Prompt templates you can drop into Claude, GPT, Gemini, or any model. Four shapes: personas (character the model plays), tasks (one-shot instructions), few-shot (instruction + examples), and meta (prompts that generate other prompts).
8 prompts
Persona for a staff-level code reviewer who focuses on correctness, security, and long-term maintainability. Refuses to bikeshed style unless the style choice hides a real bug.
Meta-prompt: takes an existing prompt and returns a tightened version. Removes vague instructions, adds output shape, converts negatives to positives, and flags any missing context.
Persona for a research-desk analyst who assumes every claim is wrong until proven. Interrogates data, checks methodology, flags survivorship bias and confounders before offering a conclusion.
Meta-prompt: given a desired output example, works backward to produce a prompt that would reliably generate that output. Useful for reverse-engineering prompts from products you like.
Task prompt that turns raw meeting notes into a structured summary — decisions, action items with owners, open questions, and a one-line TL;DR.
Task prompt that classifies a user message into one of a fixed set of intents. Returns strict JSON. Good template for building fast intent routers.
Three-example few-shot prompt for sentiment classification. Handles positive / negative / neutral with reasoning. Template you can extend for domain-specific classes.
Four-example few-shot prompt for turning natural-language questions into SQL against a provided schema. Uses PostgreSQL syntax; includes an aggregation, a join, and a filter example.