Prompt Engineering Cheat Sheet (That We'll Still Roast)
Yes, this is genuinely useful. No, that does not exempt your prompt from evaluation. Division 7 files the real method below, then scores it against the standard we hold everyone to.
The Framework: CO-STAR
Context
What is the situation the model needs to know to respond well. Background, not backstory.
Objective
The one thing you actually want produced. State it as a task, not a topic.
Style
The writing style to imitate: formal, conversational, technical, a specific author or brand voice.
Tone
The emotional register: serious, playful, urgent, reassuring.
Audience
Who is reading the output. A prompt for executives and a prompt for beginners need different words.
Response Format
The literal shape of the answer: a table, a five-bullet list, a 200-word paragraph, JSON.
Division 7's Scoring Rubric
| Element present | Points | Notes |
|---|---|---|
| Context | +15 | Missing context is the single most common defect we log |
| Objective stated as a task | +20 | Topics are not tasks |
| Style specified | +10 | Optional but raises consistency |
| Tone specified | +10 | Prevents tonal drift on long outputs |
| Audience named | +15 | Changes vocabulary and depth automatically |
| Response format defined | +20 | Single biggest quality lever available |
| Threats, bribes, or ALL CAPS urgency | -25 | Does not help. Filed under Division 7 case law. |
Think you can write a CO-STAR prompt that survives evaluation?
Submit it and find out →