TOON

Optimize token usage by adopting the compact TOON format for data storage and context.

TOON Adoption Skill

Description

Efficiently parse, generate, and store data using Token-Oriented Object Notation (TOON). TOON is designed for LLMs as a lossless, drop-in representation of JSON data that reduces token usage by approximately 40% through indentation, minimal quoting, and tabular layouts.

TOON Syntax Rules

1. Indentation-based Nesting

• Use 2-space indentation to define hierarchy, similar to YAML.
• Avoid curly braces {} and square brackets [] for nesting unless defining a tabular schema.

2. Minimal Quoting

• Keys and values do not require quotes unless they contain commas or significant leading/trailing whitespace.

3. Explicit Array Lengths

• Declare the number of items in an array using the [N] notation (e.g., friends[3]).

4. Tabular Arrays (Arrays of Objects)

• For uniform arrays of objects, use the format: key[N]{field1,field2,...}:.
• List the field names once in the header, followed by rows of values separated by commas.

5. Encoding

• Always use UTF-8.

Guidelines for Agent Use

1. Storage: Prioritize .toon files for logs, schedules, and long-term data tracking.
2. Context Compression: When summarizing history or notes, use TOON to fit more information into the context window.
3. Parsing: Interpret indentation as nested object levels and CSV rows as objects matching the header schema.

Example Document

metadata:
  name: Sample Configuration
  format: TOON
  efficiency_gain: 0.4
goals[3]{id,title,category}:
  1,Lose weight,health
  2,Increase self-esteem,personal
  3,Get out of loneliness,social