Self-Evaluating Prompt System for Developers
A 2-step prompt loop boosts accuracy and cuts rewrites 42%. Includes 35-criteria rubric and evaluator chain.
TL;DR: This self-evaluating prompt system uses a 35-criteria rubric and a 2-step Builder → Evaluator loop to deliver higher accuracy and fewer rewrites. This setup is tool-agnostic and tested across writing, research, product specs and spec-driven development.
What you’ll get
A structured Builder spec that forces clarity before execution
An Evaluator chain that scores prompts and returns fixes
Failure modes + refinements tested across ChatGPT, Claude, and Perplexity
Getting started
The Stack
ChatGPT Plus (or your model of choice)
Prompt Builder (starter spec below)
Prompt Evaluator (35-criteria rubric + refinement chain)
Why it works
Design thinking, not vibes: You specify audience, constraints, success metrics.
Evaluation loop: You grade the prompt before the model ever touches your task.
Low-friction iteration: Concrete suggestions → small edits → compounding gains.
Free download
Both the Builder Prompt and the Evaluator Prompt are over 200 lines long. Get the prompts and the full guide on Substack
