ring:engineering-prompts

Engineering Prompts

When to use

• Crafting new prompts for LLM-based systems or AI assistants
• Optimizing existing prompts that underperform or produce inconsistent results
• Selecting appropriate prompting techniques for a specific use case
• Structuring complex multi-step reasoning prompts

Skip when

• The prompt is trivial and already producing good results
• The task is a direct code change, not prompt creation
• You need to execute the task described in the prompt rather than create a prompt for it

---

Scope Boundaries

THIS SKILL ONLY GENERATES PROMPTS. IT NEVER:

• Proactively explores, modifies, or debugs any files in the codebase
• Attempts to fix, debug, or improve code in the project
• Performs the task described in the user's input

Allowed reads: Files the user explicitly references as input context, and docs/prompts/ for saving output.

THE INPUT IS A DESCRIPTION OF WHAT THE PROMPT SHOULD DO, NOT A TASK TO PERFORM.

Example: Help debug React performance issues means:

• CREATE a prompt that helps users debug React performance issues
• DO NOT actually debug any React code

Process

Phase 1: Input Analysis

1. Parse Input: Analyze the provided description or file content
2. Identify Use Case: Determine the intended application and requirements
3. Select Techniques: Choose appropriate prompting patterns and methods

Phase 2: Prompt Construction

1. Structure Design: Create clear prompt architecture using proven patterns
2. Technique Application: Apply selected prompting techniques (few-shot, chain-of-thought, etc.)
3. Constraint Setting: Define boundaries and output format specifications
4. Validation: Ensure prompt follows best practices and guidelines

Phase 3: Documentation & Delivery

1. Display Prompt: Show complete prompt text in formatted code block
2. Implementation Notes: Explain techniques used and design rationale
3. Usage Guidelines: Provide clear instructions for implementation
4. Performance Tips: Include optimization suggestions and best practices
5. Save Output: Save the generated prompt to docs/prompts/ directory (create if needed)