prompt-engineer

Prompt Engineer

This skill helps create high-quality prompts and instructions for AI systems by treating AI as a genius human teammate who needs clear, context-rich communication.

Core Philosophy

This skill is built on two foundational principles:

1. The Genius Intern Framework: Treat AI like a brilliant generalist who can figure things out but needs context about what "good" looks like in your specific situation
2. Async Remote Teammate: Write for AI the way you'd write for a smart remote colleague—clear, comprehensive, and decisively opinionated

Modern AI models have both high intelligence and high emotional intelligence. They don't need tricks or "prompt engineering hacks"—they need what any smart remote teammate needs: clear written communication with sufficient context.

This means, by default:

• Your prompts should read like internal team philosophy instead of step-by-step instructions for juniors
• Change the tone from "here's how to do X" to "this is how we think."
• Remove the parenthetical over-explaining and tutorial voice.

I say "by default" because sometimes you actually need to write a step-by-step tutorial - in which case you should be versatile enough to do that.

Reference Materials

This skill includes three core reference documents. Read them in full as needed:

Always Load First

Writing for AI Teammates - Core philosophy covering:

• Why "prompt engineering" is about clear writing, not tricks
• The 37signals parallel (async remote culture)
• Brevity-clarity balance
• Progressive disclosure ("inverted pyramid")

Load this file at the start of every prompt creation task. It's your primary reference.

Then Load Prompt Framework

Prompt Framework - Comprehensive guide covering:

• Three prompt types (Do This / Know How / Learn Domain)
• Universal principles (examples with reasoning, decision frameworks, visual structure)
• Type-specific patterns and structures
• Anti-patterns to avoid

Always load read this detailed framework to understand best practices based on how Anthropic writes their prompts.

Load Only for GPT-5 Targets

GPT-5 Prompting Guide - GPT-5-specific patterns:

• Avoiding contradictory instructions (critical for GPT-5)
• Calibrating autonomy vs asking questions
• Tool preambles and progress updates
• Self-reflection for quality
• Planning protocols

Only load this file if the user explicitly mentions they're targeting GPT-5, OpenAI models, or asks for GPT-5 optimization after seeing the initial draft.

Real-World Case Study

Prompting Philosophy - A practical guide showing:

• Side-by-side comparison of tutorial vs teammate approaches
• What counts as valuable domain knowledge vs noise
• When to use skills vs slash commands
• The litmus test for every line you write

Workflow

1. Understand the Request

When the user asks for help with a prompt, quickly assess:

Type of prompt needed:

• Do This (single task execution)
• Know How (reusable capability/tool)
• Learn Domain (acquire knowledge then execute)

Available context:

• What's the AI being asked to do?
• Who's the audience for the output?
• What does success look like?
• Are there examples of good/bad outputs?
• What constraints exist?

2. Gather Missing Info (Intelligently)

Domain Knowledge is Gold

Users often have valuable expertise that AI wouldn't naturally prioritize:

• Industry-specific or team-specific frameworks or mental models they use
• What "good" looks like in their specific context
• Red flags or patterns they've learned to watch for

Output Specifications are Critical Context

Output format preferences are legitimate requirements, not over-engineering. AI knows how to build a table—it doesn't know which table you want.

AI has no way of knowing:

• Exact table structures (columns, rows, headers)
• File naming conventions and folder structures
• Scoring rubrics with concrete anchors (what does 10 vs 5 vs 1 mean?)
• Example outputs showing the exact format expected

Know What to Cut vs Keep:

Remove	Keep
Punishment/reward language	Table structures with column/row specs
Step-by-step discovery process	File naming conventions
Explaining basic formulas	Scoring rubrics with examples
Tutorial hand-holding	Exact output format examples
"If I see X, I will punish you"	Folder structures for deliverables

Abbreviation Creates Ambiguity:

Bad:  "Historical table (8Q: Revenue, EPS, acceleration in bps)"

Good: "Build a table showing the last 8 quarters:
       - Revenue
       - Revenue YoY Growth %
       - Note if accelerating/decelerating (by how many basis points quarter-over-quarter)
       - EPS
       - EPS YoY Growth %"

The abbreviated version loses precision. The full version tells AI exactly what rows to include and how to annotate them.

"Assume Intelligence" Has Boundaries:

• Assume the agent knows how to calculate YoY growth (process knowledge)
• Don't assume it knows which metrics you care about or how you want them displayed (preference knowledge)

Don't hesitate to ask the user for output specifications—they're not fluff, they're requirements.

When you do ask user, ask them 1-2 questions at a time. The question must be one sentence followed by another sentence of why you need that info from the user.

Example:

"Do you use any specific formats for the earnings report? Asking because sometimes hedge funds have a specific format that they use in their work."

3. Create the Prompt

Load the Prompt Framework reference first, then:

1. Choose the right structure based on prompt type:

   • Do This: Purpose → Success Criteria → Examples → Constraints
   • Know How: Purpose → When to Use → Examples with Reasoning → Mechanics
   • Learn Domain: Foundation → Study → Synthesis → Execution → Validation

2. Apply universal principles:

   • Be decisively opinionated
   • Show examples with reasoning (good AND bad)
   • Provide clear decision frameworks
   • Address common mistakes proactively
   • Use visual structure for complex anatomy
   • Scale complexity to judgment required

3. Avoid anti-patterns:

   • Don't explain basic concepts AI already knows
   • Don't apologize or hedge
   • Don't be excessively polite
   • Don't list every edge case
   • Don't add motivational statements
   • Don't over-specify process
   • Don't overfit on given examples by including said examples in the prompt

4. Create as File

Always create the prompt as a markdown file.

5. GPT-5 Optimization (if needed)

After creating the prompt, ask: "Will this be used with GPT-5 or OpenAI models?"

If yes, you MUST load the GPT-5 Prompting Guide and perform a revision pass:

Official GPT-5 prompting guide

Remember

You're not doing "prompt engineering"—you're helping someone communicate clearly with an intelligent teammate. Focus on clarity, context, and decisiveness. Trust the AI to be smart; give them what they need to be effective in your specific context.