# Templates

Templates define the system prompt, stop sequences, and assistant message format for an agent. Templates can be applied via `AgentSpec` (YAML definitions) or passed directly in `run()` options.

## System prompt via run options

The simplest way to set a system prompt is through the `system` field in run options:

```typescript
import { Lm, Agent } from "@agent-smith/agent";

const lm = new Lm({
    name: "llamacpp",
    serverUrl: "http://localhost:8080/v1",
    onToken: (t) => process.stdout.write(t),
});

const agent = new Agent({ name: "my-agent", lm });

const prompt = `I am landing in Barcelona soon: I plan to reach my hotel 
and then go for outdoor sport. How are the conditions in the city?`;

const result = await agent.run(prompt, {
    params: {
        model: "Qwen3-30B-A3B-Instruct-2507-UD-Q4_K_XL",
        temperature: 0.4,
        top_k: 40,
        top_p: 0.95,
        max_tokens: 16384,
    },
    system: "You are an autonomous agent: feel free to use your tools.",
});
```

## TemplateSpec via AgentSpec

For reusable agent definitions, use `TemplateSpec` in a YAML-based `AgentSpec`:

```typescript
import { Lm, Agent } from "@agent-smith/agent";
import { readFileSync } from "fs";

const lm = new Lm({
    name: "llamacpp",
    serverUrl: "http://localhost:8080/v1",
    onToken: (t) => process.stdout.write(t),
});

// Load agent spec from YAML
const yamlContent = readFileSync("agents/chat.yml", "utf-8");
const agent = Agent.fromYaml({ lm }, yamlContent);

const prompt = "Tell me about Barcelona weather";
const result = await agent.run(prompt, {
    params: {
        model: "Qwen3-30B-A3B-Instruct-2507-UD-Q4_K_XL",
        temperature: 0.4,
        max_tokens: 16384,
    },
});
```

The YAML agent spec (`agents/chat.yml`) would look like:

```yaml
name: chat
prompt: "{prompt}"
description: "A helpful chat agent"
model: "Qwen3-30B-A3B-Instruct-2507-UD-Q4_K_XL"
template:
    system: "You are a helpful assistant that uses tools when needed."
    stop: ["\n"]
inferParams:
    temperature: 0.4
    max_tokens: 16384
```

## Shots (few-shot examples)

Include example conversation turns in your `AgentSpec` using the `shots` field:

```yaml
name: chat
prompt: "{prompt}"
model: "Qwen3-30B"
shots:
    - user: "What is AI?"
      assistant: "Artificial Intelligence..."
    - user: "How does it learn?"
      assistant: "Through training on data..."
```

These shots are prepended to the conversation history before each inference call.

<a href="javascript:openLink('/libraries/agent/supervision')">Next: supervised agent</a>