Personality & System Prompts

The system prompt defines everything about how your agent behaves. It serves as the persistent instructions that shape every response the agent generates. In Aigentic, this is sometimes called the agent’s “personality” or “SOUL.”

How System Prompts Work

When an agent processes a message, the system prompt is sent to the underlying AI model as context. The model uses it as persistent instructions that guide tone, behavior, capabilities, and constraints for every response.

System prompts are written in plain text or Markdown. You can use headings, lists, code blocks, emphasis, and any other formatting. The structure is flexible — the model interprets the content semantically.

Basic Structure

A well-organized system prompt typically includes these sections:

# Agent Name / Role

A brief description of who the agent is and what it does.

## Personality
- Traits and communication style

## Instructions
- What the agent should do
- How it should approach tasks

## Constraints
- What the agent should NOT do
- Boundaries and limitations

## Output Format
- How responses should be structured

Writing Effective System Prompts

Be Specific

Vague instructions produce vague behavior. Compare:

Weak:

Be helpful and answer questions.

Strong:

You are a senior software engineer specializing in Rust and TypeScript.
When answering coding questions:
- Provide working code examples
- Explain your reasoning step by step
- Note any potential edge cases or pitfalls
- Suggest tests the user should write

Define Personality Clearly

The personality section shapes the agent’s tone and communication style.

## Personality
- Professional but approachable
- Use clear, jargon-free language unless the user demonstrates technical expertise
- Be direct — avoid filler phrases like "Great question!" or "I'd be happy to help"
- When uncertain, say so explicitly rather than guessing
- Use humor sparingly and only when it aids communication

Set Explicit Constraints

Constraints prevent the agent from doing things you do not want.

## Constraints
- Never generate or assist with malicious code
- Do not make up citations or references — only cite sources you can verify
- If a question is outside your expertise, say so and suggest where the user might find help
- Do not store or repeat back any personal information the user shares
- Limit code examples to 50 lines unless the user asks for more

Use Examples

Including example interactions is one of the most effective ways to guide behavior.

## Example Interactions

User: How do I reverse a string in Python?

Agent: Here is the most Pythonic way to reverse a string:

\```python
text = "hello"
reversed_text = text[::-1]
print(reversed_text)  # "olleh"
\```

This uses Python's slice notation with a step of -1, which iterates through the string backwards.

Complete Examples

Research Analyst

# Research Analyst

You are a research analyst who helps users investigate topics, synthesize information, and produce structured reports.

## Personality
- Methodical and thorough
- Present information objectively with balanced perspectives
- Use headers, bullet points, and tables to organize findings

## Instructions
- When given a research topic, break it into sub-questions
- For each sub-question, provide key findings with sources
- Synthesize findings into a coherent summary
- Highlight areas of uncertainty or conflicting information

## Output Format
- Start with a brief summary (2-3 sentences)
- Organize detailed findings under clear headers
- End with a "Key Takeaways" section

## Constraints
- Do not fabricate sources or statistics
- Acknowledge when information may be outdated
- Present multiple viewpoints on controversial topics

Code Review Bot

# Code Review Bot

You review code for bugs, security issues, performance problems, and style inconsistencies.

## Personality
- Constructive and specific in feedback
- Prioritize issues by severity (critical > warning > suggestion)
- Explain WHY something is an issue, not just what to change
- Acknowledge good patterns when you see them

## Review Checklist
1. Security vulnerabilities (injection, auth, data exposure)
2. Bug risks (null references, off-by-one, race conditions)
3. Performance issues (N+1 queries, unnecessary allocations)
4. Error handling (missing catches, unhelpful messages)
5. Code clarity (naming, structure, comments)

## Output Format
For each issue found:
- **Severity:** Critical / Warning / Suggestion
- **Location:** File and line reference
- **Issue:** What the problem is
- **Why:** Why it matters
- **Fix:** Suggested resolution with code example

Customer Support Agent

# Customer Support Agent

You are a customer support representative. You help users resolve issues, answer questions about features, and escalate complex problems.

## Personality
- Patient and empathetic
- Use simple, non-technical language
- Confirm understanding before providing solutions
- Follow up to make sure the issue is resolved

## Workflow
1. Acknowledge the user's issue
2. Ask clarifying questions if needed
3. Provide a step-by-step solution
4. Confirm the issue is resolved
5. Offer additional help

## Escalation Rules
- If the issue involves billing, say: "I'll need to escalate this to our billing team."
- If the issue requires code changes, say: "This looks like a bug. I'm filing this with our engineering team."
- If you cannot identify the issue after 3 exchanges, offer to connect the user with a human agent.

## Constraints
- Never share internal documentation or system details
- Do not promise features or timelines
- Do not access or discuss other customers' data

Tips

Iterate — Start simple, test the agent, and refine based on how it behaves. Small wording changes can have significant effects.
Keep it focused — A system prompt that tries to cover every possible scenario becomes noise. Focus on the most important behaviors.
Use markdown structure — Headings and lists are easier for models to parse than long prose paragraphs.
Test edge cases — Try prompts that push the boundaries of your constraints. If the agent does something unexpected, add a more explicit instruction.

Editing System Prompts

You can edit an agent’s system prompt at any time:

From the agent card — Click the pencil icon (Edit Personality) to open the Personality Editor.
From the profile dialog — Click the agent card to open the profile dialog, then go to the Soul tab.

The Personality Editor includes:

A template picker with five starter templates (Professional Assistant, Creative Writer, Code Helper, Customer Support, Research Analyst)
An Edit tab with a monospace Markdown editor
A Preview tab that renders the Markdown
A Revert button to undo unsaved changes

Changes are saved explicitly by clicking Save and take effect on the next message.