Creating Personas

Creating a persona on KillerSkills lets you package an agent's personality, behavior, and capabilities into a reusable, shareable format. This guide walks through the creation process step by step.

Creation Methods

1. Web Wizard (Recommended)

The easiest way to create a persona is through the web-based creation wizard at /create/persona.

The wizard has three steps:

1. Describe: Enter a natural language prompt describing the agent personality you want, or provide a URL to documentation that defines the persona
2. Edit: Review and refine the AI-generated PERSONA.md with real-time validation feedback
3. Publish: Configure visibility, pricing, category, and tags before publishing to the marketplace

2. AI Generation from Prompt

In the first step, describe the persona you want to create. The AI will generate a complete PERSONA.md file including system prompt, response style, guardrails, and bundled skills.

Example prompts:

"Create a senior DevOps engineer persona specializing in Kubernetes
and CI/CD pipelines. Should be technical and educational, with
guardrails around production deployments."

"Build a creative writing assistant persona that helps with blog posts,
documentation, and technical writing. Casual tone, focused on clarity
and storytelling."

"Design a security-focused code reviewer persona that checks for
vulnerabilities, follows OWASP guidelines, and explains security
concerns in detail."

3. Manual Creation

You can also write a PERSONA.md file manually following the format specification and paste it directly into the editor in step 2.

AI Generation Rate Limits

Persona generation uses the same AI backend and rate limits as skill generation:

Rate limits reset daily at UTC midnight. If you bring your own API key (BYOK) in settings, rate limits are bypassed entirely.

Bundling Skills

One of the most powerful features of personas is the ability to bundle existing skills. This lets you compose specialized agent personalities by combining complementary capabilities.

How to Reference Skills

In the PERSONA.md frontmatter, use the skills array to list skill slugs:

---
name: Full-Stack Developer
skills:
  - react-component-generator
  - typescript-type-definitions
  - node-express-routes
  - postgresql-migrations
  - jest-testing-patterns
  - docker-compose-setup
---

When the persona is activated, all referenced skills are automatically loaded, giving the agent access to those capabilities.

Finding Skills to Bundle

Browse the marketplace to discover skills that complement your persona. Look for skills that:

• Align with your persona's expertise area
• Work well together (complementary, not overlapping)
• Match the complexity level you're targeting
• Are verified and highly rated

Skill Composition Best Practices

• Start small: Begin with 3-5 core skills rather than bundling everything
• Avoid overlap: Don't bundle skills that do the same thing
• Match the workflow: Bundle skills that are commonly used together
• Document the bundle: Explain in the Usage Guide how the skills work together

Choosing Response Style

The responseStyle field shapes how your persona communicates. Choose based on your target audience and use case:

Writing Effective Guardrails

Guardrails are behavioral constraints that ensure your persona acts safely and predictably. Good guardrails are:

• Specific: "Always run tests before committing" not "be careful with code"
• Actionable: Clear instructions the agent can follow
• Safety-focused: Prevent destructive or dangerous operations
• Context-appropriate: Relevant to the persona's domain

Example Guardrails by Domain

Development Personas

guardrails:
  - Run tests before committing code
  - Never commit directly to main branch
  - Always include error handling in new code
  - Explain complex algorithms before implementing
  - Request code review for breaking changes

DevOps Personas

guardrails:
  - Never modify production without approval
  - Always backup before schema migrations
  - Validate Kubernetes manifests before applying
  - Test in staging environment first
  - Never expose secrets in logs or outputs

Data Science Personas

guardrails:
  - Validate data quality before analysis
  - Check for bias in training data
  - Never overwrite raw data files
  - Document data transformations
  - Verify results with cross-validation

Security Personas

guardrails:
  - Never execute unverified code
  - Validate all user input
  - Use parameterized queries only
  - Never log sensitive information
  - Follow OWASP security guidelines

Writing a Strong System Prompt

The system prompt is the core of your persona. It defines the agent's identity, expertise, and approach. A good system prompt:

1. Establishes identity: "You are a senior DevOps engineer with 10+ years of experience..."
2. Lists expertise areas: "You specialize in Kubernetes, Docker, CI/CD, and cloud infrastructure..."
3. Defines approach: "You prioritize security, automation, and maintainability..."
4. Provides examples: Show how to handle common scenarios
5. Sets expectations: "Always explain your reasoning before taking actions"

System Prompt Template

# [Persona Name]

You are a [role] with [experience level] specializing in [domains].
Your expertise includes:

- [Expertise area 1]
- [Expertise area 2]
- [Expertise area 3]

## Approach

When working on [domain] tasks:

1. **[Aspect 1]**: [How to handle it]
2. **[Aspect 2]**: [How to handle it]
3. **[Aspect 3]**: [How to handle it]

## Communication

- [Communication guideline 1]
- [Communication guideline 2]
- [Communication guideline 3]

Testing Your Persona

Before publishing, test your persona thoroughly:

1. Try common tasks: Test the persona with typical use cases
2. Check guardrails: Verify behavioral constraints are enforced
3. Verify tone: Ensure response style matches your intent
4. Test bundled skills: Confirm all referenced skills work together
5. Edge cases: Try unusual or challenging requests

Publishing Options

In the final step of the wizard, configure:

• Visibility: Public (marketplace) or Private (personal use)
• Pricing: Free, one-time purchase, or subscription
• Category: Primary classification for discovery
• Tags: Keywords for search (max 10)

Public personas are discoverable in the marketplace. Private personas are only visible to you and users you share them with.

Tips for Great Personas

1. Clear Identity

Give your persona a specific role and expertise area. "Frontend React Developer" is better than "Developer".

2. Specific Guardrails

Be explicit about constraints. "Never delete files without confirmation" is better than "be careful".

3. Choose Complementary Skills

Bundle skills that work together in a typical workflow. A testing persona might bundle unit testing, integration testing, and test data generation skills.

4. Match Response Style to Audience

Educational for teaching, concise for experienced users, technical for detailed work.

5. Include Usage Examples

Show users how to interact with your persona. Provide sample prompts and expected workflows in the Usage Guide section.

6. Iterate Based on Feedback

Pay attention to reviews and ratings. Update your persona to address common issues or requests.

Example: Building a Testing Specialist

Let's walk through creating a testing-focused persona from start to finish.

Step 1: Describe the Persona

"Create a QA automation engineer persona specializing in test-driven
development and automated testing. Should be educational, helping users
learn best practices. Include guardrails around test coverage and code
quality."

Step 2: Review Generated Content

The AI generates a PERSONA.md file. Review and refine:

• Ensure system prompt captures testing philosophy
• Verify guardrails include test coverage requirements
• Add bundled skills: pytest-patterns, jest-mocking, test-data-builders, coverage-reporting
• Set response style to "educational"

Step 3: Configure Publishing

• Category: testing
• Tags: tdd, automation, pytest, jest, qa
• Visibility: public
• Pricing: free

Next Steps

• Create your first persona
• Review the format specification
• Browse existing personas for inspiration