skill-073

Worker Protocol

"Complete work, send message via pipe, exit. Supervisor restarts if needed."

When to Use This Skill

Use when:

• Implementing worker agent behavior
• Understanding when/how to exit
• Understanding supervisor auto-restart

Use proactively:

• Reference before implementing agent message loops
• After completing work, before exiting

---

Quick Start

Connect to watchdog

Connect-ToWatchdog -AgentName "developer" -SessionDir "..claude\session"

Enter message loop

Enter-AgentLoop -MessageHandler { param($Message) switch ($Message.type) { "WorkAssign" { Send-WorkComplete -TaskId $Message.payload.taskId -Result "success" } } }

Example 2: Send work complete
```powershell
Send-WorkComplete -TaskId "feat-001" -Result "success" -Notes "Complete"

Example 3: Exit conditions

• Work complete → WorkComplete → Exit (code 0)
• Need clarification → Query → Exit (code 0)
• Shutdown received → Exit (code 42)

</examples>

---

## Architecture

┌─────────────────────────────────────────────────────────────┐ │ ACTOR SUPERVISOR (Watchdog) │ │ 1. StartActor(agentName) - creates bidirectional pipe │ │ 2. Wait for agent connection via agent-runtime.ps1 │ │ 3. Supervise() - check for crashes, restart with backoff │ │ 4. Route messages between agents via pipes │ │ 5. StopAll() - graceful shutdown │ └───────────────────────────┬───────────────────────────────────┘ │ ┌───────────▼─────────────┐ │ PM ↔ Workers (pipes) │ │ Event Log (source) │ └──────────────────────────┘

---

## Agent Lifecycle

┌─────────────┐ ┌──────────────┐ ┌─────────────┐ │ CONNECT │ ──▶ │ DO WORK │ ──▶ │ EXIT │ │ (agent- │ │ (receive via │ │ (supervisor │ │ runtime) │ │ pipe, │ │ may restart)│ └─────────────┘ │ process) │ └─────────────┘ └──────────────┘

**Key Principle**: Each agent run processes messages, completes work, exits. Supervisor restarts when needed.

---

## Event-Driven vs Sequential Mode

| Aspect | Event-Driven | Sequential |
|--------|-------------|------------|
| Agent spawning | On demand | One at a time |
| Message delivery | Named pipes | Handoff file |
| Parallel work | Yes | No |
| Startup | `/ralph-coordinator-event` | `/ralph-coordinator-single` |
| Connection | `agent-runtime.ps1` | Handoff files |

---

## Exit Conditions

| Condition | Action | Exit Code | Restarted |
|-----------|--------|-----------|-----------|
| Work complete | Send `WorkComplete` | 0 | No (new task will spawn) |
| Need clarification | Send `Query` | 0 | No |
| Blocking issue | Send `WorkBlocked` | 0 | No |
| Shutdown received | - | 42 | No |
| Crash | - | Non-zero | Yes (with backoff) |

**Supervisor restart strategy**:
- Exponential backoff: 5s, 10s, 20s, 40s, 60s (max)
- Max 3 restarts per agent
- Graceful exits (0 or 42) are not restarted

---

## V2 vs V1

| Aspect | V1 (Legacy) | V2 (Current) |
|--------|-------------|--------------|
| Agent lifecycle | File queue monitoring | Pipe-based event loop |
| Delivery latency | 2-5 seconds | <10 milliseconds |
| Crash recovery | Manual health checks | Automatic (ActorSupervisor) |
| State persistence | Multiple files | Single event log |
| Message types | 47+ types | 12 core types |
| Restart strategy | None | Exponential backoff |

---

## Complete Worker Example

```powershell