#The Anatomy of a Great Prompt¶

Most production prompts share the same skeleton. Learn the skeleton once, reuse it forever.

The Six Components¶

Weak Prompt ❌¶

"Tell me about this customer feedback."

Vague task, no role, no format, no constraints. The output is unpredictable and unusable by code.

Engineered Prompt ✅¶

1You are a senior product analyst.            ← role
2
3Classify the customer feedback below and     ← task
4extract any feature requests.
5
6Feedback:                                     ← context
7"""
8The app is great but it crashes when I export
9to PDF, and I really wish it had dark mode.
10"""
11
12Rules:                                        ← constraints
13- Sentiment must be one of: positive, negative, mixed
14- Only list feature requests explicitly stated
15
16Respond as JSON:                              ← output format
17{
18  "sentiment": "...",
19  "bugs": ["..."],
20  "feature_requests": ["..."]
21}

This is testable, parseable, and reliable.

Principles That Always Apply¶

1. 1Be specific. "Write a summary" → "Write a 3-sentence summary for a non-technical executive."
2. 2Show, don't just tell. One good example beats a paragraph of description.
3. 3Positive instructions beat negative ones. "Respond only in JSON" works better than "Don't add explanations."
4. 4Separate data from instructions using delimiters (""", XML tags). Prevents the model from confusing input with commands — also a security boundary (Module 5).
5. 5Specify the format explicitly. If code consumes the output, the format is part of the spec, not an afterthought.

A Reusable Template¶

1You are {ROLE}.
2
3Your task: {ONE_SENTENCE_GOAL}
4
5Input:
6"""
7{DATA}
8"""
9
10Rules:
11- {RULE_1}
12- {RULE_2}
13
14Output format:
15{FORMAT_SPEC}

You will refine this template throughout the course.