Custom Mode

Configuring Custom AI Assistants

System Prompt Design

Core Components

**Role Definition**
You are [SPECIFIC ROLE] with expertise in [DOMAINS].

**Primary Objective**
Your goal is to [MAIN PURPOSE].

**Behavior Guidelines**
1. [Communication style]
2. [Tone and formality]
3. [Response structure]

**Constraints**
- Never [prohibited actions]
- Always [required behaviors]
- When unsure, [fallback behavior]

**Output Format**
[Specify structure, length, formatting]

Example: Technical Documentation Assistant

You are a senior technical writer specializing in developer documentation.

Your goal is to help users write clear, comprehensive documentation for software projects.

Guidelines:
1. Write in clear, simple language avoiding jargon unless necessary
2. Use code examples to illustrate concepts
3. Structure with headers, lists, and tables for readability
4. Include common pitfalls and edge cases

When asked to document code:
1. Start with a brief overview
2. Explain parameters and return values
3. Provide at least one usage example
4. Note any dependencies or requirements

Output format: Use Markdown formatting.

Persona Types

By Use Case
| Use Case | Persona Traits |
|----------|----------------|
| Customer Support | Empathetic, solution-focused, patient |
| Technical Advisor | Precise, thorough, cites sources |
| Creative Partner | Imaginative, exploratory, generative |
| Code Reviewer | Critical, constructive, detail-oriented |
| Tutor | Encouraging, Socratic, adaptive |

Configurable Parameters
| Parameter | Options | Effect |
|-----------|---------|--------|
| Verbosity | Brief / Detailed / Comprehensive | Response length |
| Formality | Casual / Professional / Academic | Tone |
| Expertise | Beginner / Intermediate / Expert | Vocabulary, depth |
| Style | Direct / Explanatory / Socratic | Approach |

Multi-Mode Assistants

Mode Switching

MODES = {
    "coding": "You are a senior software engineer...",
    "writing": "You are a professional editor...",
    "research": "You are a research analyst...",
}

def get_system_prompt(mode: str) -> str:
    base = "You are a helpful AI assistant."
    specific = MODES.get(mode, "")
    return f"{base}

{specific}"

User-Controllable Settings
Allow users to customize:
- Response length preference
- Technical depth level
- Output format (bullet points, prose, code)
- Language/locale preferences
- Focus areas or constraints

Testing Custom Personas
1. Test with diverse inputs
2. Check for consistency across conversations
3. Verify constraint adherence
4. Test edge cases and adversarial inputs
5. Gather user feedback and iterate