWorkflow(req.workflow, req.payload);
+        workflowManager.startWorkflow(req.workflow, {
-        const resp: DaemonResponse = { ok: true };
+          prompt: req.prompt,
+          maxRounds: req.maxRounds,
+          dryRun: req.dryRun,
+        });
+        const resp: DaemonIpcResponse = { ok: true };
         socket.write(`${JSON.stringify(resp)}\n`);
       } else if (req.type === "trigger-sense") {
         opts.triggerSense(req.sense);
-        const resp: DaemonResponse = { ok: true };
+        const resp: DaemonIpcResponse = { ok: true };
         socket.write(`${JSON.stringify(resp)}\n`);
       } else if (req.type === "list-senses") {
         const senses = opts.listSenses();
-        const resp: DaemonResponse = { ok: true, senses };
+        const resp: DaemonIpcResponse = { ok: true, senses };
         socket.write(`${JSON.stringify(resp)}\n`);
+      } else {
+        const _exhaustive: never = req;
+        void _exhaustive;
       }
     } catch (err) {
       const msg = err instanceof Error ? err.message : String(err);
-      const resp: DaemonResponse = { ok: false, error: msg };
+      const resp: DaemonIpcResponse = { ok: false, error: msg };
       socket.write(`${JSON.stringify(resp)}\n`);
     }
   }

packages/daemon/src/index.ts

@@ -12,6 +12,7 @@ export type {
   ResumeThreadMessage,
   ThreadEventMessage,
   WorkflowErrorMessage,
+  ThreadWorkflowMessageMessage,
 } from "./ipc.js";
 
 export type { SignalBus, SignalHandler, Unsubscribe } from "./signal-bus.js";
@@ -29,25 +30,30 @@ export {
 export { createKernel } from "./kernel.js";
 export type { Kernel, KernelOptions, KernelHealth } from "./kernel.js";
 
-export type { SenseInfo } from "./daemon-ipc.js";
+export type { SenseInfo } from "@uncaged/nerve-core";
 
 export { createFileWatcher } from "./file-watcher.js";
 export type { FileWatcher, FileChange, FileChangeHandler } from "./file-watcher.js";
 
-export { createBlobStore, normalizeBlobHash } from "./blob-store.js";
+export {
-export type { BlobStore } from "./blob-store.js";
+  createBlobStore,
-
+  createLogStore,
-export { createLogStore, LOG_ARCHIVE_META_KEY } from "./log-store.js";
+  LOG_ARCHIVE_META_KEY,
+  normalizeBlobHash,
+} from "@uncaged/nerve-store";
 export type {
-  LogStore,
-  LogEntry,
-  LogQuery,
-  WorkflowRun,
-  WorkflowRunStatus,
   ArchiveLogsDayResult,
   ArchiveLogsOptions,
   ArchiveLogsResult,
-} from "./log-store.js";
+  BlobStore,
+  GetThreadRoundsParams,
+  LogEntry,
+  LogQuery,
+  LogStore,
+  ThreadRoundRow,
+  WorkflowRun,
+  WorkflowRunStatus,
+} from "@uncaged/nerve-store";
 
 export { createWorkflowManager } from "./workflow-manager.js";
 export type { WorkflowManager } from "./workflow-manager.js";

packages/daemon/src/ipc.ts

@@ -4,7 +4,7 @@
  */
 
 import type { Result } from "@uncaged/nerve-core";
-import { err, ok } from "@uncaged/nerve-core";
+import { err, isPlainRecord, ok } from "@uncaged/nerve-core";
 
 /** Parent → Worker: trigger one compute cycle for a sense */
 export type ComputeMessage = {
@@ -31,18 +31,23 @@ export type StartThreadMessage = {
   type: "start-thread";
   runId: string;
   workflow: string;
-  /** The trigger payload from the Reflex that initiated this thread. */
+  prompt: string;
-  triggerPayload: unknown;
+  /** Safety-valve: max moderator rounds for this thread (engine launch parameter). */
+  maxRounds: number;
+  /** When true, roles may skip side effects (thread-level hint on the start frame). */
+  dryRun: boolean;
 };
 
 /** Parent → Workflow Worker: resume an existing thread after crash recovery */
 export type ResumeThreadMessage = {
   type: "resume-thread";
   runId: string;
-  /** Serialised CommandEvent history to rebuild ThreadState. */
+  /** Serialised WorkflowMessage history to rebuild chain (must begin with `__start__`). */
-  events: Array<{ type: string; [key: string]: unknown }>;
+  messages: Array<{ role: string; content: string; meta: unknown; timestamp: number }>;
-  /** Serialised trigger payload (the same value as in the original start-thread). */
+  /** Safety-valve: max moderator rounds for this thread. */
-  triggerPayload: unknown;
+  maxRounds: number;
+  /** Thread-level dry-run hint (aligns with persisted `__start__` meta when replaying). */
+  dryRun: boolean;
 };
 
 /** Union of all messages the parent sends to a worker */
@@ -103,12 +108,12 @@ export type WorkflowErrorMessage = {
   error: string;
 };
 
-/** Workflow Worker → Parent: a thread CommandEvent produced by a role (for crash recovery). */
+/** Workflow Worker → Parent: a WorkflowMessage produced by a role (for crash recovery). */
-export type ThreadCommandEventMessage = {
+export type ThreadWorkflowMessageMessage = {
-  type: "thread-command-event";
+  type: "thread-workflow-message";
   runId: string;
-  /** The CommandEvent returned by role.execute() — will be persisted for crash recovery. */
+  /** The WorkflowMessage produced by the role — persisted for crash recovery. */
-  event: { type: string; [key: string]: unknown };
+  message: { role: string; content: string; meta: unknown; timestamp: number };
 };
 
 /** Union of all messages a worker sends to the parent */
@@ -119,7 +124,7 @@ export type WorkerToParentMessage =
   | HealthResponseMessage
   | ThreadEventMessage
   | WorkflowErrorMessage
-  | ThreadCommandEventMessage;
+  | ThreadWorkflowMessageMessage;
 
 const PARENT_MSG_TYPES = new Set([
   "compute",
@@ -132,89 +137,134 @@ const PARENT_MSG_TYPES = new Set([
 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'";
+  if (typeof obj.prompt !== "string") return "'start-thread' message missing string 'prompt'";
+  if (typeof obj.maxRounds !== "number") return "'start-thread' message missing number 'maxRounds'";
+  if (typeof obj.dryRun !== "boolean") return "'start-thread' message missing boolean 'dryRun'";
   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 (!Array.isArray(obj.messages)) return "'resume-thread' message missing 'messages' array";
-  if (!("triggerPayload" in obj)) return "'resume-thread' message missing 'triggerPayload'";
+  if (typeof obj.maxRounds !== "number")
+    return "'resume-thread' message missing number 'maxRounds'";
+  if (typeof obj.dryRun !== "boolean") return "'resume-thread' message missing boolean 'dryRun'";
   return null;
 }
 
 /** Validate and parse an unknown IPC message received from the parent process. */
 export function parseParentMessage(raw: unknown): Result<ParentToWorkerMessage> {
-  if (raw === null || typeof raw !== "object") {
+  if (!isPlainRecord(raw)) {
     return err(new Error("IPC message is not an object"));
   }
-  const obj = raw as Record<string, unknown>;
+  const obj = raw;
   if (typeof obj.type !== "string") {
     return err(new Error("IPC message missing string 'type' field"));
   }
   if (!PARENT_MSG_TYPES.has(obj.type)) {
     return err(new Error(`Unknown IPC message type: "${obj.type}"`));
   }
+  if (obj.type === "compute") {
+    if (typeof obj.sense !== "string") {
+      return err(new Error("IPC 'compute' message missing string 'sense' field"));
+    }
+    return ok({ type: "compute", sense: obj.sense });
+  }
+  if (obj.type === "shutdown") {
+    return ok({ type: "shutdown" });
+  }
+  if (obj.type === "health-request") {
+    return ok({ type: "health-request" });
+  }
   if (obj.type === "start-thread") {
     const errMsg = validateStartThreadMsg(obj);
     if (errMsg !== null) return err(new Error(errMsg));
+    // Field types are validated above; `Record<string, unknown>` values stay `unknown` to TypeScript.
+    return ok({
+      type: "start-thread",
+      runId: obj.runId,
+      workflow: obj.workflow,
+      prompt: obj.prompt,
+      maxRounds: obj.maxRounds,
+      dryRun: obj.dryRun,
+    } as StartThreadMessage);
   }
   if (obj.type === "resume-thread") {
     const errMsg = validateResumeThreadMsg(obj);
     if (errMsg !== null) return err(new Error(errMsg));
+    // Elements are validated as plain objects by the kernel; trust the wire shape here.
+    return ok({
+      type: "resume-thread",
+      runId: obj.runId,
+      messages: obj.messages as ResumeThreadMessage["messages"],
+      maxRounds: obj.maxRounds,
+      dryRun: obj.dryRun,
+    } as ResumeThreadMessage);
   }
-  return ok(raw as ParentToWorkerMessage);
+  return err(new Error(`Unhandled IPC message type: "${obj.type}"`));
 }
 
-function parseSignalMsg(obj: Record<string, unknown>, raw: unknown): Result<WorkerToParentMessage> {
+function parseSignalMsg(obj: Record<string, unknown>): Result<WorkerToParentMessage> {
   if (typeof obj.sense !== "string") {
     return err(new Error("Worker 'signal' message missing string 'sense' field"));
   }
   if (!("payload" in obj)) {
     return err(new Error("Worker 'signal' message missing 'payload' field"));
   }
-  return ok(raw as SignalMessage);
+  return ok({
+    type: "signal",
+    sense: obj.sense,
+    payload: obj.payload,
+  });
 }
 
-function parseErrorMsg(obj: Record<string, unknown>, raw: unknown): Result<WorkerToParentMessage> {
+function parseErrorMsg(obj: Record<string, unknown>): Result<WorkerToParentMessage> {
   if (typeof obj.sense !== "string") {
     return err(new Error("Worker 'error' message missing string 'sense' field"));
   }
   if (typeof obj.error !== "string") {
     return err(new Error("Worker 'error' message missing string 'error' field"));
   }
-  return ok(raw as ErrorMessage);
+  return ok({
+    type: "error",
+    sense: obj.sense,
+    error: obj.error,
+  });
 }
 
-function parseHealthResponseMsg(
+function parseHealthResponseMsg(obj: Record<string, unknown>): Result<WorkerToParentMessage> {
-  obj: Record<string, unknown>,
-  raw: unknown,
-): Result<WorkerToParentMessage> {
   if (!Array.isArray(obj.senses)) {
     return err(new Error("Worker 'health-response' message missing 'senses' array"));
   }
   if (typeof obj.inFlightCount !== "number") {
     return err(new Error("Worker 'health-response' message missing 'inFlightCount' number"));
   }
-  return ok(raw as HealthResponseMessage);
+  return ok({
+    type: "health-response",
+    // Kernel only sends string[] today; keep accepting any array elements without filtering.
+    senses: obj.senses as string[],
+    inFlightCount: obj.inFlightCount,
+  });
 }
 
-const THREAD_EVENT_TYPES = new Set<string>([
+function isThreadEventType(value: string): value is ThreadEventType {
-  "queued",
+  switch (value) {
-  "started",
+    case "queued":
-  "step_complete",
+    case "started":
-  "completed",
+    case "step_complete":
-  "failed",
+    case "completed":
-]);
+    case "failed":
+      return true;
+    default:
+      return false;
+  }
+}
 
-function parseThreadEventMsg(
+function parseThreadEventMsg(obj: Record<string, unknown>): Result<WorkerToParentMessage> {
-  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)) {
+  if (typeof obj.eventType !== "string" || !isThreadEventType(obj.eventType)) {
     return err(
       new Error(`Worker 'thread-event' message has invalid 'eventType': "${obj.eventType}"`),
     );
@@ -222,20 +272,26 @@ function parseThreadEventMsg(
   if (!("payload" in obj)) {
     return err(new Error("Worker 'thread-event' message missing 'payload' field"));
   }
-  return ok(raw as ThreadEventMessage);
+  return ok({
+    type: "thread-event",
+    runId: obj.runId,
+    eventType: obj.eventType,
+    payload: obj.payload,
+  });
 }
 
-function parseWorkflowErrorMsg(
+function parseWorkflowErrorMsg(obj: Record<string, unknown>): Result<WorkerToParentMessage> {
-  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);
+  return ok({
+    type: "workflow-error",
+    runId: obj.runId,
+    error: obj.error,
+  });
 }
 
 const WORKER_MSG_TYPES = new Set([
@@ -245,43 +301,61 @@ const WORKER_MSG_TYPES = new Set([
   "health-response",
   "thread-event",
   "workflow-error",
-  "thread-command-event",
+  "thread-workflow-message",
 ]);
 
-function parseThreadCommandEventMsg(
+function parseThreadWorkflowMessageMsg(
   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"));
+    return err(new Error("Worker 'thread-workflow-message' missing string 'runId' field"));
   }
-  if (obj.event === null || typeof obj.event !== "object") {
+  if (!isPlainRecord(obj.message)) {
-    return err(new Error("Worker 'thread-command-event' message missing object 'event' field"));
+    return err(new Error("Worker 'thread-workflow-message' missing object 'message' field"));
   }
-  const event = obj.event as Record<string, unknown>;
+  const msg = obj.message;
-  if (typeof event.type !== "string") {
+  if (typeof msg.role !== "string") {
-    return err(new Error("Worker 'thread-command-event' event missing string 'type' field"));
+    return err(new Error("Worker 'thread-workflow-message' message missing string 'role' field"));
   }
-  return ok(raw as ThreadCommandEventMessage);
+  if (typeof msg.content !== "string") {
+    return err(
+      new Error("Worker 'thread-workflow-message' message missing string 'content' field"),
+    );
+  }
+  if (typeof msg.timestamp !== "number") {
+    return err(
+      new Error("Worker 'thread-workflow-message' message missing number 'timestamp' field"),
+    );
+  }
+  return ok({
+    type: "thread-workflow-message",
+    runId: obj.runId,
+    message: {
+      role: msg.role,
+      content: msg.content,
+      meta: "meta" in msg ? msg.meta : undefined,
+      timestamp: msg.timestamp,
+    },
+  });
 }
 
 /** Validate and parse an unknown IPC message received from a worker process. */
 export function parseWorkerMessage(raw: unknown): Result<WorkerToParentMessage> {
-  if (raw === null || typeof raw !== "object") {
+  if (!isPlainRecord(raw)) {
     return err(new Error("Worker IPC message is not an object"));
   }
-  const obj = raw as Record<string, unknown>;
+  const obj = raw;
   if (typeof obj.type !== "string") {
     return err(new Error("Worker IPC message missing string 'type' field"));
   }
   if (!WORKER_MSG_TYPES.has(obj.type)) {
     return err(new Error(`Unknown worker IPC message type: "${obj.type}"`));
   }
-  if (obj.type === "signal") return parseSignalMsg(obj, raw);
+  if (obj.type === "signal") return parseSignalMsg(obj);
-  if (obj.type === "error") return parseErrorMsg(obj, raw);
+  if (obj.type === "error") return parseErrorMsg(obj);
-  if (obj.type === "health-response") return parseHealthResponseMsg(obj, raw);
+  if (obj.type === "health-response") return parseHealthResponseMsg(obj);
-  if (obj.type === "thread-event") return parseThreadEventMsg(obj, raw);
+  if (obj.type === "thread-event") return parseThreadEventMsg(obj);
-  if (obj.type === "workflow-error") return parseWorkflowErrorMsg(obj, raw);
+  if (obj.type === "workflow-error") return parseWorkflowErrorMsg(obj);
-  if (obj.type === "thread-command-event") return parseThreadCommandEventMsg(obj, raw);
+  if (obj.type === "thread-workflow-message") return parseThreadWorkflowMessageMsg(obj);
   return ok({ type: "ready" });
 }

packages/daemon/src/kernel-file-watch.ts

@@ -0,0 +1,92 @@
+/**
+ * File-watcher callbacks for nerve.yaml / sense / workflow sources (hot reload wiring).
+ */
+
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+
+import type { NerveConfig } from "@uncaged/nerve-core";
+import { parseNerveConfig } from "@uncaged/nerve-core";
+
+import type { LogStore } from "@uncaged/nerve-store";
+import type { WorkflowManager } from "./workflow-manager.js";
+
+export type KernelFileWatchDeps = {
+  nerveRoot: string;
+  getConfig: () => NerveConfig;
+  logStore: LogStore;
+  workflowManager: WorkflowManager;
+  restartGroup: (group: string) => Promise<void>;
+  reloadConfig: (newConfig: NerveConfig) => void;
+};
+
+export type KernelFileWatchHandlers = {
+  onSenseFileChange: (senseName: string) => void;
+  onWorkflowFileChange: (workflowName: string) => void;
+  onConfigFileChange: () => void;
+};
+
+export function createKernelFileWatchHandlers(deps: KernelFileWatchDeps): KernelFileWatchHandlers {
+  function onSenseFileChange(senseName: string): void {
+    const sc = deps.getConfig().senses[senseName];
+    if (sc === undefined) return;
+    process.stderr.write(
+      `[kernel] sense file changed: "${senseName}", restarting group "${sc.group}"\n`,
+    );
+    deps.logStore.append({
+      source: "system",
+      type: "sense_reload",
+      refId: senseName,
+      payload: null,
+      ts: Date.now(),
+    });
+    deps.restartGroup(sc.group).catch((e) => {
+      const msg = e instanceof Error ? e.message : String(e);
+      process.stderr.write(`[kernel] restartGroup error: ${msg}\n`);
+    });
+  }
+
+  function onWorkflowFileChange(workflowName: string): void {
+    process.stderr.write(
+      `[kernel] workflow file changed: "${workflowName}", draining and respawning worker\n`,
+    );
+    deps.logStore.append({
+      source: "system",
+      type: "workflow_reload",
+      refId: workflowName,
+      payload: null,
+      ts: Date.now(),
+    });
+    deps.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 onConfigFileChange(): void {
+    process.stderr.write("[kernel] nerve.yaml changed, reloading config\n");
+    deps.logStore.append({
+      source: "system",
+      type: "config_reload",
+      refId: null,
+      payload: null,
+      ts: Date.now(),
+    });
+    try {
+      const raw = readFileSync(join(deps.nerveRoot, "nerve.yaml"), "utf8");
+      const parseResult = parseNerveConfig(raw);
+      if (!parseResult.ok) {
+        process.stderr.write(
+          `[kernel] config parse error, keeping current config: ${parseResult.error.message}\n`,
+        );
+        return;
+      }
+      deps.reloadConfig(parseResult.value);
+    } catch (e) {
+      const msg = e instanceof Error ? e.message : String(e);
+      process.stderr.write(`[kernel] failed to read nerve.yaml, keeping current config: ${msg}\n`);
+    }
+  }
+
+  return { onSenseFileChange, onWorkflowFileChange, onConfigFileChange };
+}

packages/daemon/src/kernel-sense-groups.ts

@@ -0,0 +1,29 @@
+import type { NerveConfig } from "@uncaged/nerve-core";
+
+export function groupForSense(config: NerveConfig, senseName: string): string | null {
+  const senseConfig = config.senses[senseName];
+  if (senseConfig === undefined) return null;
+  return senseConfig.group;
+}
+
+export function senseNamesInGroup(config: NerveConfig, group: string): string[] {
+  return Object.entries(config.senses)
+    .filter(([, sc]) => sc.group === group)
+    .map(([name]) => name);
+}
+
+export function collectSenseGroups(cfg: NerveConfig): Set<string> {
+  const result = new Set<string>();
+  for (const sc of Object.values(cfg.senses)) {
+    result.add(sc.group);
+  }
+  return result;
+}
+
+export function senseNamesInGroupAsSet(cfg: NerveConfig, group: string): Set<string> {
+  const result = new Set<string>();
+  for (const [name, sc] of Object.entries(cfg.senses)) {
+    if (sc.group === group) result.add(name);
+  }
+  return result;
+}

packages/daemon/src/kernel.ts

@@ -1,43 +1,32 @@
 /**
- * Kernel — the main orchestrator that ties sense workers, signal bus, and
+ * Kernel — ties sense workers, signal bus, reflex scheduler, workflow manager,
- * reflex scheduler together.
+ * optional file watcher, and daemon IPC.
- *
- * Responsibilities:
- *   - Spawn one child process per sense group (via fork)
- *   - Route SignalMessage from workers → SignalBus
- *   - Route ErrorMessage from workers → stderr log
- *   - Drive compute triggers via ReflexScheduler
- *   - Graceful shutdown: stop scheduler, send shutdown to all workers
- *   - Hot reload: restartGroup, reloadConfig, file watcher integration
- *   - Health reporting: getHealth
  */
 
-import { fork } from "node:child_process";
+import { join } from "node:path";
-import type { ChildProcess } from "node:child_process";
-import { readFileSync } from "node:fs";
-import { dirname, join } from "node:path";
-import { fileURLToPath } from "node:url";
 
 import type { NerveConfig, SenseInfo, Signal } from "@uncaged/nerve-core";
-import { parseNerveConfig } from "@uncaged/nerve-core";
+import { routeSenseComputeOutput } from "@uncaged/nerve-core";
 
+import { createLogStore } from "@uncaged/nerve-store";
+import type { LogStore } from "@uncaged/nerve-store";
 import { createDaemonIpcServer } from "./daemon-ipc.js";
 import type { DaemonIpcServer } from "./daemon-ipc.js";
 import { createFileWatcher } from "./file-watcher.js";
 import type { FileWatcher } from "./file-watcher.js";
-import type { ComputeMessage, ShutdownMessage } from "./ipc.js";
 import { parseWorkerMessage } from "./ipc.js";
-import { createLogStore } from "./log-store.js";
+import { createKernelFileWatchHandlers } from "./kernel-file-watch.js";
-import type { LogStore } from "./log-store.js";
+import {
+  collectSenseGroups,
+  groupForSense,
+  senseNamesInGroup,
+  senseNamesInGroupAsSet,
+} from "./kernel-sense-groups.js";
 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 {
+import { createSenseWorkerPool, resolveWorkerScript } from "./worker-pool.js";
-  formatCapturedStderrTail,
-  formatChildExitSummary,
-  teeCapturedStderr,
-} from "./worker-fork-support.js";
 import { createWorkflowManager } from "./workflow-manager.js";
 import type { WorkflowManager } from "./workflow-manager.js";
 
@@ -57,93 +46,19 @@ export type Kernel = {
   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. */
   getWorkerPid: (group: string) => number | null;
-  /** Sends a compute message to the worker responsible for the given sense. */
   triggerCompute: (senseName: string) => void;
-  /**
-   * On-demand sense trigger — looks up the group for `senseName`, finds its worker,
-   * and sends a compute message. Throws if the sense is unknown.
-   */
   triggerSense: (senseName: string) => void;
-  /** Gracefully restart a group worker (wait for exit, then respawn). */
   restartGroup: (group: string) => Promise<void>;
-  /** Reload config from a new NerveConfig, incrementally updating scheduler and workers.
-   * Note: any pending/throttled computes in the old scheduler are silently dropped on reload.
-   * In-flight state is not preserved across reloadConfig. */
   reloadConfig: (newConfig: NerveConfig) => void;
-  /** Return daemon health info. */
   getHealth: () => KernelHealth;
 };
 
-type WorkerEntry = {
-  group: string;
-  process: ChildProcess;
-};
-
-function resolveWorkerScript(): string {
-  const __filename = fileURLToPath(import.meta.url);
-  const __dir = dirname(__filename);
-  return join(__dir, "sense-worker.js");
-}
-
-function spawnWorker(
-  nerveRoot: string,
-  group: string,
-  workerScript: string,
-  stderrTail: { value: string },
-): ChildProcess {
-  const child = fork(workerScript, ["--group", group, "--root", nerveRoot], {
-    stdio: ["ignore", "inherit", "pipe", "ipc"],
-  });
-  teeCapturedStderr(child, stderrTail);
-  // Prevent unhandled EPIPE when writing to a child whose IPC channel closed
-  child.on("error", (err) => {
-    if ((err as NodeJS.ErrnoException).code !== "EPIPE") {
-      console.error("[worker] error:", err.message);
-    }
-  });
-  return child;
-}
-
-function sendCompute(worker: ChildProcess, senseName: string): void {
-  // worker.connected is false when the IPC channel has been closed (e.g. worker crashed)
-  if (worker.connected === false) return;
-  const msg: ComputeMessage = { type: "compute", sense: senseName };
-  try {
-    worker.send(msg);
-  } catch {
-    // IPC channel closed between connected check and send
-  }
-}
-
-function sendShutdown(worker: ChildProcess): void {
-  if (worker.connected === false) return;
-  const msg: ShutdownMessage = { type: "shutdown" };
-  try {
-    worker.send(msg);
-  } catch {
-    // IPC channel closed between connected check and send
-  }
-}
-
-function groupForSense(config: NerveConfig, senseName: string): string | null {
-  const senseConfig = config.senses[senseName];
-  if (senseConfig === undefined) return null;
-  return senseConfig.group;
-}
-
 export type KernelOptions = {
   workerScript?: string | null;
   enableFileWatcher?: boolean;
-  /** Override the LogStore instance (useful for testing). */
   logStore?: LogStore;
-  /**
-   * Unix socket path for the daemon IPC server (used by CLI to send trigger-workflow).
-   * When null, the IPC server is not started (e.g. during tests).
-   */
   ipcSocketPath?: string | null;
 };
 
@@ -184,9 +99,9 @@ export function createKernel(
     groups.add(senseConfig.group);
   }
 
-  const workers = new Map<string, WorkerEntry>();
   let stopped = false;
-  let scheduler: ReflexScheduler = null as unknown as ReflexScheduler;
+  /** Assigned before workers start; `handleWorkerMessage` only runs after this is set. */
+  let scheduler!: ReflexScheduler;
 
   let readyResolve: (() => void) | undefined;
   const ready = new Promise<void>((resolve) => {
@@ -194,10 +109,10 @@ export function createKernel(
   });
   let pendingReadyCount = groups.size > 0 ? groups.size : 0;
 
-  function sensesForGroup(group: string): string[] {
+  function clearSchedulerForGroup(group: string): void {
-    return Object.entries(config.senses)
+    for (const senseName of senseNamesInGroup(config, group)) {
-      .filter(([, sc]) => sc.group === group)
+      scheduler.onComputeComplete(senseName);
-      .map(([name]) => name);
+    }
   }
 
   function handleWorkerMessage(raw: unknown): void {
@@ -230,66 +145,46 @@ export function createKernel(
     }
 
     if (msg.type === "signal") {
-      const signal: Signal = {
+      const route = routeSenseComputeOutput(msg.payload);
-        id: nextSignalId(),
+      if (route.kind === "launch") {
-        senseId: msg.sense,
+        const { workflowName, maxRounds, prompt } = route.launch;
-        payload: msg.payload,
+        workflowManager.startWorkflow(workflowName, { prompt, maxRounds, dryRun: false });
-        ts: Date.now(),
+        logStore.append({
-      };
+          source: "sense",
-      logStore.append({
+          type: "workflow-launch",
-        source: "sense",
+          refId: msg.sense,
-        type: "signal",
+          payload: JSON.stringify(route.launch),
-        refId: msg.sense,
+          ts: Date.now(),
-        payload: JSON.stringify(msg.payload),
+        });
-        ts: signal.ts,
+      } else {
-      });
+        const signal: Signal = {
-      bus.emit(signal);
+          id: nextSignalId(),
+          senseId: msg.sense,
+          payload: route.payload,
+          ts: Date.now(),
+        };
+        logStore.append({
+          source: "sense",
+          type: "signal",
+          refId: msg.sense,
+          payload: JSON.stringify(route.payload),
+          ts: signal.ts,
+        });
+        bus.emit(signal);
+      }
       scheduler.onComputeComplete(msg.sense);
     }
-
-    // health-response is handled externally by the caller; no action needed here
   }
 
-  function startWorker(group: string): Promise<void> {
+  const senseWorkerPool = createSenseWorkerPool({
-    const stderrTail = { value: "" };
+    nerveRoot,
-    const child = spawnWorker(nerveRoot, group, workerScript, stderrTail);
+    workerScript,
-
+    onWorkerMessage: handleWorkerMessage,
-    let workerReadyResolve: (() => void) | undefined;
+    sensesForGroup: (group) => senseNamesInGroup(config, group),
-    const workerReady = new Promise<void>((resolve) => {
+    onWorkerCrashed: clearSchedulerForGroup,
-      workerReadyResolve = resolve;
+    onBeforeGroupRestart: clearSchedulerForGroup,
-    });
+    isStopped: () => stopped,
-
+  });
-    child.on("message", (raw: unknown) => {
-      const result = parseWorkerMessage(raw);
-      if (result.ok && result.value.type === "ready") {
-        workerReadyResolve?.();
-      }
-      handleWorkerMessage(raw);
-    });
-
-    child.on("exit", (code, signal) => {
-      const summary = formatChildExitSummary(code, signal ?? null);
-      process.stderr.write(
-        `[kernel] worker for group "${group}" exited (${summary})${formatCapturedStderrTail(stderrTail.value)}\n`,
-      );
-      // Resolve ready in case the worker exits before sending ready (prevents hangs)
-      workerReadyResolve?.();
-      if (!stopped && code !== 0) {
-        process.stderr.write(`[kernel] respawning worker for group "${group}" in 1s\n`);
-        for (const senseName of sensesForGroup(group)) {
-          scheduler.onComputeComplete(senseName);
-        }
-        setTimeout(() => {
-          if (!stopped) {
-            startWorker(group);
-          }
-        }, 1000);
-      }
-    });
-
-    workers.set(group, { group, process: child });
-    return workerReady;
-  }
 
   function triggerFn(senseName: string): void {
     const group = groupForSense(config, senseName);
@@ -297,12 +192,7 @@ export function createKernel(
       process.stderr.write(`[kernel] triggerFn: unknown sense "${senseName}"\n`);
       return;
     }
-    const entry = workers.get(group);
+    senseWorkerPool.sendCompute(group, senseName);
-    if (entry === undefined) {
-      process.stderr.write(`[kernel] triggerFn: no worker for group "${group}"\n`);
-      return;
-    }
-    sendCompute(entry.process, senseName);
   }
 
   function triggerSense(senseName: string): void {
@@ -310,18 +200,14 @@ export function createKernel(
     if (group === null) {
       throw new Error(`Unknown sense: "${senseName}"`);
     }
-    const entry = workers.get(group);
+    if (!senseWorkerPool.hasWorkerForGroup(group)) {
-    if (entry === undefined) {
       throw new Error(`No worker running for group "${group}" (sense: "${senseName}")`);
     }
-    sendCompute(entry.process, senseName);
+    senseWorkerPool.sendCompute(group, senseName);
   }
 
   scheduler = createReflexScheduler(config, bus, triggerFn, {
     logStore,
-    workflowTriggerFn: (workflowName, payload) => {
-      workflowManager.startWorkflow(workflowName, payload);
-    },
   });
 
   if (groups.size === 0) {
@@ -329,63 +215,13 @@ export function createKernel(
   }
 
   for (const group of groups) {
-    startWorker(group);
+    senseWorkerPool.startWorker(group);
-  }
-
-  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();
-      });
-    });
-  }
-
-  // --- restartGroup: gracefully stop worker, then respawn and await ready ---
-  async function restartGroup(group: string): Promise<void> {
-    const entry = workers.get(group);
-    if (entry === undefined) return;
-
-    for (const senseName of sensesForGroup(group)) {
-      scheduler.onComputeComplete(senseName);
-    }
-
-    sendShutdown(entry.process);
-    await waitForExit(entry.process, 5000);
-
-    if (!stopped) {
-      await startWorker(group);
-    }
-  }
-
-  function collectGroups(cfg: NerveConfig): Set<string> {
-    const result = new Set<string>();
-    for (const sc of Object.values(cfg.senses)) {
-      result.add(sc.group);
-    }
-    return result;
-  }
-
-  function sensesForGroupInConfig(cfg: NerveConfig, group: string): Set<string> {
-    const result = new Set<string>();
-    for (const [name, sc] of Object.entries(cfg.senses)) {
-      if (sc.group === group) result.add(name);
-    }
-    return result;
   }
 
   function removeStaleGroups(oldGroups: Set<string>, newGroups: Set<string>): void {
     for (const g of oldGroups) {
       if (newGroups.has(g)) continue;
-      const entry = workers.get(g);
+      senseWorkerPool.evictGroup(g);
-      if (entry !== undefined) {
-        sendShutdown(entry.process);
-        workers.delete(g);
-      }
       groups.delete(g);
     }
   }
@@ -394,30 +230,25 @@ export function createKernel(
     for (const g of newGroups) {
       if (oldGroups.has(g)) continue;
       groups.add(g);
-      if (!stopped) startWorker(g);
+      if (!stopped) {
+        senseWorkerPool.startWorker(g);
+      }
     }
   }
 
   function reloadConfig(newConfig: NerveConfig): void {
-    const oldGroups = collectGroups(config);
+    const oldGroups = collectSenseGroups(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,
-      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(
@@ -432,20 +263,17 @@ export function createKernel(
       }
     }
 
-    const newGroups = collectGroups(newConfig);
+    const newGroups = collectSenseGroups(newConfig);
     removeStaleGroups(oldGroups, newGroups);
     addNewGroups(oldGroups, newGroups);
 
-    // Restart existing groups that gained new senses — the running worker process
-    // was spawned with the old config and will report "Unknown sense" for any newly
-    // added sense until it is restarted.
     for (const g of newGroups) {
-      if (!oldGroups.has(g)) continue; // already handled by addNewGroups
+      if (!oldGroups.has(g)) continue;
-      const oldSenses = sensesForGroupInConfig(oldConfig, g);
+      const oldSenses = senseNamesInGroupAsSet(oldConfig, g);
-      const newSenses = sensesForGroupInConfig(newConfig, g);
+      const newSenses = senseNamesInGroupAsSet(newConfig, g);
       const gained = [...newSenses].some((s) => !oldSenses.has(s));
       if (gained) {
-        restartGroup(g).catch((e) => {
+        senseWorkerPool.restartGroup(g).catch((e) => {
           const msg = e instanceof Error ? e.message : String(e);
           process.stderr.write(`[kernel] reloadConfig restartGroup error for "${g}": ${msg}\n`);
         });
@@ -457,80 +285,28 @@ export function createKernel(
     return {
       uptime: Date.now() - startTime,
       activeSenses: Object.keys(config.senses).length,
-      activeGroups: workers.size,
+      activeGroups: senseWorkerPool.activeGroupCount(),
       pendingComputes: 0,
       activeWorkflows: workflowManager.totalActiveCount(),
       memoryUsage: process.memoryUsage(),
     };
   }
 
-  function handleSenseFileChange(senseName: string): void {
+  const fileWatchHandlers = createKernelFileWatchHandlers({
-    const sc = config.senses[senseName];
+    nerveRoot,
-    if (sc === undefined) return;
+    getConfig: () => config,
-    process.stderr.write(
+    logStore,
-      `[kernel] sense file changed: "${senseName}", restarting group "${sc.group}"\n`,
+    workflowManager,
-    );
+    restartGroup: (group) => senseWorkerPool.restartGroup(group),
-    logStore.append({
+    reloadConfig,
-      source: "system",
+  });
-      type: "sense_reload",
-      refId: senseName,
-      payload: null,
-      ts: Date.now(),
-    });
-    restartGroup(sc.group).catch((e) => {
-      const msg = e instanceof Error ? e.message : String(e);
-      process.stderr.write(`[kernel] restartGroup error: ${msg}\n`);
-    });
-  }
-
-  function handleWorkflowFileChange(workflowName: string): void {
-    process.stderr.write(
-      `[kernel] workflow file changed: "${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({
-      source: "system",
-      type: "config_reload",
-      refId: null,
-      payload: null,
-      ts: Date.now(),
-    });
-    try {
-      const raw = readFileSync(join(nerveRoot, "nerve.yaml"), "utf8");
-      const parseResult = parseNerveConfig(raw);
-      if (!parseResult.ok) {
-        process.stderr.write(
-          `[kernel] config parse error, keeping current config: ${parseResult.error.message}\n`,
-        );
-        return;
-      }
-      reloadConfig(parseResult.value);
-    } catch (e) {
-      const msg = e instanceof Error ? e.message : String(e);
-      process.stderr.write(`[kernel] failed to read nerve.yaml, keeping current config: ${msg}\n`);
-    }
-  }
 
   let fileWatcher: FileWatcher | null = null;
   if (options.enableFileWatcher) {
     fileWatcher = createFileWatcher(nerveRoot, (change) => {
-      if (change.kind === "sense") handleSenseFileChange(change.senseName);
+      if (change.kind === "sense") fileWatchHandlers.onSenseFileChange(change.senseName);
-      if (change.kind === "config") handleConfigFileChange();
+      if (change.kind === "config") fileWatchHandlers.onConfigFileChange();
-      if (change.kind === "workflow") handleWorkflowFileChange(change.workflowName);
+      if (change.kind === "workflow") fileWatchHandlers.onWorkflowFileChange(change.workflowName);
     });
   }
 
@@ -570,12 +346,7 @@ export function createKernel(
     }
     scheduler.stop();
     await workflowManager.stop();
-    const exitPromises: Promise<void>[] = [];
+    await senseWorkerPool.shutdownAll();
-    for (const entry of workers.values()) {
-      sendShutdown(entry.process);
-      exitPromises.push(waitForExit(entry.process, 5000));
-    }
-    await Promise.all(exitPromises);
     logStore.append({
       source: "system",
       type: "stop",
@@ -587,7 +358,7 @@ export function createKernel(
   }
 
   function getWorkerPid(group: string): number | null {
-    return workers.get(group)?.process.pid ?? null;
+    return senseWorkerPool.getWorkerPid(group);
   }
 
   const senseCount = Object.keys(config.senses).length;
@@ -603,7 +374,7 @@ export function createKernel(
     getWorkerPid,
     triggerCompute: triggerFn,
     triggerSense,
-    restartGroup,
+    restartGroup: (group) => senseWorkerPool.restartGroup(group),
     reloadConfig,
     getHealth,
   };

packages/daemon/src/reflex-scheduler.ts

@@ -10,15 +10,12 @@
  */
 
 import type { NerveConfig } from "@uncaged/nerve-core";
-import type { LogStore } from "./log-store.js";
+import type { LogStore } from "@uncaged/nerve-store";
 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;
@@ -40,7 +37,6 @@ function makeSenseState(): SenseState {
 
 export type ReflexSchedulerOptions = {
   logStore?: LogStore;
-  workflowTriggerFn?: WorkflowTriggerFn;
 };
 
 /**
@@ -157,21 +153,6 @@ 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

@@ -1,17 +1,17 @@
 import { mkdirSync, readFileSync, readdirSync } from "node:fs";
 import { dirname, join } from "node:path";
+import { DatabaseSync } from "node:sqlite";
 
-import Database from "better-sqlite3";
+import { drizzle } from "drizzle-orm/node-sqlite";
-import { drizzle } from "drizzle-orm/better-sqlite3";
+import type { NodeSQLiteDatabase } from "drizzle-orm/node-sqlite";
-import type { BetterSQLite3Database } from "drizzle-orm/better-sqlite3";
 
 import type { Result } from "@uncaged/nerve-core";
-import { err, ok } from "@uncaged/nerve-core";
+import { err, isPlainRecord, ok } from "@uncaged/nerve-core";
 
-import type { BlobStore } from "./blob-store.js";
+import type { BlobStore } from "@uncaged/nerve-store";
 
 /** A Drizzle DB instance (schema-generic) */
-export type DrizzleDB = BetterSQLite3Database<Record<string, never>>;
+export type DrizzleDB = NodeSQLiteDatabase<Record<string, never>>;
 
 /** Read-only map of peer sense name → their Drizzle DB */
 export type PeerMap = Readonly<Record<string, DrizzleDB>>;
@@ -42,7 +42,7 @@ export type SenseRuntime = {
   compute: ComputeFn;
 };
 
-function ensureMigrationsTable(sqlite: Database.Database): Result<void> {
+function ensureMigrationsTable(sqlite: DatabaseSync): Result<void> {
   try {
     sqlite.exec(
       `CREATE TABLE IF NOT EXISTS _migrations (
@@ -69,11 +69,7 @@ function listMigrationFiles(migrationsDir: string): Result<string[]> {
   }
 }
 
-function applyMigrationFile(
+function applyMigrationFile(sqlite: DatabaseSync, file: string, filePath: string): Result<void> {
-  sqlite: Database.Database,
-  file: string,
-  filePath: string,
-): Result<void> {
   let sql: string;
   try {
     sql = readFileSync(filePath, "utf8");
@@ -83,13 +79,18 @@ function applyMigrationFile(
   }
 
   const insertJournal = sqlite.prepare("INSERT INTO _migrations (name, applied_at) VALUES (?, ?)");
+  sqlite.exec("BEGIN IMMEDIATE");
   try {
-    sqlite.transaction(() => {
+    sqlite.exec(sql);
-      sqlite.exec(sql);
+    insertJournal.run(file, Date.now());
-      insertJournal.run(file, Date.now());
+    sqlite.exec("COMMIT");
-    })();
     return ok(undefined);
   } catch (e) {
+    try {
+      sqlite.exec("ROLLBACK");
+    } catch {
+      // ignore secondary errors during rollback
+    }
     const msg = e instanceof Error ? e.message : String(e);
     return err(new Error(`Migration "${file}" failed: ${msg}`));
   }
@@ -97,20 +98,21 @@ function applyMigrationFile(
 
 /**
  * Run all *.sql migration files in the given directory against a
- * better-sqlite3 Database, in lexicographic order.
+ * `node:sqlite` DatabaseSync, in lexicographic order.
  * Tracks applied migrations in _migrations table to avoid re-running.
  */
-export function runMigrations(sqlite: Database.Database, migrationsDir: string): Result<void> {
+export function runMigrations(sqlite: DatabaseSync, migrationsDir: string): Result<void> {
   const tableResult = ensureMigrationsTable(sqlite);
   if (!tableResult.ok) return tableResult;
 
   const filesResult = listMigrationFiles(migrationsDir);
   if (!filesResult.ok) return filesResult;
 
+  const migrationRows = sqlite.prepare("SELECT name FROM _migrations").all();
   const applied = new Set<string>(
-    (sqlite.prepare("SELECT name FROM _migrations").all() as Array<{ name: string }>).map(
+    migrationRows
-      (r) => r.name,
+      .filter((r): r is { name: string } => isPlainRecord(r) && typeof r.name === "string")
-    ),
+      .map((r) => r.name),
   );
 
   for (const file of filesResult.value) {
@@ -129,14 +131,13 @@ export function runMigrations(sqlite: Database.Database, migrationsDir: string):
 export function openSenseDb(
   dbPath: string,
   migrationsDir: string,
-): Result<{ sqlite: Database.Database; db: DrizzleDB }> {
+): Result<{ sqlite: DatabaseSync; db: DrizzleDB }> {
-  let sqlite: Database.Database;
+  let sqlite: DatabaseSync;
 
   try {
     mkdirSync(dirname(dbPath), { recursive: true });
-    sqlite = new Database(dbPath);
+    sqlite = new DatabaseSync(dbPath);
-    // WAL mode for better concurrent read performance
+    sqlite.exec("PRAGMA journal_mode=WAL");
-    sqlite.pragma("journal_mode = WAL");
   } catch (e) {
     const msg = e instanceof Error ? e.message : String(e);
     return err(new Error(`Failed to open database "${dbPath}": ${msg}`));
@@ -145,7 +146,8 @@ export function openSenseDb(
   const migResult = runMigrations(sqlite, migrationsDir);
   if (!migResult.ok) return migResult;
 
-  const db = drizzle(sqlite) as DrizzleDB;
+  // Drizzle infers a schema-specific DB type; senses are schema-agnostic at this layer.
+  const db = drizzle({ client: sqlite }) as DrizzleDB;
   return ok({ sqlite, db });
 }
 
@@ -153,16 +155,17 @@ export function openSenseDb(
  * Open a peer sense DB in read-only mode (no migrations).
  */
 export function openPeerDb(dbPath: string): Result<DrizzleDB> {
-  let sqlite: Database.Database;
+  let sqlite: DatabaseSync;
 
   try {
-    sqlite = new Database(dbPath, { readonly: true });
+    sqlite = new DatabaseSync(dbPath, { readOnly: true });
   } catch (e) {
     const msg = e instanceof Error ? e.message : String(e);
     return err(new Error(`Failed to open peer database "${dbPath}" (readonly): ${msg}`));
   }
 
-  return ok(drizzle(sqlite) as DrizzleDB);
+  // Same schema-agnostic Drizzle wrapper as openSenseDb.
+  return ok(drizzle({ client: sqlite }) as DrizzleDB);
 }
 
 /**
@@ -180,18 +183,13 @@ export async function loadComputeFn(senseIndexPath: string): Promise<Result<Comp
     return err(new Error(`Failed to import sense module "${senseIndexPath}": ${msg}`));
   }
 
-  if (
+  if (!isPlainRecord(mod) || !("compute" in mod) || typeof mod.compute !== "function") {
-    mod === null ||
-    typeof mod !== "object" ||
-    !("compute" in mod) ||
-    typeof (mod as Record<string, unknown>).compute !== "function"
-  ) {
     return err(
       new Error(`Sense module "${senseIndexPath}" must export a named "compute" function`),
     );
   }
 
-  return ok((mod as { compute: ComputeFn }).compute);
+  return ok(mod.compute as ComputeFn);
 }
 
 /**
@@ -232,7 +230,9 @@ export async function executeCompute(
   } catch (e) {
     const msg = e instanceof Error ? e.message : String(e);
     if (controller.signal.aborted) {
-      return err(new Error(`compute("${runtime.name}") timed out after ${timeoutMs as number}ms`));
+      return err(
+        new Error(`compute("${runtime.name}") timed out after ${String(timeoutMs ?? "?")}ms`),
+      );
     }
     return err(new Error(`compute("${runtime.name}") threw: ${msg}`));
   } finally {

packages/daemon/src/sense-worker.ts

@@ -20,7 +20,7 @@ import { join, resolve } from "node:path";
 import { parseNerveConfig } from "@uncaged/nerve-core";
 import type { NerveConfig } from "@uncaged/nerve-core";
 
-import { createBlobStore } from "./blob-store.js";
+import { createBlobStore } from "@uncaged/nerve-store";
 import type { WorkerToParentMessage } from "./ipc.js";
 import { parseParentMessage } from "./ipc.js";
 import { executeCompute, loadComputeFn, openPeerDb, openSenseDb } from "./sense-runtime.js";

packages/daemon/src/worker-fork-support.ts

@@ -26,10 +26,7 @@ export function teeCapturedStderr(child: ChildProcess, tail: { value: string }):
   });
 }
 
-export function formatChildExitSummary(
+export function formatChildExitSummary(code: number | null, signal: NodeJS.Signals | null): string {
-  code: number | null,
-  signal: NodeJS.Signals | null,
-): string {
   const codeStr = code === null || code === undefined ? "null" : String(code);
   if (signal) {
     return `code=${codeStr} signal=${signal}`;

packages/daemon/src/worker-pool.ts

@@ -0,0 +1,211 @@
+/**
+ * Sense worker pool — forked child processes per sense group (IPC lifecycle).
+ */
+
+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 { ComputeMessage, ShutdownMessage } from "./ipc.js";
+import { parseWorkerMessage } from "./ipc.js";
+import {
+  formatCapturedStderrTail,
+  formatChildExitSummary,
+  teeCapturedStderr,
+} from "./worker-fork-support.js";
+
+export function resolveWorkerScript(): string {
+  const __filename = fileURLToPath(import.meta.url);
+  const __dir = dirname(__filename);
+  return join(__dir, "sense-worker.js");
+}
+
+type WorkerEntry = {
+  group: string;
+  process: ChildProcess;
+};
+
+export type SenseWorkerPoolOptions = {
+  nerveRoot: string;
+  workerScript: string;
+  /** Invoked for every IPC message from a worker (including ready / signal / error). */
+  onWorkerMessage: (raw: unknown) => void;
+  /** Sense names in a group — used when clearing scheduler state on crash or restart. */
+  sensesForGroup: (group: string) => string[];
+  /**
+   * Called when a worker exits with non-zero code before scheduling a respawn
+   * (scheduler should release pending computes for senses in that group).
+   */
+  onWorkerCrashed: (group: string) => void;
+  /**
+   * Called at the beginning of `restartGroup` before shutdown
+   * (same scheduler cleanup as crash path).
+   */
+  onBeforeGroupRestart: (group: string) => void;
+  isStopped: () => boolean;
+};
+
+export type SenseWorkerPool = {
+  startWorker: (group: string) => Promise<void>;
+  restartGroup: (group: string) => Promise<void>;
+  /** Send shutdown and drop the entry without waiting (matches reloadConfig stale-group removal). */
+  evictGroup: (group: string) => void;
+  shutdownAll: () => Promise<void>;
+  sendCompute: (group: string, senseName: string) => void;
+  getWorkerPid: (group: string) => number | null;
+  hasWorkerForGroup: (group: string) => boolean;
+  activeGroupCount: () => number;
+};
+
+function spawnWorker(
+  nerveRoot: string,
+  group: string,
+  workerScript: string,
+  stderrTail: { value: string },
+): ChildProcess {
+  const child = fork(workerScript, ["--group", group, "--root", nerveRoot], {
+    stdio: ["ignore", "inherit", "pipe", "ipc"],
+  });
+  teeCapturedStderr(child, stderrTail);
+  child.on("error", (err) => {
+    if ((err as NodeJS.ErrnoException).code !== "EPIPE") {
+      console.error("[worker] error:", err.message);
+    }
+  });
+  return child;
+}
+
+function sendComputeToProcess(worker: ChildProcess, senseName: string): void {
+  if (worker.connected === false) return;
+  const msg: ComputeMessage = { type: "compute", sense: senseName };
+  try {
+    worker.send(msg);
+  } catch {
+    // IPC channel closed between connected check and send
+  }
+}
+
+function sendShutdownToProcess(worker: ChildProcess): void {
+  if (worker.connected === false) return;
+  const msg: ShutdownMessage = { type: "shutdown" };
+  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 createSenseWorkerPool(options: SenseWorkerPoolOptions): SenseWorkerPool {
+  const workers = new Map<string, WorkerEntry>();
+
+  function startWorker(group: string): Promise<void> {
+    const stderrTail = { value: "" };
+    const child = spawnWorker(options.nerveRoot, group, options.workerScript, stderrTail);
+
+    let workerReadyResolve: (() => void) | undefined;
+    const workerReady = new Promise<void>((resolve) => {
+      workerReadyResolve = resolve;
+    });
+
+    child.on("message", (raw: unknown) => {
+      const result = parseWorkerMessage(raw);
+      if (result.ok && result.value.type === "ready") {
+        workerReadyResolve?.();
+      }
+      options.onWorkerMessage(raw);
+    });
+
+    child.on("exit", (code, signal) => {
+      const summary = formatChildExitSummary(code, signal ?? null);
+      process.stderr.write(
+        `[kernel] worker for group "${group}" exited (${summary})${formatCapturedStderrTail(stderrTail.value)}\n`,
+      );
+      workerReadyResolve?.();
+      if (!options.isStopped() && code !== 0) {
+        process.stderr.write(`[kernel] respawning worker for group "${group}" in 1s\n`);
+        options.onWorkerCrashed(group);
+        setTimeout(() => {
+          if (!options.isStopped()) {
+            startWorker(group);
+          }
+        }, 1000);
+      }
+    });
+
+    workers.set(group, { group, process: child });
+    return workerReady;
+  }
+
+  async function restartGroup(group: string): Promise<void> {
+    const entry = workers.get(group);
+    if (entry === undefined) return;
+
+    options.onBeforeGroupRestart(group);
+
+    sendShutdownToProcess(entry.process);
+    await waitForExit(entry.process, 5000);
+
+    if (!options.isStopped()) {
+      await startWorker(group);
+    }
+  }
+
+  function evictGroup(group: string): void {
+    const entry = workers.get(group);
+    if (entry === undefined) return;
+    sendShutdownToProcess(entry.process);
+    workers.delete(group);
+  }
+
+  async function shutdownAll(): Promise<void> {
+    const exitPromises: Promise<void>[] = [];
+    for (const entry of workers.values()) {
+      sendShutdownToProcess(entry.process);
+      exitPromises.push(waitForExit(entry.process, 5000));
+    }
+    await Promise.all(exitPromises);
+  }
+
+  function sendCompute(group: string, senseName: string): void {
+    const entry = workers.get(group);
+    if (entry === undefined) return;
+    sendComputeToProcess(entry.process, senseName);
+  }
+
+  function getWorkerPid(group: string): number | null {
+    return workers.get(group)?.process.pid ?? null;
+  }
+
+  function hasWorkerForGroup(group: string): boolean {
+    return workers.has(group);
+  }
+
+  function activeGroupCount(): number {
+    return workers.size;
+  }
+
+  return {
+    startWorker,
+    restartGroup,
+    evictGroup,
+    shutdownAll,
+    sendCompute,
+    getWorkerPid,
+    hasWorkerForGroup,
+    activeGroupCount,
+  };
+}

packages/daemon/src/workflow-manager.ts

@@ -11,8 +11,10 @@ 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 { NerveConfig, WorkflowConfig, WorkflowMessage } from "@uncaged/nerve-core";
+import { START, isPlainRecord } from "@uncaged/nerve-core";
 
+import type { LogStore, WorkflowRunStatus } from "@uncaged/nerve-store";
 import type {
   ResumeThreadMessage,
   ShutdownMessage,
@@ -20,17 +22,21 @@ import type {
   ThreadEventMessage,
 } from "./ipc.js";
 import { parseWorkerMessage } from "./ipc.js";
-import type { LogStore } from "./log-store.js";
-import type { WorkflowRunStatus } from "./log-store.js";
 import {
   formatCapturedStderrTail,
   formatChildExitSummary,
   teeCapturedStderr,
 } from "./worker-fork-support.js";
 
+export type WorkflowLaunchParams = {
+  prompt: string;
+  maxRounds: number;
+  dryRun: boolean;
+};
+
 export type WorkflowManager = {
-  /** Trigger a new workflow thread (called by Reflex scheduler). */
+  /** Trigger a new workflow thread (Sense-driven launch or CLI / IPC). */
-  startWorkflow: (workflowName: string, payload: unknown) => void;
+  startWorkflow: (workflowName: string, launch: WorkflowLaunchParams) => 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. */
@@ -51,7 +57,9 @@ export type WorkflowManager = {
 
 type PendingThread = {
   runId: string;
-  payload: unknown;
+  prompt: string;
+  maxRounds: number;
+  dryRun: boolean;
 };
 
 type WorkflowState = {
@@ -81,6 +89,44 @@ const WORKER_SHUTDOWN_TIMEOUT_MS = 10_000;
 
 const DEFAULT_MAX_QUEUE = 100;
 
+function readLaunchFromTriggerPayload(
+  raw: unknown,
+  engineDefaultMaxRounds: number,
+): { prompt: string; maxRounds: number; dryRun: boolean } {
+  if (isPlainRecord(raw)) {
+    const o = raw;
+    if (typeof o.prompt === "string" && typeof o.maxRounds === "number") {
+      const dryRun = typeof o.dryRun === "boolean" ? o.dryRun : false;
+      return { prompt: o.prompt, maxRounds: o.maxRounds, dryRun };
+    }
+  }
+  return { prompt: "", maxRounds: engineDefaultMaxRounds, dryRun: false };
+}
+
+function ensureThreadMessagesWithStart(
+  messages: Array<{ role: string; content: string; meta: unknown; timestamp: number }>,
+  fallbackPrompt: string,
+  fallbackMaxRounds: number,
+  fallbackDryRun: boolean,
+): WorkflowMessage[] {
+  const mapped: WorkflowMessage[] = messages.map((m) => ({
+    role: m.role,
+    content: m.content,
+    meta: m.meta,
+    timestamp: m.timestamp,
+  }));
+  if (mapped.length > 0 && mapped[0].role === START) {
+    return mapped;
+  }
+  const start: WorkflowMessage = {
+    role: START,
+    content: fallbackPrompt,
+    meta: { maxRounds: fallbackMaxRounds, dryRun: fallbackDryRun },
+    timestamp: Date.now(),
+  };
+  return [start, ...mapped];
+}
+
 function resolveWorkerScript(): string {
   const __filename = fileURLToPath(import.meta.url);
   const __dir = dirname(__filename);
@@ -213,7 +259,13 @@ export function createWorkflowManager(
     }
   }
 
-  function dispatchThread(workflowName: string, runId: string, payload: unknown): void {
+  function dispatchThread(
+    workflowName: string,
+    runId: string,
+    prompt: string,
+    maxRounds: number,
+    dryRun: boolean,
+  ): void {
     const state = getOrCreateState(workflowName);
     state.active.add(runId);
 
@@ -222,11 +274,12 @@ export function createWorkflowManager(
       type: "start-thread",
       runId,
       workflow: workflowName,
-      triggerPayload: payload,
+      prompt,
+      maxRounds,
+      dryRun,
     };
     sendStartThread(worker.process, msg);
-    // Store triggerPayload in the log so it can be recovered after a crash
+    logWorkflowEvent(workflowName, runId, "started", { prompt, maxRounds, dryRun });
-    logWorkflowEvent(workflowName, runId, "started", { triggerPayload: payload });
   }
 
   function dequeueNext(workflowName: string): void {
@@ -239,7 +292,7 @@ export function createWorkflowManager(
     if (state.active.size < concurrency) {
       const next = state.queue.shift();
       if (next !== undefined) {
-        dispatchThread(workflowName, next.runId, next.payload);
+        dispatchThread(workflowName, next.runId, next.prompt, next.maxRounds, next.dryRun);
       }
     }
   }
@@ -260,8 +313,16 @@ export function createWorkflowManager(
 
   function recoverQueuedRun(workflowName: string, runId: string, state: WorkflowState): void {
     if (state.queue.some((q) => q.runId === runId)) return;
-    const triggerPayload = logStore.getTriggerPayload(runId);
+    const launch = readLaunchFromTriggerPayload(
-    state.queue.push({ runId, payload: triggerPayload });
+      logStore.getTriggerPayload(runId),
+      config.maxRounds,
+    );
+    state.queue.push({
+      runId,
+      prompt: launch.prompt,
+      maxRounds: launch.maxRounds,
+      dryRun: launch.dryRun,
+    });
     process.stderr.write(
       `[workflow-manager] crash-recovery: re-queued thread "${runId}" for "${workflowName}"\n`,
     );
@@ -274,18 +335,28 @@ export function createWorkflowManager(
     worker: WorkerEntry,
   ): void {
     if (state.active.has(runId)) return;
-    const events = logStore.getThreadEvents(runId);
+    const rawMessages = logStore.getThreadMessages(runId);
-    const triggerPayload = logStore.getTriggerPayload(runId);
+    const launch = readLaunchFromTriggerPayload(
+      logStore.getTriggerPayload(runId),
+      config.maxRounds,
+    );
+    const messages = ensureThreadMessagesWithStart(
+      rawMessages,
+      launch.prompt,
+      launch.maxRounds,
+      launch.dryRun,
+    );
     state.active.add(runId);
     const msg: ResumeThreadMessage = {
       type: "resume-thread",
       runId,
-      events,
+      messages,
-      triggerPayload,
+      maxRounds: launch.maxRounds,
+      dryRun: launch.dryRun,
     };
     sendResumeThread(worker.process, msg);
     process.stderr.write(
-      `[workflow-manager] crash-recovery: resuming thread "${runId}" for "${workflowName}" (${events.length} events)\n`,
+      `[workflow-manager] crash-recovery: resuming thread "${runId}" for "${workflowName}" (${messages.length} messages)\n`,
     );
   }
 
@@ -363,12 +434,12 @@ export function createWorkflowManager(
       return;
     }
 
-    if (msg.type === "thread-command-event") {
+    if (msg.type === "thread-workflow-message") {
       logStore.append({
         source: "workflow",
-        type: "thread_command_event",
+        type: "thread_workflow_message",
         refId: msg.runId,
-        payload: JSON.stringify(msg.event),
+        payload: JSON.stringify(msg.message),
         ts: Date.now(),
       });
       return;
@@ -464,7 +535,7 @@ export function createWorkflowManager(
     return entry;
   }
 
-  function startWorkflow(workflowName: string, payload: unknown): void {
+  function startWorkflow(workflowName: string, launch: WorkflowLaunchParams): void {
     if (stopped) return;
 
     const wfConfig = workflowConfig(workflowName);
@@ -477,9 +548,10 @@ export function createWorkflowManager(
 
     const state = getOrCreateState(workflowName);
     const runId = crypto.randomUUID();
+    const { prompt, maxRounds, dryRun } = launch;
 
     if (state.active.size < wfConfig.concurrency) {
-      dispatchThread(workflowName, runId, payload);
+      dispatchThread(workflowName, runId, prompt, maxRounds, dryRun);
       return;
     }
 
@@ -504,7 +576,7 @@ export function createWorkflowManager(
       }
     }
 
-    state.queue.push({ runId, payload });
+    state.queue.push({ runId, prompt, maxRounds, dryRun });
     logWorkflowEvent(workflowName, runId, "queued");
     process.stderr.write(
       `[workflow-manager] queued thread for "${workflowName}" runId "${runId}" (queue length: ${state.queue.length})\n`,

packages/daemon/src/workflow-worker.ts

@@ -13,13 +13,19 @@ import { existsSync } from "node:fs";
 import { join, resolve } from "node:path";
 
 import type {
-  CommandEvent,
+  ModeratorContext,
-  ThreadState,
+  RoleMeta,
-  WorkflowContext,
+  StartSignal,
   WorkflowDefinition,
+  WorkflowMessage,
 } from "@uncaged/nerve-core";
+import { END, START, isPlainRecord } from "@uncaged/nerve-core";
 
-import type { ThreadCommandEventMessage, ThreadEventType, WorkerToParentMessage } from "./ipc.js";
+import type {
+  ThreadEventType,
+  ThreadWorkflowMessageMessage,
+  WorkerToParentMessage,
+} from "./ipc.js";
 import { parseParentMessage } from "./ipc.js";
 import { ignoreSessionBroadcastSignals } from "./worker-fork-support.js";
 
@@ -45,147 +51,223 @@ function sendWorkflowError(runId: string, error: string): void {
   send({ type: "workflow-error", runId, error });
 }
 
-function sendCommandEvent(runId: string, event: CommandEvent): void {
+function sendWorkflowMessage(runId: string, message: WorkflowMessage): void {
-  const msg: ThreadCommandEventMessage = {
+  const msg: ThreadWorkflowMessageMessage = {
-    type: "thread-command-event",
+    type: "thread-workflow-message",
     runId,
-    event: event as { type: string; [key: string]: unknown },
+    message: {
+      role: message.role,
+      content: message.content,
+      meta: message.meta,
+      timestamp: message.timestamp,
+    },
   };
   send(msg);
 }
 
 // ---------------------------------------------------------------------------
-// Thread loop (RFC-002 §5.4)
+// Thread loop (signal-driven automaton, issue #80)
 // ---------------------------------------------------------------------------
 
-/**
+function validateRoleResult(
- * Replay persisted events through moderate() to reconstruct ThreadState,
+  result: { content: string; meta: Record<string, unknown> },
- * then execute the next role and return the resulting CommandEvent.
+  roleName: string,
- * Returns null if the thread is already complete (moderate returned null).
- */
-async function replayAndResume(
-  def: WorkflowDefinition,
   runId: string,
-  ctx: WorkflowContext,
+): boolean {
-  state: ThreadState,
+  if (typeof result.content !== "string") {
-  resumeEvents: CommandEvent[],
+    sendWorkflowError(runId, `Role "${roleName}" returned non-string content`);
-): Promise<CommandEvent | null> {
+    return false;
-  let lastNext: ReturnType<typeof def.moderate> = null;
+  }
-  for (const ev of resumeEvents) {
+  if (result.meta === null || typeof result.meta !== "object" || Array.isArray(result.meta)) {
-    state.events.push(ev);
+    sendWorkflowError(runId, `Role "${roleName}" returned invalid meta (must be a plain object)`);
-    lastNext = def.moderate(state, ev);
+    return false;
-    if (lastNext === null) {
+  }
-      sendThreadEvent(runId, "completed", null);
+  return true;
-      return null;
+}
+
+function isStartMeta(meta: unknown): meta is StartSignal["meta"] {
+  return (
+    isPlainRecord(meta) && typeof meta.maxRounds === "number" && typeof meta.dryRun === "boolean"
+  );
+}
+
+function normalizeStartMeta(meta: unknown, maxRoundsFallback: number): StartSignal["meta"] {
+  if (!isPlainRecord(meta)) {
+    return { maxRounds: maxRoundsFallback, dryRun: false };
+  }
+  const maxRounds = typeof meta.maxRounds === "number" ? meta.maxRounds : maxRoundsFallback;
+  const dryRun = typeof meta.dryRun === "boolean" ? meta.dryRun : false;
+  return { maxRounds, dryRun };
+}
+
+function startSignalFromWorkflowMessage(
+  msg: WorkflowMessage,
+  maxRoundsFallback: number,
+): StartSignal {
+  if (msg.role !== START) {
+    return {
+      role: START,
+      content: "",
+      meta: { maxRounds: maxRoundsFallback, dryRun: false },
+      timestamp: Date.now(),
+    };
+  }
+  const meta = isStartMeta(msg.meta) ? msg.meta : normalizeStartMeta(msg.meta, maxRoundsFallback);
+  return {
+    role: START,
+    content: msg.content,
+    meta,
+    timestamp: msg.timestamp,
+  };
+}
+
+type ThreadMessagesState = {
+  start: StartSignal;
+  /** Role outputs only; never includes the `__start__` frame. */
+  messages: WorkflowMessage[];
+};
+
+function initThreadMessages(
+  runId: string,
+  resumeMessages: WorkflowMessage[],
+  freshPrompt: string | null,
+  maxRounds: number,
+  dryRun: boolean,
+): ThreadMessagesState {
+  if (resumeMessages.length > 0) {
+    const [first, ...rest] = resumeMessages;
+    if (first.role === START) {
+      return {
+        start: startSignalFromWorkflowMessage(first, maxRounds),
+        messages: [...rest],
+      };
     }
+    const prompt = freshPrompt ?? "";
+    return {
+      start: {
+        role: START,
+        content: prompt,
+        meta: { maxRounds, dryRun },
+        timestamp: Date.now(),
+      },
+      messages: [...resumeMessages],
+    };
   }
+  const prompt = freshPrompt ?? "";
+  const start: StartSignal = {
+    role: START,
+    content: prompt,
+    meta: { maxRounds, dryRun },
+    timestamp: Date.now(),
+  };
+  sendWorkflowMessage(runId, {
+    role: start.role,
+    content: start.content,
+    meta: start.meta,
+    timestamp: start.timestamp,
+  });
+  return { start, messages: [] };
+}
 
-  const next = lastNext;
+async function executeRole(
-  if (next === null) {
+  def: WorkflowDefinition<RoleMeta>,
-    sendThreadEvent(runId, "completed", null);
+  nextRole: string,
-    return null;
+  start: StartSignal,
-  }
+  messages: WorkflowMessage[],
-
+  runId: string,
-  const role = def.roles[next.role];
+): Promise<{ content: string; meta: Record<string, unknown> } | null> {
+  const role = def.roles[nextRole];
   if (!role) {
-    sendWorkflowError(runId, `Unknown role: ${next.role}`);
+    sendWorkflowError(runId, `Unknown role: ${nextRole}`);
     return null;
   }
 
+  let result: { content: string; meta: Record<string, unknown> };
   try {
-    const event = await role.execute(next.prompt, ctx);
+    result = await role(start, messages);
-    sendCommandEvent(runId, event);
-    return event;
   } catch (e: unknown) {
     const errMsg = e instanceof Error ? e.message : String(e);
     sendThreadEvent(runId, "failed", { error: errMsg });
     return null;
   }
+
+  if (!validateRoleResult(result, nextRole, runId)) return null;
+  return result;
 }
 
 async function runThread(
-  def: WorkflowDefinition,
+  def: WorkflowDefinition<RoleMeta>,
-  workflowName: string,
   runId: string,
-  triggerPayload: unknown,
+  maxRounds: number,
-  /** Pre-existing event history for crash-recovery resume. Empty for a fresh thread. */
+  resumeMessages: WorkflowMessage[] = [],
-  resumeEvents: CommandEvent[] = [],
+  freshPrompt: string | null = null,
+  dryRun = false,
 ): Promise<void> {
-  const state: ThreadState = { runId, events: [] };
+  const { start, messages: roleMessages } = initThreadMessages(
-  const ctx: WorkflowContext = {
     runId,
-    workflowName,
+    resumeMessages,
-    log: (msg) => sendThreadEvent(runId, "step_complete", { message: msg }),
+    freshPrompt,
-  };
+    maxRounds,
+    dryRun,
+  );
 
-  const initialEvent: CommandEvent = {
+  let roleRound = roleMessages.length;
-    type: "thread_start",
+  let nextRole = def.moderator({ kind: "start", start }, roleRound, maxRounds);
-    triggerPayload:
-      triggerPayload != null && typeof triggerPayload === "object" ? triggerPayload : {},
-  };
 
-  // On resume: replay persisted events, run the next un-executed role, then continue.
+  if (nextRole === END) {
-  if (resumeEvents.length > 0) {
+    sendThreadEvent(runId, "completed", null);
-    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.
+  while (roleRound < maxRounds) {
-  sendCommandEvent(runId, initialEvent);
+    const result = await executeRole(def, nextRole, start, roleMessages, runId);
-  await continueThread(def, runId, ctx, state, initialEvent);
+    if (result === null) return;
-}
 
-async function continueThread(
+    const message: WorkflowMessage = {
-  def: WorkflowDefinition,
+      role: nextRole,
-  runId: string,
+      content: result.content,
-  ctx: WorkflowContext,
+      meta: result.meta,
-  state: ThreadState,
+      timestamp: Date.now(),
-  firstEvent: CommandEvent,
+    };
-): Promise<void> {
+    roleMessages.push(message);
-  let event = firstEvent;
+    sendWorkflowMessage(runId, message);
 
-  const MAX_STEPS = 1000;
+    roleRound += 1;
-  let step = 0;
-  while (step < MAX_STEPS) {
-    step++;
-    state.events.push(event);
-    const next = def.moderate(state, event);
 
-    if (next === null) {
+    const stepContext: ModeratorContext<RoleMeta> = {
+      kind: "step",
+      signal: { role: nextRole, meta: result.meta },
+    };
+    nextRole = def.moderator(stepContext, roleRound, maxRounds);
+
+    if (nextRole === END) {
       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})`);
   }
+
+  sendWorkflowError(runId, `Thread exceeded maximum rounds (${maxRounds})`);
 }
 
 // ---------------------------------------------------------------------------
 // Workflow definition loader
 // ---------------------------------------------------------------------------
 
+function isWorkflowDefinitionShape(def: unknown): def is WorkflowDefinition<RoleMeta> {
+  if (!isPlainRecord(def)) return false;
+  return (
+    typeof def.moderator === "function" &&
+    typeof def.roles === "object" &&
+    def.roles !== null &&
+    !Array.isArray(def.roles) &&
+    typeof def.name === "string"
+  );
+}
+
 async function loadWorkflowDefinition(
   nerveRoot: string,
   workflowName: string,
-): Promise<WorkflowDefinition> {
+): Promise<WorkflowDefinition<RoleMeta>> {
   const candidates = [
     resolve(join(nerveRoot, "workflows", workflowName, "index.ts")),
     resolve(join(nerveRoot, "workflows", workflowName, "index.js")),
@@ -202,18 +284,13 @@ async function loadWorkflowDefinition(
   const mod = await import(indexPath);
   const def: unknown = mod.default ?? mod;
 
-  if (
+  if (!isWorkflowDefinitionShape(def)) {
-    def === null ||
-    typeof def !== "object" ||
-    typeof (def as WorkflowDefinition).moderate !== "function" ||
-    typeof (def as WorkflowDefinition).roles !== "object"
-  ) {
     throw new Error(
-      `Workflow "${workflowName}" must export a WorkflowDefinition with "roles" and "moderate".`,
+      `Workflow "${workflowName}" must export a WorkflowDefinition with "name", "roles", and "moderator".`,
     );
   }
 
-  return def as WorkflowDefinition;
+  return def;
 }
 
 // ---------------------------------------------------------------------------
@@ -222,8 +299,7 @@ async function loadWorkflowDefinition(
 
 function handleMessage(
   raw: unknown,
-  def: WorkflowDefinition,
+  def: WorkflowDefinition<RoleMeta>,
-  workflowName: string,
   inFlight: Map<string, Promise<void>>,
   shuttingDown: { value: boolean },
 ): void {
@@ -246,11 +322,11 @@ function handleMessage(
 
   if (msg.type === "start-thread") {
     if (shuttingDown.value) return;
-    const { runId, triggerPayload } = msg;
+    const { runId, prompt, maxRounds, dryRun } = msg;
 
     const previous = inFlight.get(runId) ?? Promise.resolve();
     const next = previous
-      .then(() => runThread(def, workflowName, runId, triggerPayload))
+      .then(() => runThread(def, runId, maxRounds, [], prompt, dryRun))
       .catch((e: unknown) => {
         const errMsg = e instanceof Error ? e.message : String(e);
         sendWorkflowError(runId, errMsg);
@@ -265,11 +341,11 @@ function handleMessage(
 
   if (msg.type === "resume-thread") {
     if (shuttingDown.value) return;
-    const { runId, events, triggerPayload } = msg;
+    const { runId, messages, maxRounds, dryRun } = msg;
 
     const previous = inFlight.get(runId) ?? Promise.resolve();
     const next = previous
-      .then(() => runThread(def, workflowName, runId, triggerPayload, events as CommandEvent[]))
+      .then(() => runThread(def, runId, maxRounds, messages, null, dryRun))
       .catch((e: unknown) => {
         const errMsg = e instanceof Error ? e.message : String(e);
         sendWorkflowError(runId, errMsg);
@@ -288,7 +364,7 @@ function handleMessage(
 // ---------------------------------------------------------------------------
 
 async function bootstrap(nerveRoot: string, workflowName: string): Promise<void> {
-  let def: WorkflowDefinition;
+  let def: WorkflowDefinition<RoleMeta>;
   try {
     def = await loadWorkflowDefinition(nerveRoot, workflowName);
   } catch (e: unknown) {
@@ -303,7 +379,7 @@ async function bootstrap(nerveRoot: string, workflowName: string): Promise<void>
   sendReady();
 
   process.on("message", (raw: unknown) => {
-    handleMessage(raw, def, workflowName, inFlight, shuttingDown);
+    handleMessage(raw, def, inFlight, shuttingDown);
   });
 }
 

packages/daemon/tsup.config.ts

@@ -1,8 +0,0 @@
-import { defineConfig } from "tsup";
-
-export default defineConfig({
-  entry: ["src/index.ts", "src/sense-worker.ts"],
-  format: ["esm"],
-  dts: true,
-  clean: true,
-});

packages/store/README.md

@@ -0,0 +1,48 @@
+# @uncaged/nerve-store
+
+Persistent storage for the [nerve](../../README.md) daemon — append-only structured logs, optional JSONL cold archive, and content-addressable blobs.
+
+## LogStore (`createLogStore`, `log-store.ts`)
+
+- **Append-only log table** — rows with `source`, `type`, `refId`, `payload`, `ts` (string payloads for ad hoc fields)
+- **SQLite WAL** — `DatabaseSync` from `node:sqlite`
+- **Workflow run tracking** — materialized `workflow_runs` table plus helpers to list active runs, upsert status transitions, and read **thread messages** / **role rounds** for CLI and crash recovery
+- **Meta key-value** — small `meta` table (e.g. archive watermarks)
+
+Public exports include `LogStore`, `LogEntry`, `LogQuery`, `WorkflowRun`, `WorkflowRunStatus`, `ThreadRoundRow`, `GetThreadRoundsParams`, and archive-related types re-exported from `log-archive`.
+
+## WorkflowRunStatus
+
+Runs progress through a small state machine. Typical paths:
+
+1. **`queued`** → **`started`** when a worker picks up the thread
+2. **`started`** → **`completed`** | **`failed`** | **`crashed`** | **`interrupted`** | **`dropped`**
+
+Semantics in the daemon/store layer:
+
+- **`completed` / `failed`** — normal terminal outcomes from the workflow worker
+- **`crashed`** — worker exited unexpectedly; manager may respawn and **`resume-thread`** eligible **`started`** runs
+- **`interrupted`** — e.g. hot-reload drain killed an in-flight thread after timeout
+- **`dropped`** — concurrency **`overflow: drop`** rejected a new run, or **`overflow: queue`** evicted an queued item when the queue was full
+
+## LogArchive (`log-archive.ts`)
+
+- **`archiveLogs`** / helpers — export eligible UTC days of old rows to **`data/archive/logs/YYYY-MM-DD.jsonl`**, delete archived rows from SQLite, optional **`VACUUM`**
+- Used by **`nerve store archive`** in `@uncaged/nerve-cli`
+
+## BlobStore (`createBlobStore`, `blob-store.ts`)
+
+- **Content-addressable storage** — `write` returns lowercase **sha256** hex; files live under **`data/blobs/<2-hex>/<62-hex>`**
+- **`read` / `exists`** — path must match digest on disk (tamper detection)
+
+## Install
+
+```bash
+pnpm add @uncaged/nerve-store
+```
+
+Requires Node.js ≥ 22.5 (same as the rest of the stack).
+
+## License
+
+MIT

packages/store/package.json

@@ -0,0 +1,24 @@
+{
+  "name": "@uncaged/nerve-store",
+  "version": "0.4.0",
+  "type": "module",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "files": ["dist"],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "prepublishOnly": "bash ../../scripts/prepublish-check.sh",
+    "build": "rslib build",
+    "test": "vitest run"
+  },
+  "dependencies": {
+    "@uncaged/nerve-core": "workspace:*"
+  },
+  "devDependencies": {
+    "@rslib/core": "^0.21.3",
+    "@types/node": "^22.0.0",
+    "vitest": "^4.1.5"
+  }
+}

packages/store/rslib.config.ts

@@ -0,0 +1,19 @@
+import { defineConfig } from "@rslib/core";
+
+export default defineConfig({
+  lib: [
+    {
+      format: "esm",
+      dts: true,
+    },
+  ],
+  source: {
+    entry: {
+      index: "src/index.ts",
+    },
+  },
+  output: {
+    target: "node",
+    cleanDistPath: true,
+  },
+});

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

@@ -195,4 +195,65 @@ describe("LogStore — crash recovery helpers (Phase 3)", () => {
       expect(result8[0].type).toBe("event_for_8");
     });
   });
+
+  describe("getThreadRoundCount / getThreadRounds", () => {
+    it("excludes thread_start from rounds and assigns ROW_NUMBER in chronological order", () => {
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-tr",
+        payload: JSON.stringify({ type: "thread_start", triggerPayload: { x: 1 } }),
+        ts: 100,
+      });
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-tr",
+        payload: JSON.stringify({
+          type: "step_a",
+          role: "alpha",
+          content: "hello",
+          meta: 1,
+        }),
+        ts: 101,
+      });
+      store.append({
+        source: "workflow",
+        type: "thread_command_event",
+        refId: "run-tr",
+        payload: JSON.stringify({ type: "step_b", role: "beta", content: "world" }),
+        ts: 102,
+      });
+
+      expect(store.getThreadRoundCount("run-tr")).toBe(2);
+
+      const all = store.getThreadRounds("run-tr", { before: 0, limit: 50 });
+      expect(all).toHaveLength(2);
+      expect(all.map((r) => r.round)).toEqual([2, 1]);
+      expect(all[0].message.role).toBe("beta");
+      expect(all[1].message.role).toBe("alpha");
+    });
+
+    it("getThreadRounds respects exclusive before bound", () => {
+      for (let i = 0; i < 3; i++) {
+        store.append({
+          source: "workflow",
+          type: "thread_command_event",
+          refId: "run-b4",
+          payload: JSON.stringify({ type: `ev_${i}`, role: "r", content: String(i) }),
+          ts: 200 + i,
+        });
+      }
+
+      expect(store.getThreadRoundCount("run-b4")).toBe(3);
+
+      const page = store.getThreadRounds("run-b4", { before: 3, limit: 50 });
+      expect(page.map((r) => r.round)).toEqual([2, 1]);
+    });
+
+    it("returns empty when no role rounds for runId", () => {
+      expect(store.getThreadRoundCount("missing")).toBe(0);
+      expect(store.getThreadRounds("missing", { before: 0, limit: 10 })).toHaveLength(0);
+    });
+  });
 });

packages/store/src/index.ts

@@ -0,0 +1,16 @@
+/**
+ * @uncaged/nerve-store — append-only log storage, cold-archive helpers, CAS blob store.
+ */
+
+export * from "./blob-store.js";
+export * from "./log-archive.js";
+export { createLogStore } from "./log-store.js";
+export type {
+  GetThreadRoundsParams,
+  LogEntry,
+  LogQuery,
+  LogStore,
+  ThreadRoundRow,
+  WorkflowRun,
+  WorkflowRunStatus,
+} from "./log-store.js";

packages/store/src/log-store.ts

@@ -9,8 +9,9 @@
 
 import { mkdirSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
-import Database from "better-sqlite3";
+import { DatabaseSync, type StatementSync } from "node:sqlite";
-import type BetterSqlite3 from "better-sqlite3";
+
+import { isPlainRecord } from "@uncaged/nerve-core";
 
 import {
   DEFAULT_LOG_RETENTION_MS,
@@ -69,11 +70,15 @@ const VALID_WORKFLOW_STATUSES = new Set<string>([
   "interrupted",
 ]);
 
+function isWorkflowRunStatus(value: string): value is WorkflowRunStatus {
+  return VALID_WORKFLOW_STATUSES.has(value);
+}
+
 function validateWorkflowRunStatus(status: string): WorkflowRunStatus {
-  if (!VALID_WORKFLOW_STATUSES.has(status)) {
+  if (!isWorkflowRunStatus(status)) {
     throw new Error(`Invalid workflow run status from DB: "${status}"`);
   }
-  return status as WorkflowRunStatus;
+  return status;
 }
 
 /** One row in the workflow_runs materialized table. */
@@ -84,6 +89,25 @@ export type WorkflowRun = {
   ts: number;
 };
 
+/** One role-produced workflow-message row with 1-based round index (ROW_NUMBER over role messages only). */
+export type ThreadRoundRow = {
+  round: number;
+  logId: number;
+  ts: number;
+  message: { role: string; content: string; meta: unknown; timestamp: number };
+};
+
+/** Parameters for {@link LogStore.getThreadRounds}. */
+export type GetThreadRoundsParams = {
+  /**
+   * Exclusive upper bound on round index (1-based among role events).
+   * Use `0` to include all rounds (subject to `limit`).
+   */
+  before: number;
+  /** Maximum rows returned from the DB (DESC by round). */
+  limit: number;
+};
+
 export type LogStore = {
   append: (entry: Omit<LogEntry, "id">) => LogEntry;
   query: (filter?: LogQuery) => LogEntry[];
@@ -118,9 +142,27 @@ export type LogStore = {
   getTriggerPayload: (runId: string) => unknown;
   /**
    * Get all workflow CommandEvents for a specific run, ordered by id ASC.
-   * Used for crash recovery to rebuild ThreadState.
+   * @deprecated Use getThreadMessages for the new WorkflowMessage format.
    */
   getThreadEvents: (runId: string) => Array<{ type: string; [key: string]: unknown }>;
+  /**
+   * Get all WorkflowMessages for a specific run, ordered by id ASC.
+   * Used for crash recovery to rebuild the message chain.
+   */
+  getThreadMessages: (
+    runId: string,
+  ) => Array<{ role: string; content: string; meta: unknown; timestamp: number }>;
+  /**
+   * Count role command events for a run (excludes `thread_start`/`__start__` messages and invalid payloads).
+   * Round indices for {@link getThreadRounds} are 1..count in chronological order.
+   */
+  getThreadRoundCount: (runId: string) => number;
+  /**
+   * Role rounds for agent-oriented retrieval: each row is one `thread_command_event` or
+   * `thread_workflow_message` whose JSON `type` is not `thread_start` and `role` is not `__start__`,
+   * with `round` from ROW_NUMBER() OVER (ORDER BY id ASC). No schema migration — numbering is computed in SQL.
+   */
+  getThreadRounds: (runId: string, params: GetThreadRoundsParams) => ThreadRoundRow[];
   /**
    * Export logs older than the retention window to `data/archive/logs/YYYY-MM-DD.jsonl`,
    * then delete those rows and advance `meta.archived_up_to` in one transaction per day
@@ -184,7 +226,58 @@ function buildJsonlBody(rows: SqlLogRow[]): string {
   return `${lines.join("\n")}\n`;
 }
 
-function runOptionalVacuum(sqlite: BetterSqlite3.Database, vacuum?: boolean): boolean {
+function runInTransaction<T>(db: DatabaseSync, fn: () => T): T {
+  db.exec("BEGIN IMMEDIATE");
+  try {
+    const out = fn();
+    db.exec("COMMIT");
+    return out;
+  } catch (e) {
+    try {
+      db.exec("ROLLBACK");
+    } catch {
+      // ignore rollback errors
+    }
+    throw e;
+  }
+}
+
+function launchShapeFromRecord(rec: Record<string, unknown>): {
+  prompt: string;
+  maxRounds: number;
+  dryRun: boolean;
+} | null {
+  if (typeof rec.prompt !== "string" || typeof rec.maxRounds !== "number") return null;
+  return {
+    prompt: rec.prompt,
+    maxRounds: rec.maxRounds,
+    dryRun: typeof rec.dryRun === "boolean" ? rec.dryRun : false,
+  };
+}
+
+/** Parse JSON from a workflow `started` log row into a trigger / launch payload for crash recovery. */
+function triggerPayloadFromStartedLogJson(payload: string): unknown | null {
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(payload);
+  } catch {
+    return null;
+  }
+  if (!isPlainRecord(parsed)) return null;
+
+  const direct = launchShapeFromRecord(parsed);
+  if (direct !== null) return direct;
+
+  const inner = parsed.triggerPayload;
+  if (inner !== null && isPlainRecord(inner)) {
+    const fromInner = launchShapeFromRecord(inner);
+    if (fromInner !== null) return fromInner;
+    return inner;
+  }
+  return null;
+}
+
+function runOptionalVacuum(sqlite: DatabaseSync, vacuum?: boolean): boolean {
   if (vacuum !== true) return false;
   sqlite.exec("VACUUM");
   return true;
@@ -199,7 +292,7 @@ function resolveArchiveStartDay(watermark: string | null, minDay: string): strin
 function runArchiveDayLoop(
   dbPath: string,
   options: ArchiveLogsOptions,
-  selectLogsForDayStmt: BetterSqlite3.Statement,
+  selectLogsForDayStmt: StatementSync,
   archiveDayTx: (day: string, start: number, endExclusive: number) => void,
   startDay: string,
   lastDay: string,
@@ -235,8 +328,8 @@ function runArchiveDayLoop(
 export function createLogStore(dbPath: string): LogStore {
   mkdirSync(dirname(dbPath), { recursive: true });
 
-  const sqlite: BetterSqlite3.Database = new Database(dbPath);
+  const sqlite = new DatabaseSync(dbPath);
-  sqlite.pragma("journal_mode = WAL");
+  sqlite.exec("PRAGMA journal_mode=WAL");
   sqlite.exec(SCHEMA_SQL);
 
   const insertStmt = sqlite.prepare(
@@ -264,6 +357,34 @@ export function createLogStore(dbPath: string): LogStore {
     "SELECT payload FROM logs WHERE source = 'workflow' AND type = 'thread_command_event' AND ref_id = ? ORDER BY id ASC",
   );
 
+  const getThreadMessagesStmt = sqlite.prepare(
+    "SELECT payload FROM logs WHERE source = 'workflow' AND type = 'thread_workflow_message' AND ref_id = ? ORDER BY id ASC",
+  );
+
+  const getThreadRoundCountStmt = sqlite.prepare(
+    `SELECT COUNT(*) AS c FROM logs
+     WHERE source = 'workflow' AND type IN ('thread_command_event', 'thread_workflow_message') AND ref_id = ?
+       AND payload IS NOT NULL AND json_valid(payload) = 1
+       AND COALESCE(json_extract(payload, '$.type'), '') != 'thread_start'
+       AND COALESCE(json_extract(payload, '$.role'), '') != '__start__'`,
+  );
+
+  const getThreadRoundsStmt = sqlite.prepare(
+    `WITH numbered AS (
+       SELECT id, ts, payload,
+         ROW_NUMBER() OVER (ORDER BY id ASC) AS rn
+       FROM logs
+       WHERE source = 'workflow' AND type IN ('thread_command_event', 'thread_workflow_message') AND ref_id = @runId
+         AND payload IS NOT NULL AND json_valid(payload) = 1
+         AND COALESCE(json_extract(payload, '$.type'), '') != 'thread_start'
+         AND COALESCE(json_extract(payload, '$.role'), '') != '__start__'
+     )
+     SELECT id, ts, payload, rn FROM numbered
+     WHERE (@before = 0 OR rn < @before)
+     ORDER BY rn DESC
+     LIMIT @lim`,
+  );
+
   const getActiveWorkflowRunsStmt = sqlite.prepare(
     "SELECT run_id, workflow, status, ts FROM workflow_runs WHERE status IN ('queued', 'started') ORDER BY ts ASC",
   );
@@ -288,8 +409,8 @@ export function createLogStore(dbPath: string): LogStore {
     "DELETE FROM logs WHERE ts >= @start AND ts < @endExclusive",
   );
 
-  const upsertWorkflowRunTx = sqlite.transaction(
+  function upsertWorkflowRunTx(entry: Omit<LogEntry, "id">, run: WorkflowRun): LogEntry {
-    (entry: Omit<LogEntry, "id">, run: WorkflowRun) => {
+    return runInTransaction(sqlite, () => {
       const info = insertStmt.run({
         source: entry.source,
         type: entry.type,
@@ -304,8 +425,8 @@ export function createLogStore(dbPath: string): LogStore {
         ts: run.ts,
       });
       return { ...entry, id: Number(info.lastInsertRowid) };
-    },
+    });
-  );
+  }
 
   function append(entry: Omit<LogEntry, "id">): LogEntry {
     const info = insertStmt.run({
@@ -320,7 +441,7 @@ export function createLogStore(dbPath: string): LogStore {
 
   function query(filter: LogQuery = {}): LogEntry[] {
     const conditions: string[] = [];
-    const params: Record<string, unknown> = {};
+    const params: Record<string, string | number> = {};
 
     if (filter.source !== undefined) {
       conditions.push("source = @source");
@@ -376,11 +497,11 @@ export function createLogStore(dbPath: string): LogStore {
   }
 
   function upsertWorkflowRun(entry: Omit<LogEntry, "id">, run: WorkflowRun): LogEntry {
-    return upsertWorkflowRunTx(entry, run) as LogEntry;
+    return upsertWorkflowRunTx(entry, run);
   }
 
   function appendWithWorkflowUpdate(entry: Omit<LogEntry, "id">, run: WorkflowRun): LogEntry {
-    return upsertWorkflowRunTx(entry, run) as LogEntry;
+    return upsertWorkflowRunTx(entry, run);
   }
 
   function getWorkflowRun(runId: string): WorkflowRun | null {
@@ -427,16 +548,7 @@ export function createLogStore(dbPath: string): LogStore {
   function getTriggerPayload(runId: string): unknown {
     const row = getTriggerPayloadStmt.get(runId) as { payload: string | null } | undefined;
     if (row === undefined || row.payload === null) return null;
-    try {
+    return triggerPayloadFromStartedLogJson(row.payload);
-      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 }> {
@@ -445,12 +557,8 @@ export function createLogStore(dbPath: string): LogStore {
     for (const row of rows) {
       if (row.payload === null) continue;
       try {
-        const parsed = JSON.parse(row.payload) as unknown;
+        const parsed: unknown = JSON.parse(row.payload);
-        if (
+        if (isPlainRecord(parsed) && typeof parsed.type === "string") {
-          parsed !== null &&
-          typeof parsed === "object" &&
-          typeof (parsed as Record<string, unknown>).type === "string"
-        ) {
           result.push(parsed as { type: string; [key: string]: unknown });
         }
       } catch {
@@ -460,10 +568,107 @@ export function createLogStore(dbPath: string): LogStore {
     return result;
   }
 
-  const archiveDayTx = sqlite.transaction((day: string, start: number, endExclusive: number) => {
+  function tryParseWorkflowMessage(
-    deleteLogsForDayStmt.run({ start, endExclusive });
+    payload: string,
-    setMetaStmt.run({ key: LOG_ARCHIVE_META_KEY, value: day });
+  ): { role: string; content: string; meta: unknown; timestamp: number } | null {
-  });
+    try {
+      const parsed: unknown = JSON.parse(payload);
+      if (!isPlainRecord(parsed)) return null;
+      const obj = parsed;
+      if (typeof obj.role !== "string" || typeof obj.content !== "string") return null;
+      return {
+        role: obj.role,
+        content: obj.content,
+        meta: obj.meta,
+        timestamp: typeof obj.timestamp === "number" ? obj.timestamp : 0,
+      };
+    } catch {
+      return null;
+    }
+  }
+
+  function getThreadMessages(
+    runId: string,
+  ): Array<{ role: string; content: string; meta: unknown; timestamp: number }> {
+    const rows = getThreadMessagesStmt.all(runId) as Array<{ payload: string | null }>;
+    const result: Array<{ role: string; content: string; meta: unknown; timestamp: number }> = [];
+    for (const row of rows) {
+      if (row.payload === null) continue;
+      const msg = tryParseWorkflowMessage(row.payload);
+      if (msg !== null) result.push(msg);
+    }
+    return result;
+  }
+
+  function getThreadRoundCount(runId: string): number {
+    const row = getThreadRoundCountStmt.get(runId) as { c: number } | undefined;
+    const c = row?.c;
+    if (c === null || c === undefined) return 0;
+    return Number(c);
+  }
+
+  function recordToRoundMessage(
+    obj: Record<string, unknown>,
+    fallbackTs: number,
+  ): { role: string; content: string; meta: unknown; timestamp: number } | null {
+    if (typeof obj.role === "string" && typeof obj.content === "string") {
+      return {
+        role: obj.role,
+        content: obj.content,
+        meta: obj.meta,
+        timestamp: typeof obj.timestamp === "number" ? obj.timestamp : 0,
+      };
+    }
+    if (typeof obj.type === "string") {
+      return {
+        role: typeof obj.role === "string" ? obj.role : obj.type,
+        content: typeof obj.content === "string" ? obj.content : JSON.stringify(obj),
+        meta: obj,
+        timestamp: fallbackTs,
+      };
+    }
+    return null;
+  }
+
+  function parseRoundPayload(
+    payload: string,
+    fallbackTs: number,
+  ): { role: string; content: string; meta: unknown; timestamp: number } | null {
+    try {
+      const parsed: unknown = JSON.parse(payload);
+      if (!isPlainRecord(parsed)) return null;
+      return recordToRoundMessage(parsed, fallbackTs);
+    } catch {
+      return null;
+    }
+  }
+
+  function getThreadRounds(runId: string, params: GetThreadRoundsParams): ThreadRoundRow[] {
+    if (params.limit < 1) return [];
+
+    const rows = getThreadRoundsStmt.all({
+      runId,
+      before: params.before,
+      lim: params.limit,
+    }) as Array<{ id: number; ts: number; payload: string | null; rn: number }>;
+
+    const out: ThreadRoundRow[] = [];
+    for (const row of rows) {
+      if (row.payload === null) continue;
+      const message = parseRoundPayload(row.payload, row.ts);
+      if (message !== null) {
+        out.push({ round: row.rn, logId: row.id, ts: row.ts, message });
+      }
+    }
+    return out;
+  }
+
+  function archiveDayTx(day: string, start: number, endExclusive: number): void {
+    runInTransaction(sqlite, () => {
+      deleteLogsForDayStmt.run({ start, endExclusive });
+      setMetaStmt.run({ key: LOG_ARCHIVE_META_KEY, value: day });
+    });
+  }
 
   function readWatermark(): string | null {
     const raw = getMeta(LOG_ARCHIVE_META_KEY);
@@ -522,6 +727,9 @@ export function createLogStore(dbPath: string): LogStore {
     getAllWorkflowRuns,
     getTriggerPayload,
     getThreadEvents,
+    getThreadMessages,
+    getThreadRoundCount,
+    getThreadRounds,
     archiveLogs,
     close,
   };

packages/store/tsconfig.json

@@ -0,0 +1,9 @@
+{
+  "extends": "../../tsconfig.json",
+  "compilerOptions": {
+    "outDir": "dist",
+    "rootDir": "src",
+    "composite": false
+  },
+  "include": ["src"]
+}

packages/workflow-utils/package.json

@@ -0,0 +1,25 @@
+{
+  "name": "@uncaged/nerve-workflow-utils",
+  "version": "0.4.0",
+  "type": "module",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "files": ["dist"],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "prepublishOnly": "bash ../../scripts/prepublish-check.sh",
+    "build": "rslib build",
+    "test": "vitest run"
+  },
+  "dependencies": {
+    "@uncaged/nerve-core": "workspace:*",
+    "zod": "^4.3.6"
+  },
+  "devDependencies": {
+    "@rslib/core": "^0.21.3",
+    "@types/node": "^22.0.0",
+    "vitest": "^4.1.5"
+  }
+}

packages/workflow-utils/rslib.config.ts

@@ -0,0 +1,19 @@
+import { defineConfig } from "@rslib/core";
+
+export default defineConfig({
+  lib: [
+    {
+      format: "esm",
+      dts: true,
+    },
+  ],
+  source: {
+    entry: {
+      index: "src/index.ts",
+    },
+  },
+  output: {
+    target: "node",
+    cleanDistPath: true,
+  },
+});

packages/workflow-utils/src/__tests__/llm-extract.test.ts

@@ -0,0 +1,133 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { z } from "zod";
+
+import { llmExtract } from "../llm-extract.js";
+
+describe("llmExtract", () => {
+  afterEach(() => {
+    vi.unstubAllGlobals();
+    vi.restoreAllMocks();
+  });
+
+  it("parses tool call arguments and validates with the zod schema", async () => {
+    const schema = z
+      .object({
+        name: z.string(),
+        description: z.string(),
+      })
+      .describe("Extract sense metadata from plan");
+
+    const fetchMock = vi.fn().mockResolvedValue({
+      ok: true,
+      status: 200,
+      text: async () =>
+        JSON.stringify({
+          choices: [
+            {
+              message: {
+                tool_calls: [
+                  {
+                    function: {
+                      name: "extract",
+                      arguments: JSON.stringify({ name: "cpu-usage", description: "CPU load" }),
+                    },
+                  },
+                ],
+              },
+            },
+          ],
+        }),
+    });
+
+    vi.stubGlobal("fetch", fetchMock);
+
+    const result = await llmExtract({
+      text: "some plan",
+      schema,
+      provider: {
+        baseUrl: "https://example.com/v1",
+        apiKey: "k",
+        model: "m",
+      },
+    });
+
+    expect(result.ok).toBe(true);
+    if (!result.ok) {
+      return;
+    }
+    expect(result.value).toEqual({ name: "cpu-usage", description: "CPU load" });
+
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+    const [, init] = fetchMock.mock.calls[0] as [string, RequestInit];
+    expect(init.method).toBe("POST");
+    expect(init.headers).toMatchObject({
+      Authorization: "Bearer k",
+      "Content-Type": "application/json",
+    });
+    const body = JSON.parse(init.body as string) as {
+      model: string;
+      tool_choice: { function: { name: string } };
+    };
+    expect(body.model).toBe("m");
+    expect(body.tool_choice.function.name).toBeDefined();
+  });
+
+  it("returns schema_validation_failed when arguments do not match the schema", async () => {
+    const schema = z.object({ n: z.number() });
+
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockResolvedValue({
+        ok: true,
+        status: 200,
+        text: async () =>
+          JSON.stringify({
+            choices: [
+              {
+                message: {
+                  tool_calls: [
+                    { function: { name: "extract", arguments: JSON.stringify({ n: "oops" }) } },
+                  ],
+                },
+              },
+            ],
+          }),
+      }),
+    );
+
+    const result = await llmExtract({
+      text: "x",
+      schema,
+      provider: { baseUrl: "https://example.com", apiKey: "k", model: "m" },
+    });
+
+    expect(result.ok).toBe(false);
+    if (result.ok) {
+      return;
+    }
+    expect(result.error.kind).toBe("schema_validation_failed");
+  });
+
+  it("dryRun skips fetch and returns an empty stub value", async () => {
+    const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+    const fetchMock = vi.fn();
+    vi.stubGlobal("fetch", fetchMock);
+
+    const schema = z.object({ n: z.number() });
+    const result = await llmExtract({
+      text: "ignored",
+      schema,
+      provider: { baseUrl: "https://example.com", apiKey: "k", model: "m" },
+      dryRun: true,
+    });
+
+    logSpy.mockRestore();
+
+    expect(fetchMock).not.toHaveBeenCalled();
+    expect(result.ok).toBe(true);
+    if (!result.ok) {
+      return;
+    }
+    expect(result.value).toEqual({});
+  });
+});

packages/workflow-utils/src/__tests__/spawn-safe.test.ts

@@ -0,0 +1,63 @@
+import { describe, expect, it, vi } from "vitest";
+
+import { spawnSafe } from "../spawn-safe.js";
+
+describe("spawnSafe", () => {
+  it("passes argv literally without shell interpretation (injection-safe)", async () => {
+    const injection = "$(echo BAD)";
+    const result = await spawnSafe(
+      process.execPath,
+      ["-e", "process.stdout.write(process.argv[1] ?? '')", injection],
+      { cwd: null, env: null, timeoutMs: 10_000 },
+    );
+
+    expect(result.ok).toBe(true);
+    if (!result.ok) {
+      return;
+    }
+    expect(result.value.stdout).toBe(injection);
+    expect(result.value.exitCode).toBe(0);
+  });
+
+  it("returns err on non-zero exit", async () => {
+    const result = await spawnSafe(process.execPath, ["-e", "process.exit(7)"], {
+      cwd: null,
+      env: null,
+      timeoutMs: 10_000,
+    });
+
+    expect(result.ok).toBe(false);
+    if (result.ok) {
+      return;
+    }
+    expect(result.error.kind).toBe("non_zero_exit");
+    if (result.error.kind !== "non_zero_exit") {
+      return;
+    }
+    expect(result.error.exitCode).toBe(7);
+  });
+
+  it("dryRun skips spawn and returns a zero-exit stub", async () => {
+    const logSpy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+    const result = await spawnSafe(process.execPath, ["-e", "process.exit(1)"], {
+      cwd: null,
+      env: null,
+      timeoutMs: 10_000,
+      dryRun: true,
+    });
+
+    logSpy.mockRestore();
+
+    expect(result.ok).toBe(true);
+    if (!result.ok) {
+      return;
+    }
+    expect(result.value).toEqual({
+      stdout: "[dryRun] skipped",
+      stderr: "",
+      exitCode: 0,
+      signal: null,
+    });
+  });
+});

packages/workflow-utils/src/context.ts

@@ -0,0 +1,47 @@
+import { readFileSync } from "node:fs";
+import { resolve } from "node:path";
+import { type Result, err, ok } from "@uncaged/nerve-core";
+
+export type ReadNerveYamlOptions = {
+  nerveRoot: string;
+};
+
+export type NerveYamlError = {
+  code: "PATH_TRAVERSAL" | "READ_FAILED";
+  message: string;
+};
+
+/**
+ * Reads `nerve.yaml` from a Nerve data directory (typically `~/.uncaged-nerve`).
+ * Returns Result to avoid throwing on expected failures (missing file, bad perms).
+ * Validates that the resolved path stays within nerveRoot to prevent path traversal.
+ */
+export function readNerveYaml(options: ReadNerveYamlOptions): Result<string, NerveYamlError> {
+  const root = resolve(options.nerveRoot);
+  const target = resolve(root, "nerve.yaml");
+
+  if (!target.startsWith(root)) {
+    return err({
+      code: "PATH_TRAVERSAL",
+      message: `Resolved path "${target}" escapes nerveRoot "${root}"`,
+    });
+  }
+
+  try {
+    return ok(readFileSync(target, "utf-8"));
+  } catch (e) {
+    return err({
+      code: "READ_FAILED",
+      message: e instanceof Error ? e.message : String(e),
+    });
+  }
+}
+
+/**
+ * Shared context for workflow agents: how Nerve fits together and common CLI verbs.
+ */
+export const nerveAgentContext = `
+Nerve observes the world through **Senses** (each has its own SQLite DB and a \`compute()\` function).
+**Reflexes** (YAML) schedule sense runs or start **Workflows** on intervals or signals.
+The \`nerve\` CLI manages config, triggers, and queries; keep paths and commands aligned with the host nerve.yaml and senses directory.
+`.trim();

packages/workflow-utils/src/cursor-agent.ts

@@ -0,0 +1,62 @@
+import { type Result, ok } from "@uncaged/nerve-core";
+
+import { type SpawnEnv, type SpawnError, spawnSafe } from "./spawn-safe.js";
+
+export type CursorAgentMode = "plan" | "ask" | "default";
+
+export type CursorAgentOptions = {
+  prompt: string;
+  mode: CursorAgentMode;
+  cwd: string;
+  env: SpawnEnv | null;
+  timeoutMs: number | null;
+  dryRun: boolean;
+};
+
+type CursorAgentOptionsInput = CursorAgentOptions | Omit<CursorAgentOptions, "dryRun">;
+
+function resolveCursorAgentDryRun(options: CursorAgentOptionsInput): boolean {
+  return "dryRun" in options ? options.dryRun : false;
+}
+
+/**
+ * Invokes `cursor-agent` with the prompt passed as a single argv slot (`shell: false`).
+ */
+export async function cursorAgent(
+  options: CursorAgentOptionsInput,
+): Promise<Result<string, SpawnError>> {
+  const dryRun = resolveCursorAgentDryRun(options);
+  if (dryRun) {
+    console.log("[dryRun] cursorAgent:", options.prompt, JSON.stringify(options));
+    return ok("[dryRun] skipped");
+  }
+
+  const args: string[] = [
+    "-p",
+    options.prompt,
+    "--model",
+    "auto",
+    "--output-format",
+    "text",
+    "--trust",
+    "--force",
+  ];
+  if (options.mode === "plan") {
+    args.push("--mode=plan");
+  } else if (options.mode === "ask") {
+    args.push("--mode=ask");
+  }
+
+  const run = await spawnSafe("cursor-agent", args, {
+    cwd: options.cwd,
+    env: options.env,
+    timeoutMs: options.timeoutMs,
+    dryRun: false,
+  });
+
+  if (!run.ok) {
+    return run;
+  }
+
+  return ok(run.value.stdout);
+}

packages/workflow-utils/src/index.ts

@@ -0,0 +1,22 @@
+export { cursorAgent, type CursorAgentMode, type CursorAgentOptions } from "./cursor-agent.js";
+export {
+  nerveAgentContext,
+  readNerveYaml,
+  type NerveYamlError,
+  type ReadNerveYamlOptions,
+} from "./context.js";
+export {
+  llmExtract,
+  type LlmError,
+  type LlmExtractOptions,
+  type LlmProvider,
+} from "./llm-extract.js";
+export {
+  nerveCommandEnv,
+  spawnSafe,
+  type SpawnEnv,
+  type SpawnError,
+  type SpawnResult,
+  type SpawnSafeOptions,
+} from "./spawn-safe.js";
+export { isDryRun } from "./start-signal.js";

packages/workflow-utils/src/llm-extract.ts

@@ -0,0 +1,189 @@
+import { type Result, err, ok } from "@uncaged/nerve-core";
+import { toJSONSchema, type z } from "zod";
+
+export type LlmProvider = {
+  baseUrl: string;
+  apiKey: string;
+  model: string;
+};
+
+export type LlmExtractOptions<T> = {
+  text: string;
+  schema: z.ZodType<T>;
+  provider: LlmProvider;
+  dryRun: boolean;
+};
+
+type LlmExtractOptionsInput<T> = LlmExtractOptions<T> | Omit<LlmExtractOptions<T>, "dryRun">;
+
+function resolveLlmExtractDryRun<T>(options: LlmExtractOptionsInput<T>): boolean {
+  return "dryRun" in options ? options.dryRun : false;
+}
+
+export type LlmError =
+  | { kind: "http_error"; status: number; body: string }
+  | { kind: "invalid_response_json"; message: string }
+  | { kind: "no_tool_call"; preview: string }
+  | { kind: "tool_arguments_invalid_json"; message: string }
+  | { kind: "schema_validation_failed"; message: string }
+  | { kind: "network_error"; message: string };
+
+function chatCompletionsUrl(baseUrl: string): string {
+  const trimmed = baseUrl.replace(/\/+$/, "");
+  return `${trimmed}/chat/completions`;
+}
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+
+function stripJsonSchemaMeta(json: Record<string, unknown>): Record<string, unknown> {
+  const { $schema: _drop, ...rest } = json;
+  return rest;
+}
+
+function readToolName(parametersSchema: Record<string, unknown>): string {
+  const title = parametersSchema.title;
+  if (typeof title === "string" && title.trim().length > 0) {
+    return title.trim();
+  }
+  return "extract";
+}
+
+function readToolArgumentsJson(parsed: unknown, previewSource: string): Result<string, LlmError> {
+  if (!isRecord(parsed)) {
+    return err({ kind: "invalid_response_json", message: "Top-level JSON is not an object" });
+  }
+
+  const choices = parsed.choices;
+  if (!Array.isArray(choices) || choices.length === 0) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const first = choices[0];
+  if (!isRecord(first)) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const messageObj = first.message;
+  if (!isRecord(messageObj)) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const toolCalls = messageObj.tool_calls;
+  if (!Array.isArray(toolCalls) || toolCalls.length === 0) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const call0 = toolCalls[0];
+  if (!isRecord(call0)) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const fn = call0.function;
+  if (!isRecord(fn)) {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  const argsRaw = fn.arguments;
+  if (typeof argsRaw !== "string") {
+    return err({ kind: "no_tool_call", preview: previewSource.slice(0, 500) });
+  }
+
+  return ok(argsRaw);
+}
+
+/**
+ * Calls an OpenAI-compatible chat completions API with `tool_choice` forced to a single function
+ * derived from a Zod v4 schema (`toJSONSchema`). Uses `fetch()` only (no shell).
+ */
+export async function llmExtract<T>(
+  options: LlmExtractOptionsInput<T>,
+): Promise<Result<T, LlmError>> {
+  const dryRun = resolveLlmExtractDryRun(options);
+  if (dryRun) {
+    console.log("[dryRun] llmExtract:", options.text, JSON.stringify(options.schema));
+    return ok({} as T);
+  }
+
+  const rawJsonSchema = toJSONSchema(options.schema) as Record<string, unknown>;
+  const parameters = stripJsonSchemaMeta(rawJsonSchema);
+  const toolName = readToolName(parameters);
+  const toolDescription =
+    typeof options.schema.description === "string" && options.schema.description.trim().length > 0
+      ? options.schema.description.trim()
+      : "Extract structured data from the input text.";
+
+  const body = {
+    model: options.provider.model,
+    messages: [
+      {
+        role: "system" as const,
+        content: "Extract the requested information from the provided text. Be precise.",
+      },
+      { role: "user" as const, content: options.text },
+    ],
+    tools: [
+      {
+        type: "function" as const,
+        function: {
+          name: toolName,
+          description: toolDescription,
+          parameters,
+        },
+      },
+    ],
+    tool_choice: { type: "function" as const, function: { name: toolName } },
+  };
+
+  let response: Response;
+  try {
+    response = await fetch(chatCompletionsUrl(options.provider.baseUrl), {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${options.provider.apiKey}`,
+        "Content-Type": "application/json",
+      },
+      body: JSON.stringify(body),
+    });
+  } catch (cause) {
+    const message = cause instanceof Error ? cause.message : String(cause);
+    return err({ kind: "network_error", message });
+  }
+
+  const responseText = await response.text();
+  if (!response.ok) {
+    return err({ kind: "http_error", status: response.status, body: responseText.slice(0, 4000) });
+  }
+
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(responseText) as unknown;
+  } catch (cause) {
+    const message = cause instanceof Error ? cause.message : String(cause);
+    return err({ kind: "invalid_response_json", message });
+  }
+
+  const argsJson = readToolArgumentsJson(parsed, responseText);
+  if (!argsJson.ok) {
+    return argsJson;
+  }
+
+  let argsParsed: unknown;
+  try {
+    argsParsed = JSON.parse(argsJson.value) as unknown;
+  } catch (cause) {
+    const message = cause instanceof Error ? cause.message : String(cause);
+    return err({ kind: "tool_arguments_invalid_json", message });
+  }
+
+  const validated = options.schema.safeParse(argsParsed);
+  if (!validated.success) {
+    return err({
+      kind: "schema_validation_failed",
+      message: validated.error.message,
+    });
+  }
+
+  return ok(validated.data);
+}

packages/workflow-utils/src/spawn-safe.ts

@@ -0,0 +1,160 @@
+import { spawn } from "node:child_process";
+import { homedir } from "node:os";
+import { join } from "node:path";
+import { type Result, err, ok } from "@uncaged/nerve-core";
+
+/** Compatible with `process.env` for `child_process.spawn`. */
+export type SpawnEnv = Record<string, string | undefined>;
+
+export type SpawnResult = {
+  stdout: string;
+  stderr: string;
+  exitCode: number;
+  /** OS signal name (e.g. `"SIGTERM"`) when terminated by signal; otherwise `null`. */
+  signal: string | null;
+};
+
+export type SpawnError =
+  | {
+      kind: "non_zero_exit";
+      stdout: string;
+      stderr: string;
+      exitCode: number;
+      signal: string | null;
+    }
+  | { kind: "timeout"; stdout: string; stderr: string }
+  | { kind: "spawn_failed"; message: string };
+
+export type SpawnSafeOptions = {
+  cwd: string | null;
+  /** When null, merges {@link nerveCommandEnv} over `process.env`. When set, merges over that default. */
+  env: SpawnEnv | null;
+  timeoutMs: number | null;
+  dryRun: boolean;
+};
+
+type SpawnSafeOptionsInput = SpawnSafeOptions | Omit<SpawnSafeOptions, "dryRun">;
+
+const DEFAULT_TIMEOUT_MS = 300_000;
+
+/**
+ * PATH and PNPM_HOME for running `pnpm` and `nerve` from workflow roles.
+ * Uses the pnpm store home only (no npm user bin); binaries must resolve via PATH.
+ */
+export function nerveCommandEnv(): SpawnEnv {
+  const home = homedir();
+  const pnpmHome = join(home, ".local/share/pnpm");
+  return {
+    ...process.env,
+    PNPM_HOME: pnpmHome,
+    PATH: `${pnpmHome}:${process.env.PATH ?? ""}`,
+  };
+}
+
+function mergeEnv(user: SpawnEnv | null): SpawnEnv {
+  const base = nerveCommandEnv();
+  if (user === null) {
+    return base;
+  }
+  return { ...base, ...user };
+}
+
+function resolveTimeout(timeoutMs: number | null): number {
+  if (timeoutMs === null) {
+    return DEFAULT_TIMEOUT_MS;
+  }
+  return timeoutMs;
+}
+
+function resolveDryRun(options: SpawnSafeOptionsInput): boolean {
+  return "dryRun" in options ? options.dryRun : false;
+}
+
+/**
+ * Spawn a process with `shell: false` (argv only), default {@link nerveCommandEnv}, and optional timeout.
+ * Returns `ok` only when the process exits with code 0.
+ */
+export function spawnSafe(
+  command: string,
+  args: ReadonlyArray<string>,
+  options: SpawnSafeOptionsInput,
+): Promise<Result<SpawnResult, SpawnError>> {
+  const dryRun = resolveDryRun(options);
+  if (dryRun) {
+    console.log("[dryRun] spawnSafe:", command, args, JSON.stringify(options));
+    return Promise.resolve(
+      ok({
+        stdout: "[dryRun] skipped",
+        stderr: "",
+        exitCode: 0,
+        signal: null,
+      }),
+    );
+  }
+
+  return new Promise((resolve) => {
+    const cwd = options.cwd === null ? process.cwd() : options.cwd;
+    const env = mergeEnv(options.env);
+    const timeoutMs = resolveTimeout(options.timeoutMs);
+
+    const child = spawn(command, args, {
+      cwd,
+      env,
+      shell: false,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+
+    let stdout = "";
+    let stderr = "";
+    let settled = false;
+
+    const finish = (outcome: Result<SpawnResult, SpawnError>) => {
+      if (settled) {
+        return;
+      }
+      settled = true;
+      clearTimeout(timer);
+      resolve(outcome);
+    };
+
+    const timer = setTimeout(() => {
+      child.kill("SIGTERM");
+      finish(err({ kind: "timeout", stdout, stderr }));
+    }, timeoutMs);
+
+    child.stdout?.on("data", (chunk: Buffer | string) => {
+      stdout += typeof chunk === "string" ? chunk : chunk.toString("utf-8");
+    });
+    child.stderr?.on("data", (chunk: Buffer | string) => {
+      stderr += typeof chunk === "string" ? chunk : chunk.toString("utf-8");
+    });
+
+    child.on("error", (cause: Error) => {
+      finish(err({ kind: "spawn_failed", message: cause.message }));
+    });
+
+    child.on("close", (code, signal) => {
+      const exitCode = code ?? 1;
+      const sig = signal === undefined || signal === null ? null : String(signal);
+      const result: SpawnResult = {
+        stdout: stdout.trimEnd(),
+        stderr: stderr.trimEnd(),
+        exitCode,
+        signal: sig,
+      };
+      if (exitCode !== 0) {
+        finish(
+          err({
+            kind: "non_zero_exit",
+            stdout: result.stdout,
+            stderr: result.stderr,
+            exitCode,
+            signal: sig,
+          }),
+        );
+        return;
+      }
+      finish(ok(result));
+    });
+  });
+}

packages/workflow-utils/src/start-signal.ts

@@ -0,0 +1,6 @@
+import type { StartSignal } from "@uncaged/nerve-core";
+
+/** Returns the thread-level dry-run flag from the workflow start frame. */
+export function isDryRun(start: StartSignal): boolean {
+  return start.meta.dryRun;
+}

packages/workflow-utils/tsconfig.json

@@ -0,0 +1,9 @@
+{
+  "extends": "../../tsconfig.json",
+  "compilerOptions": {
+    "outDir": "dist",
+    "rootDir": "src",
+    "composite": false
+  },
+  "include": ["src"]
+}

pnpm-workspace.yaml

@@ -3,5 +3,4 @@ packages:
 
 onlyBuiltDependencies:
   - "@biomejs/biome"
-  - better-sqlite3
   - esbuild

tsconfig.json

@@ -17,6 +17,7 @@
   },
   "references": [
     { "path": "packages/core" },
+    { "path": "packages/store" },
     { "path": "packages/cli" },
     { "path": "packages/daemon" }
   ]