---
name: datoon
description: Smart TOON conversion workflow for structured data in Claude Code. Converts JSON-like payloads to TOON only when structure is suitable and token savings are meaningful.
---

# datoon

## Core Rule

Before sending structured payloads to the model:

1. Detect whether payload is a good TOON candidate.
2. Convert only when expected savings are meaningful.
3. Keep JSON when conversion likely hurts clarity or efficiency.

## Trigger Phrases

- `/datoon`
- `datoon mode`
- `convert to TOON`
- `optimize tokens for this JSON`
- `use TOON for this dataset`

## Decision Policy

Convert when all conditions pass:

- payload has at least one uniform object array (table-like records);
- nesting depth is moderate;
- estimated savings are above threshold (default 15%).

Skip conversion when any condition fails:

- deeply nested or highly non-uniform structures;
- tiny payloads with negligible savings;
- free-form text tasks without structured data.

## Execution Workflow

1. Normalize input to valid JSON.
2. Run:
   - `echo '<json>' | datoon --report-stdout`
   - or `datoon input.json -o output.toon --report report.json`
3. Read report:
   - if `decision == "convert"`, pass TOON to model;
   - if `decision == "skip"`, pass normalized JSON.
4. Keep report metadata for observability.

## Reliability Rules

- Never change semantic values or key names.
- Fail loudly for invalid JSON.
- If TOON CLI dependency is unavailable, keep JSON and report fallback reason.
- Use `--force` only for controlled experiments.

## Boundaries

- This skill is for structured data payload optimization, not general prose rewriting.
- For sensitive data, preserve exact fields and avoid unnecessary reformatting outside conversion path.