Filter Functions

AgentMark provides a set of built-in filters that you can use to manipulate and transform data within your prompts. Filters are functions that take an input value and return a transformed output.

​

Basic Usage

Use filters by calling them on values in your prompts:

---
name: greeting
metadata:
  model:
    name: gpt-4
test_settings:
  props:
    userName: "alice"
    items: ["apple", "banana", "cherry"]
---

<System>
Your name is {capitalize(props.userName)}!
Your items are: {join(props.items, ", ")}
</System>
<User>
What is your name, and what items do you have?
</User>

​

Built-in Filters

AgentMark includes several built-in filters:

​

Text Manipulation

• capitalize(str): Capitalizes first character of string
• lower(str): Converts string to lowercase
• upper(str): Converts string to uppercase
• truncate(str, length): Truncates string to specified length
• replace(str, search, replace): Replaces text in string

​

Array Operations

• join(array, separator): Joins array elements with separator

​

Number Formatting

• abs(num): Returns absolute value
• round(num, decimals): Rounds number to decimal places

​

Data Transformation

• dump(obj): Converts object to JSON string
• urlencode(str): URL encodes a string

​

Creating Custom Filters

You can create and register custom filters:

import { FilterRegistry } from "@puzzlet/agentmark";

// Create filter function
const reverse = (input: string): string => {
  return input.split('').reverse().join('');
};

// Register the filter
FilterRegistry.register("reverse", reverse);

Then use in your prompts:

---
name: reverse-text
metadata:
  model:
    name: gpt-4
test_settings:
  props:
    text: "Hello World"
---

<System>
Original: {props.text}
Reversed: {reverse(props.text)}
</System>

​

Best Practices

1. Use descriptive filter names
2. Handle invalid input types gracefully
3. Keep filters focused on single transformations