Seedlings

Small context documents that give AI agents enough information to build their own skills.

A seedling is a tiny markdown file — a few API endpoints, a lookup table, a code snippet — just the stuff an agent can't figure out on its own. The agent reads it and grows the rest: working code, error handling, tests, everything.

Follows the Agent Skills spec.

How it works

You write this:

---
name: weather-fetcher
description: Get current weather and forecasts for any location using the free Open-Meteo API.
---

## API: Open-Meteo (free, no key)

Geocode: GET https://geocoding-api.open-meteo.com/v1/search?name={city}&count=1
Current: GET https://api.open-meteo.com/v1/forecast?latitude={lat}&longitude={lon}&current=temperature_2m,weather_code,wind_speed_10m

## Weather Codes (WMO)

0=Clear, 1-3=Cloudy, 45=Fog, 51-55=Drizzle, 61-65=Rain, 71-75=Snow, 95=Thunderstorm

The agent grows it into a full working skill.

Sprout CLI

sprout search "get the weather"        # find a seedling
sprout plant weather-fetcher           # install it as a local skill
sprout plant weather-fetcher -g        # install globally
sprout list                            # see what you've planted

Seedlings are installed as Agent Skills:

• Local: ./agents/skills/<name>/SKILL.md
• Global: ~/.agents/skills/<name>/SKILL.md

Structure

Each seedling is a folder with a SKILL.md and optional references:

seedlings/
├── weather-fetcher/
│   └── SKILL.md
├── csv-to-chart/
│   └── SKILL.md
├── stock-price/
│   └── SKILL.md
└── qr-code-generator/
    └── SKILL.md

Writing a seedling

A good seedling is as short as possible. Only include what an agent genuinely cannot infer:

Include	Skip
API endpoints and params	Error handling patterns
Auth method (key, OAuth, none)	Code structure
Lookup tables and code mappings	File I/O boilerplate
Library name and install command	Logging and validation
Response format quirks	Test scaffolding

Contributing

1. Create seedlings/your-skill/SKILL.md
2. Add frontmatter with name (must match folder name) and description
3. Write only what an agent can't infer
4. Open a PR