Prompt Engineering

Specificity Questions

• What's the exact problem you're trying to solve? (Not "help me code" but "implement pagination in my React table")
• Are there constraints or preferences? (Performance requirements, tech stack, code style)
• What should success look like? (Working code, explanation, architecture diagram, etc.)

Output Format Questions

• What format do you want the response in? (Code snippet, detailed explanation, step-by-step guide, architecture diagram)
• How long should the response be? (Quick 1-minute answer, thorough explanation, full implementation)
• What level of detail do you need? (High-level overview, implementation details, edge cases)

Risk/Assumption Questions

• Are there known gotchas or common mistakes? (Edge cases, performance pitfalls, security concerns)
• What should the AI explicitly avoid? (Over-engineering, certain patterns, performance anti-patterns)
• Do you need validation or testing included? (Unit tests, integration tests, none)

The 5-Part Prompt Framework

Structure optimized prompts with this pattern:

1. ROLE
   "Act as a [specific role] experienced in [domain]"

2. CONTEXT
   "We are [situation]. The goal is [objective]."

3. TASK
   "Create [specific deliverable]. It should [key requirements]."

4. CONSTRAINTS
   - Use [technology/language]
   - Avoid [anti-patterns]
   - Optimize for [priority: performance/readability/maintainability]

5. OUTPUT FORMAT
   "Format: [code/markdown/explanation]. Include [specific elements]."

Token Optimization Tips

• Be specific about deliverables: "Generate a React hook" not "help with React"
• Mention the tech stack early: Saves AI from asking clarifications
• State constraints upfront: Avoids multiple iterations
• Specify output format: Prevents verbose unnecessary explanations
• Use examples sparingly: Only include if the AI might misunderstand

Common Patterns

Pattern: Code Implementation

Role: Expert [framework] developer
Context: We're building [feature] in [project type]
Task: Write a [component/function] that [specific behavior]
Constraints: Use TypeScript, optimize for performance, follow [patterns]
Output: Code with brief inline comments explaining key sections

Pattern: Bug Investigation

Role: Senior debugger with [framework] expertise
Context: [Observed behavior]. Expected: [correct behavior]
Task: Identify the root cause and suggest fixes
Constraints: No breaking changes, maintain backward compatibility
Output: Explanation + code fix

Pattern: Architecture Review

Role: Architect experienced in [domain]
Context: Current: [description]. Problem: [what's not working]
Task: Propose a better architecture that [desired outcomes]
Constraints: Works with [tech stack], team familiar with [level]
Output: Diagram (Mermaid) + explanation + migration path

Pattern: Explanation/Learning

Role: Patient educator in [domain]
Context: User level: [beginner/intermediate/expert]
Task: Explain [concept] in the context of [specific problem]
Constraints: Use [analogies/examples], avoid [jargon/over-simplification]
Output: Step-by-step explanation with code examples

When This Skill Helps Most

✅ Use for:

• First-time requests where requirements aren't crystal clear
• Complex features requiring multiple iterations
• When previous AI responses were too generic or missed the mark
• Teaching mode where you want specific explanation style
• Token-heavy projects where efficiency matters

❌ Skip for:

• Simple syntax questions ("How do I import X?")
• Quick code snippets you already know how to specify
• When you've already run through a successful prompt once

Quick Checklist

Before delivering your final prompt, verify:

• Specific role/expertise identified
• Problem clearly stated (not vague)
• Success criteria defined
• Tech stack/constraints listed
• Output format explicit
• No ambiguous pronouns or undefined terms
• Token-efficient (no redundant explanations)