Guides

Skills & Templates

Understand the difference between templates and skills, and how to use them to build better AI agents.

| View as Markdown
1 min read

What are Templates?

Templates are full AI agent blueprints — pre-built configurations that include a system prompt, model selection, suggested messages, and sometimes required integrations. When you use a template, it creates a new app in your workspace with everything pre-configured.

Templates are great for:

  • Getting started quickly with a proven use case
  • Learning best practices for specific agent types
  • Launching a working agent in minutes instead of hours

What are Skills?

Skills are composable prompt chunks — small pieces of a system prompt that introduce specific behaviors into your AI agent. Unlike templates, skills don’t create new apps. Instead, you add them to existing apps to enhance their capabilities.

Think of skills like plugins: each one teaches your agent a new behavior.

Examples of skills:

  • Chain of Thought — makes the agent reason step by step
  • Socratic Tutor — guides users to answers through questions
  • Anti-Hallucination — prevents the agent from making things up
  • Ultra-Concise — keeps responses short and direct

How Skills Work

Skills are stored separately from your system prompt and composed at runtime. This means:

  1. You can add and remove skills without editing your system prompt
  2. Skills stack — add as many as you want, they all apply
  3. Skills are visible in the system prompt editor under the “Skills” section
  4. The agent sees the skill instructions alongside your system prompt during every conversation

Personalities

Personalities are a special category of skills that define how your AI communicates — its voice, tone, and character. Examples include “Patient Mentor”, “Corporate Strategist”, and “Sardonic Realist”.

Pick one personality as a foundation, then stack behavioral skills on top.

Quality Scores

Some templates include a Quality Score — an automated evaluation showing how well the template performs across test scenarios. The score is generated by running the agent through realistic test cases and having an AI judge evaluate the responses for accuracy and relevance.

A score of 90%+ means the template consistently produces high-quality responses for its intended use case.