Hugin Guide

Hugin Agent Creation Guide

This skill helps you create Hugin AI agents. It provides an overview and directs you to detailed reference documentation.

How to Use This Skill

1. Quick Start: Follow the minimal agent example below
2. Detailed Info: Read the appropriate reference file for schemas and examples
3. Scaffolding: Use /hugin-agent-creator:hugin-scaffold to generate starter files

Reference Files (Read These for Details)

When the user needs detailed information about a specific topic, read the appropriate reference file from this plugin directory:

Important: These reference files contain detailed schemas, field descriptions, and complete examples. Read them when answering specific questions.

Starter Templates

When the user wants to create files, read and adapt the starter templates from this plugin:

Quick Start: Minimal Agent (3 Files)

A working agent needs just 3 files:

my_agent/
├── configs/my_agent.yaml      # Agent configuration
├── tasks/my_task.yaml         # Task definition
└── templates/my_system.yaml   # System prompt template

Run Command

uv run hugin run --task my_task --task-path ./my_agent

Core Concepts (Overview)

Config

Defines an agent's identity: model, system template, available tools.

• Read references/config-reference.md for full schema

Task

Defines what an agent should do: prompt, parameters, optional pipeline.

• Read references/task-reference.md for full schema

Template

System prompt that sets agent behavior and personality.

• Read references/template-reference.md for examples

Tool

Extends agent capabilities with Python code.

• Read references/tool-reference.md for implementation guide

Built-in Tools

Reference with builtins.<tool>:<alias>:

Decision Tree: Which Pattern?

Do you need custom tools?
├── No → Minimal pattern
└── Yes → Tool pattern
    └── Does the tool spawn another agent?
        ├── No → Simple tool
        └── Yes → AgentCall pattern

Do you need human input during execution?
├── No → Set interactive: false
└── Yes → AskHuman pattern (interactive: true)

Do you need multiple processing stages?
├── No → Single task
└── Yes → Task sequence pattern (task_sequence + pass_result_as)

Do you need multiple agents running together?
└── Yes → Multi-agent pattern (shared state via env_vars)

For detailed pattern examples, read references/patterns.md

Common Patterns (Summary)

1. Minimal Agent - Config + task + template, built-in tools only
2. Tool Agent - Custom tools for specific capabilities
3. Pipeline Agent - Multi-stage processing with result passing
4. Human-in-the-Loop - Requires human approval during execution
5. Agent Delegation - Spawns specialized sub-agents
6. Multi-Agent - Multiple agents sharing state

Workflow for Helping Users

1. Understand what they want to build - Ask clarifying questions if needed
2. Recommend a pattern - Use the decision tree above
3. Read the relevant reference - Get detailed schema information
4. Read the starter template - Use as base for their files
5. Customize for their use case - Replace placeholders, add their logic
6. Provide the run command - Show how to test the agent

Converted and distributed by TomeVault — claim your Tome and manage your conversions.