KillerSkills
Sign in
Documentation

Personas Overview

Personas are reusable agent personalities that bundle a system prompt, response style, behavioral guardrails, and a curated set of skills into a single shareable package.

What Are Personas?

While skills define what an agent can do (specific capabilities and tools), personas define who the agent is — its personality, tone, constraints, and default behaviors.

A persona combines:

  • System Prompt: Core instructions that shape the agent's identity and behavior
  • Response Style: How the agent communicates (technical, casual, formal, concise, verbose, educational, creative)
  • Guardrails: Behavioral constraints and safety rules (e.g., "never execute destructive commands", "always explain before acting")
  • Bundled Skills: A curated set of skills referenced by slug that define the agent's capabilities

PERSONA.md Format

Personas are defined using the PERSONA.md format, which consists of YAML frontmatter followed by markdown content. The frontmatter specifies metadata, response style, guardrails, and bundled skills, while the body contains the system prompt, skill configuration details, and usage guidance.

---
name: Python Dev Assistant
description: Expert Python developer with testing focus
version: "1.0.0"
author: KillerSkills Team
defaultRuntime: claude-code
runtimes: [claude-code, cursor, copilot]
category: development
tags: [python, testing, debugging]
responseStyle: technical
guardrails:
  - Always run tests before committing code
  - Explain complex algorithms step-by-step
  - Never execute rm -rf or destructive commands
skills:
  - python-testing-pytest
  - python-linting-ruff
  - git-commit-conventions
pricing:
  type: free
---

# Python Dev Assistant

You are an expert Python developer specializing in test-driven
development and code quality...

Response Styles

The responseStyle field shapes how the agent communicates:

  • technical: Precise, detailed explanations with technical terminology
  • casual: Friendly, conversational tone
  • formal: Professional, structured communication
  • concise: Brief, to-the-point responses
  • verbose: Detailed explanations with context
  • educational: Teaching-focused with examples
  • creative: Innovative solutions and approaches

Guardrails

Guardrails are behavioral constraints that ensure safe, predictable agent behavior. Common examples include:

  • Never execute destructive commands without confirmation
  • Always explain reasoning before taking actions
  • Require approval before making API calls
  • Never commit changes without running tests first
  • Only access files within the project directory
  • Ask for clarification when requirements are ambiguous

Bundling Skills

Personas can reference existing skills by their slug. This allows you to compose specialized agent personalities by combining complementary capabilities:

skills:
  - react-component-generator
  - typescript-type-definitions
  - jest-testing-patterns
  - git-conventional-commits

The bundled skills are automatically loaded when the persona is activated, giving the agent access to all referenced capabilities.

Learn More