A collection of behavioral blueprints and skills for agents.
These are just recipes. They are not part of the SOUP protocol or infrastructure; they are standardized behavioral templates that agents can "install" to achieve specific outcomes.
A Recipe is typically a skill.md file that an agent reads to adopt a specific coordination pattern.
Each recipe includes a front-matter block with:
-
author: The creator of the skill. -
summary: A high-level overview for humans. -
description: The critical "Semantic Trigger" for the agent. It defines when and how the agent should invoke this specific skill. -
out_of_scope: The "Safety Barrier." Explicitly defines what the agent should not do, preventing scope creep and hallucinations. -
version: The current iteration of the logic. -
Decoupled: The soup engine doesn't care how you coordinate, but these recipes ensure you do it well.
-
Functional: From "Graph Integrity" to "Project Management," these are the blueprints for the agentic era.
-
Community Driven: Standardized ways of working that anyone can contribute to.
This is a placeholder for the upcoming community library of skill.md blueprints. Later, these recipes will live as the manifest of each SOUP workspace.
There are multiple ways to "install" a recipe into an agent's brain. Each has pros and cons depending on your workflow.
Open a skill.md file, copy its entire content, and paste it into your agent's chat window (ChatGPT, Claude, etc.).
- Instructions: "Adopt this skill and wait for my next command."
- Pros: Instant, works with any LLM, zero setup required.
- Cons: Volatile (lost when the session ends), consumes context window, manual effort.
Place the skill.md file into your agent's shared workspace or project directory.
- Instructions: "Search for available skills in this workspace and adopt the relevant ones for this task."
- Pros: Persistent & Shared. All agents working in the same environment automatically adopt the same protocols. No manual copy-pasting required for new sessions.
- Cons: Requires an agent capable of reading local or remote files.
Upload the skill.md to your agent's "Knowledge" files or paste it directly into its System Instructions.
- Pros: The agent is the skill. Low overhead per session, permanently available.
- Cons: Harder to version-control or update across multiple agents. Locked to a specific platform.
Follow the recipe. Reach the Purpose.
Maintained by heysoup.co — Secure shared context for the agentic era.