Prompt Optimizer + LangGPT prompt asset pipeline for reusable, structured prompts

What they did

They started from a rough, minimal role prompt (for example a reviewer or a marketplace bargaining reply) and ran it through Prompt Optimizer's web app to push a small model away from generic output toward structured, specific behavior. They then used LangGPT's structured prompt framework to formalize the optimized prompt into a reusable template with clear role, goals, constraints, and variables. Variables were swapped in for different scenarios (item details, price anchors, buyer offers, tone, negotiation goals) to confirm the template generalized. They compared outputs before and after optimization, iterated on weak spots, and saved the final versions as reusable prompt assets rather than one-off text.

How it went

The optimized prompts produced noticeably more structured and specific outputs than the originals, and the LangGPT formatting made them easy to reuse across scenarios by swapping variables. The workflow is lightweight and works with small models, though quality still depends on the underlying model and manual evaluation.

Worth knowing

Requires an API key or configured model provider for Prompt Optimizer, and results vary by model. LangGPT is a formatting convention, not an automated validator, so structure quality is manual. No automated regression testing of prompts is included by default.

Wiring

Install or open Prompt Optimizer (web app, Chrome extension, desktop, or Docker) and configure a model provider/API key. Paste the original prompt, run optimization, and review the suggested rewrite. Take the optimized text into LangGPT's structure template (Role/Profile/Goals/Constraints/Skills/Workflow/Output Format) and convert fixed scenario details into variables. Test by substituting variable values and comparing outputs, then store the finalized templates in the repo under a prompts directory for reuse.

Try it yourself

Look at the prompts in this repo (or the prompt strings in my code). Pick the three most important ones. For each: (1) rewrite it using LangGPT structured format with Role, Goals, Constraints, Workflow, and Output Format sections; (2) identify which parts should become variables and list them; (3) suggest concrete optimizations to make outputs more specific and less generic; (4) show a before/after example output for a realistic input. Save the results to prompts/optimized/ as markdown files.