devharness: Command for Claude Code

TOON (Token-Oriented Object Notation)

A lightweight data format designed to minimize token usage when passing structured data to LLMs.

Syntax Reference

Type	Syntax	Example
Key-value	key:value;	name:Luna;age:3
Spaces in values	(...)	title:(Chief Snack Manager)
Arrays (unordered)	[item|item]	pets:[cat|dog|ferret]
Lists (ordered)	<item|item>	shopping:<milk|eggs|bread>
Nested objects	{...}	user:{name:Luna;stats:{speed:9;stealth:10}}

Rules

1. Key-value pairs separated by semicolons
2. No quotes needed for simple strings
3. Use parentheses () for values containing spaces or special characters
4. Arrays [] for unordered collections, Lists <> for ordered sequences
5. Pipe | separates items in arrays/lists
6. Curly braces {} for nested structures

Examples

Simple object:

name:Luna;age:3;color:silver

With spaces:

name:Luna;title:(Chief Snack Manager);active:true

With array:

name:Luna;pets:[cat|dog|ferret];active:true

Nested:

user:{name:Luna;stats:{speed:9;stealth:10}};timestamp:123456

When to Use

• Data passed directly into LLM prompts
• Internal tool outputs within AI workflows
• RAG systems and metadata encoding
• Bulk classification tasks
• Scenarios prioritizing compactness over schema validation

When NOT to Use

• Public APIs (use JSON)
• External systems requiring schema validation
• Anywhere standardized formats are required

Token Efficiency

~58% reduction compared to JSON (26 tokens → 11 tokens for equivalent data).

---

Use TOON format when the user requests token-efficient data representation or when outputting structured data that will be consumed by LLMs.