This prompt engineering framework for ChatGPT acts as a guide for providing a structured set of instructions to the AI model in order to receive fuller, more accurate, and more desirable results.

I call this SPICIER because it is the best anagram I could think of, but it doesn't perfectly match the order of categories.

S.P.I.C.I.E.R.

ℹ️ This, especially Input, can and should be written for a computer, not a human.

1. Role: Give ChatGPT a persona to give a response through the lens of
2. Instruction: What you want to get out of ChatGPT
3. Context: Help ChatGPT understand the context
4. Input: The input data you want ChatGPT to respond to
5. Source: Reference websites to help shape the input to results
6. Exclusions: Further define parameters by omitting undesired results
7. Parameters: Specify specific options or constraints, i.e. the tone of voice, formatting, length

Example

ROLE:

Act as a software engineer expert in the following:

• TypeScript
• Node.js
• Koa
• Prisma
• Zod

INSTRUCTIONS:

• I will give you my code for a specific Koa middleware and the related Prisma schema
• You will change the code to do the following:
  • Check if something123 exists.
  • If it does, send a patch to X API and ignore both Prisma and the Airtable response
  • If it does not, send a post to Y API and create the record in Prisma

CONTEXT:

• The submit Koa middleware to handles form submissions
• The following are arbitrary unique identifier variables:
  • something123
  • something456

CODE:

INSERT CODE

EXCLUSIONS:

• Assume my web application code is working and is setup correctly