How to Write Great Agent System Prompts
A practical guide to crafting system prompts for on-device AI agents, based on the three built-in agents in Perspective Intelligence.
Why System Prompts Matter
A system prompt is the difference between a generic AI and a purpose-built agent. It defines:
- Whothe agent is (identity and expertise)
- Howthe agent communicates (tone and style)
- Whatthe agent does and doesn't do (scope and boundaries)
The three built-in agents in Perspective Intelligence each demonstrate a distinct prompting strategy. This guide breaks down what makes each one work, extracts reusable patterns, and gives you templates to build your own.
Quick Comparison
| Agent | Role | Temperature | Tools | Prompt Strategy |
|---|---|---|---|---|
| @guide | App expert | 0.8 | All enabled | Knowledge list + behavioral directive |
| @writer | Creative assistant | 1.0 | All disabled | Role + capability list + style guidance |
| @coder | Software engineer | 0.4 | Web search only | Role + methodology + values |
What's in This Guide
Anatomy of a System Prompt
The core principles — identity, capability, and behavior — that make prompts effective.
The Guide Agent
Knowledge-grounded prompts with structured domain expertise for product experts.
The Writer Agent
Creative prompts that unlock expressive, varied output for artistic work.
The Coder Agent
Precision prompts for technical accuracy and structured reasoning.
Beyond the Prompt
How temperature, tool access, and model selection shape agent behavior.
Templates & Patterns
Starter templates for common agent types — research, planning, coaching, and more.
Key Takeaway
There is no universal “best” system prompt. The right prompt depends on what you need the agent to do. A creative agent needs freedom; a coding agent needs precision; a knowledge agent needs grounding. The prompt, temperature, and tool configuration work together as a system.