Prompt Perfect — System Prompt Engineering Templates

What it is

Prompt Perfect is a collection of system prompt engineering templates designed for building LLM personas, agents, and workflows. It provides structured formats for defining roles, capabilities, constraints, and output formatting. Instead of writing system prompts from scratch, you start with a proven template and customize it.

This resource targets developers building LLM-powered applications who need consistent, reliable system prompts. It covers common patterns: customer support agents, code reviewers, content writers, data analysts, and task-specific assistants.

How it saves time or tokens

A well-structured system prompt reduces the need for follow-up corrections. Prompt Perfect templates encode best practices: clear role definition, explicit constraints, output format specification, and edge case handling. The estimated token cost is around 2,800 tokens per template, but this investment eliminates iterative prompt refinement that can consume thousands of tokens.

The templates also include anti-patterns to avoid, helping you skip common mistakes.

How to use

1. Copy the base template and customize:

# IDENTITY AND PURPOSE
You are a [ROLE] specializing in [DOMAIN].
Your audience is [TARGET USERS].

# CAPABILITIES
- You CAN: [list specific abilities]
- You CANNOT: [list restrictions]

# OUTPUT FORMAT
- Format: [markdown/json/plain text]
- Length: [specific range]
- Tone: [professional/casual/technical]

# CONSTRAINTS
- Never [specific prohibition]
- Always [required behavior]
- When uncertain, [fallback behavior]

1. Add domain-specific rules:

# EXAMPLES
User: [sample input]
Assistant: [expected output]

# EDGE CASES
- If asked about [topic], respond with [approach]
- If input is ambiguous, ask for clarification

1. Test the prompt with representative inputs and iterate on constraints.

Example

A complete system prompt for a code review agent:

# IDENTITY
You are a senior software engineer performing code reviews.

# REVIEW CRITERIA
1. Correctness: Does the code do what it claims?
2. Security: Are there injection, auth, or data exposure risks?
3. Performance: Are there O(n^2) loops or N+1 queries?
4. Readability: Can a new team member understand this?

# OUTPUT FORMAT
For each issue found:
- Severity: critical / warning / suggestion
- Line: [file:line]
- Issue: [one sentence]
- Fix: [code snippet]

# CONSTRAINTS
- Do not rewrite the entire file
- Do not comment on style unless it impacts readability
- Limit to top 5 issues per review

Related on TokRepo

• Prompt library -- Curated prompts and templates for AI workflows
• AI tools for coding -- AI coding assistants and configuration

Common pitfalls

• Vague role definitions. 'You are a helpful assistant' gives the LLM no useful constraints. Be specific: 'You are a PostgreSQL DBA who reviews slow queries.'
• Missing output format specification. Without explicit format rules, LLM output varies unpredictably. Specify format, length, and structure.
• Overloading a single prompt with too many roles. If your prompt tries to be a coder, reviewer, and project manager simultaneously, quality drops. Create separate prompts for distinct tasks.