# Agent Prompts

LLM agent prompts for Poimen RoutingWorkflow.

## Agents

| Agent | Model | Purpose | Prompt File |
|-------|-------|---------|-------------|
| **Router** | `reasoning` | Natural language → WorkflowSpec | [router/AGENTS.md](router/AGENTS.md) |

---

## Architecture

```
User Request: "Scan repo X for security issues"
     │
     ▼
┌─────────────────────────────────────────────────┐
│            RetrieveMemoryActivity               │
│  Query poimen-memory for relevant skills/lessons│
└─────────────────────────────────────────────────┘
     │
     ▼
┌─────────────────────────────────────────────────┐
│              Router Agent (LLM)                 │
│  Input: message + memory context + activities   │
│  Output: WorkflowSpec JSON                      │
└─────────────────────────────────────────────────┘
     │
     ▼
┌─────────────────────────────────────────────────┐
│              RoutingWorkflow                    │
│  Executes WorkflowSpec as state machine         │
│  Clone → Scan → Report → Notify                 │
└─────────────────────────────────────────────────┘
```

---

## Memory Integration

Router receives context from `RetrieveMemoryActivity`:

```
1. User: "scan repo for security"
2. RetrieveMemoryActivity queries poimen-memory
3. Returns: skills, lessons, references
4. Injected into Router prompt
5. Router generates smarter WorkflowSpec
```

---

## LLM Endpoint

Default: `https://api.riotpiao.com/v1/chat/completions`

Override: `LOCAL_LLM_BASE_URL=http://localhost:11434`