skill
toon-skill
Claude skill: compresses heavy AI-coding-agent prompts into lean, TOON-encoded, token-efficient specs
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/sheikhsajid69/toon-skill itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 17, 2026toon-skill
A Claude skill that compresses large, heavy prompts into lean, unambiguous specs for AI coding agents — encoding any structured data inside the prompt as TOON (Token-Oriented Object Notation) to cut token usage.
Triggered by starting a message with /toon, or by asking to compress/optimize/reduce the token cost of a prompt headed for a coding agent (Claude Code, Cursor, Copilot, Antigravity, OpenCode, Windsurf, etc.).
What it does
Heavy prompts are usually two things tangled together, and this skill treats them differently:
- Prose (instructions, context, requirements) is rewritten tighter — filler and hedging cut, exact file paths / constraints / done-criteria always preserved — so the result is still safe for an autonomous agent to act on without asking a follow-up question.
- Structured data (file lists, test cases, config, sample records) is re-encoded in TOON, which benchmarks ~30–60% fewer tokens than pretty-printed JSON on uniform, tabular data.
It also knows when not to use TOON — deeply nested or non-uniform data, or a handful of key-value pairs, are left as compact JSON, since TOON's savings only materialize on uniform tabular arrays.
See SKILL.md for the full workflow and references/ for the TOON syntax cheat sheet and the prompt-compression rules.
Install
Requires Python 3. The bundled converter (scripts/to_toon.py) prefers the official @toon-format/cli via npx (Node.js) when available, and falls back to a bundled pure-Python encoder otherwise.
git clone https://github.com/sheikhsajid69/toon-skill.git
cd toon-skill
./install.sh # personal install: ~/.claude/skills/toon — available in every project
./install.sh --project # this repo only: ./.claude/skills/toon — commit it to share with a team
Claude Code / Claude Desktop
install.sh drops the skill straight into the directory these tools already scan (~/.claude/skills/ or <project>/.claude/skills/). No further setup — the skill triggers automatically once the agent's task matches its description.
Claude.ai (web/desktop/mobile)
Zip the folder (SKILL.md, references/, scripts/) into a .skill file and upload it from Settings → Capabilities → Skills.
Other IDE agents (Cursor, Windsurf, Copilot Chat, etc.)
These tools don't read a skills/ folder natively. Point the agent's custom-instructions / rules file at the installed SKILL.md, e.g. in .cursor/rules or your agent's system-prompt config:
Follow the workflow in ~/.claude/skills/toon/SKILL.md whenever asked to
compress, optimize, or /toon a prompt.
Usage
/toon <paste your heavy prompt here>
or just ask directly: "toonify this prompt for my coding agent", "compress this so it costs fewer tokens".
The converter script can also be run standalone:
python3 scripts/to_toon.py input.json --stats
Repo structure
toon-skill/
├── SKILL.md # workflow the agent follows
├── references/
│ ├── toon-cheatsheet.md # TOON syntax + tabular-eligibility check
│ └── prompt-compression-guide.md # what to cut vs. never cut, worked example
├── scripts/
│ └── to_toon.py # JSON -> TOON converter (CLI-backed, Python fallback)
├── install.sh
├── LICENSE
└── SECURITY.md
Testing
Run the test suite:
pytest tests/