Core Concepts: Skills

What is a Skill?

A Skill represents a specific capability or task that an Agent can perform in Reactive Agents. While an Agent is like a specialized AI assistant, a Skill is a particular function within that assistant's expertise.

Examples:

• english-grammar-check - analyzes English text for grammar errors
• spanish-translation - translates text to Spanish
• code-debugging - identifies and fixes code issues
• customer-inquiry-routing - categorizes and routes customer questions

Key Properties

Every skill has:

Basic Parameters

Basic Parameters

• Name: A unique identifier within the agent (e.g., "english-grammar-check")
• Description: What the skill does and how Reactive Agents should optimize it
• Partition: Split the skill into multiple parts for optimization
• System Prompts: Number of AI system prompts to generate for optimization

Advanced Parameters

NOTE: These parameters are not recommended to be changed.

• Grouping Interval: How often to re-optimize based on new data
• Minimum Requests Per Arm Reflection Threshold: Minimum number of requests per arm reflection to trigger a system prompt regeneration

How Skills Work

With an agent created, you can select a skill now:

1. Click on the "Create Skill" tab:

2. Fill out the Skill Optimization Form and click "Create Skill":

You can create skills by providing:

• Skill Name: What you want to call this specific capability
• Description: Detailed explanation of what the skill should do (minimum 25 characters)
• Optimization Settings: How many configurations and prompts to test

Important: The description is crucial - Reactive Agents uses it to automatically generate system prompts and optimize the skill's performance.

Skill Optimization

Skills use advanced optimization techniques:

Multi-Armed Bandit Testing

• Arms: Different configurations (system prompt + AI model + parameters)
• Testing: Automatically tests multiple approaches simultaneously
• Selection: Gradually favors better-performing configurations

Grouping

• Groups Similar Requests: Organizes requests by patterns and complexity
• Specialized Optimization: Different groups can have different optimal configurations
• Adaptive Learning: Adjusts group boundaries as more data comes in

Continuous Improvement

• Performance Tracking: Measures success rates and quality metrics
• Auto-Generation: Creates new system prompts based on performance data
• Dynamic Adjustment: Modifies configurations based on real-world usage

Why Use Skills?

• Task Specialization: Each skill is optimized for a specific type of work
• Automatic Optimization: Performance improves over time without manual tuning
• Intelligent Routing: Best AI provider and configuration selected automatically
• Measurable Results: Track how well each skill performs
• Flexible Organization: Break complex agents into manageable, specialized skills

Best Practices

• Specific Descriptions: Write detailed descriptions of what each skill should accomplish
• Single Responsibility: Each skill should handle one clear, specific task
• Meaningful Names: Use descriptive names that explain the skill's purpose
• Let It Learn: Allow the optimization system time to learn and improve performance