Prompt Design

Prompt Design Skill

Bonded to: prompt-fundamentals-agent

Quick Start

Skill("custom-plugin-prompt-engineering:prompt-design")

Parameter Schema

parameters:
  task_type:
    type: enum
    values: [generation, classification, extraction, transformation, analysis]
    required: true

  output_format:
    type: enum
    values: [text, json, markdown, code, structured]
    default: text

  constraints:
    type: object
    properties:
      max_length: number
      tone: string
      language: string

Core Patterns

1. ROLE-CONTEXT-TASK-FORMAT (RCTF)

Best for: General-purpose prompts

You are a [ROLE] with expertise in [DOMAIN].

## Context
[RELEVANT BACKGROUND INFORMATION]

## Task
[SPECIFIC REQUEST WITH CLEAR OBJECTIVE]

## Output Format
[STRUCTURE AND FORMAT REQUIREMENTS]

## Constraints
[LIMITATIONS AND GUARDRAILS]

Example:

You are a senior software architect with expertise in distributed systems.

## Context
We're designing a new microservices architecture for an e-commerce platform
that needs to handle 10,000 concurrent users.

## Task
Design the service boundaries and communication patterns for the checkout flow.

## Output Format
- Service diagram (ASCII)
- API contracts (OpenAPI style)
- Data flow description

## Constraints
- Use event-driven architecture where possible
- Minimize synchronous calls
- Consider eventual consistency

2. INSTRUCTION-INPUT-OUTPUT (IIO)

Best for: Data processing and transformation

## Instruction
[WHAT TO DO WITH THE INPUT]

## Input
[DATA TO PROCESS]

## Expected Output
[FORMAT AND STRUCTURE OF RESULT]

3. PERSONA-SCENARIO-GOAL (PSG)

Best for: Creative and roleplay tasks

## Persona
[WHO YOU ARE - background, expertise, personality]

## Scenario
[SITUATION AND CONTEXT]

## Goal
[WHAT TO ACHIEVE]

4. CONSTRAINT-FIRST (CF)

Best for: Safety-critical applications

## RULES (Must follow absolutely)
1. [CRITICAL CONSTRAINT 1]
2. [CRITICAL CONSTRAINT 2]

## Your Role
[ROLE DEFINITION]

## Task
[WHAT TO DO WITHIN CONSTRAINTS]

Pattern Selection Guide

Anti-Patterns to Avoid

anti_patterns:
  vague_instructions:
    bad: "Make it better"
    good: "Improve readability by using shorter sentences (max 20 words)"

  no_output_format:
    bad: "Analyze this code"
    good: "Analyze this code. For each issue found, provide: issue, severity, fix"

  conflicting_instructions:
    bad: "Be concise. Explain everything in detail."
    good: "Be concise but complete - include all critical details"

  missing_context:
    bad: "Fix the bug"
    good: "Fix the null pointer exception in the user authentication module"

Validation Checklist

validation:
  structure:
    - [ ] Clear role/persona defined
    - [ ] Task objective is specific and measurable
    - [ ] Output format is explicit
    - [ ] Constraints are stated upfront

  clarity:
    - [ ] No ambiguous terms
    - [ ] One instruction per line
    - [ ] Action verbs used
    - [ ] Examples provided if complex

  completeness:
    - [ ] All necessary context included
    - [ ] Edge cases addressed
    - [ ] Error handling specified

Troubleshooting

Integration

integrates_with:
  - prompt-templates: Provides reusable patterns
  - prompt-evaluation: Tests pattern effectiveness
  - chain-of-thought: Adds reasoning capability

usage_example: |
  # Combine with few-shot for complex tasks
  [RCTF Pattern]
  +
  [2-3 examples]
  +
  [Current input]

References

See references/GUIDE.md for detailed methodology. See assets/config.yaml for configuration options.