How I built a meta-agent to write my prompts for me

Jenil Saija — Fri, 16 Jan 2026 17:08:36 +0000

The Problem: "Blank Page Syndrome" for Agents

I've been building a lot of AI Agents lately (specifically for Replit and Gemini). The hardest part isn't the code—it's the System Prompt.

To get an agent to reliably perform complex tasks, you need a prompt that defines:

The Persona
The Tool Capabilities (and when to use them)
The Logic Constraints (what not to do)

Writing this from scratch every time was killing my momentum.

The Solution: A "Meta-Agent" 🤖

I decided to build a tool to automate this. I call it OpenBuilder.

It's a "Text-to-Prompt" engine.

Input: "I need an agent that researches SEO keywords and writes blog outlines."
Output: A structured, 100+ line System Prompt ready to copy-paste.

How it works (The Stack)

I built this using Next.js on the frontend and a Node/Express backend.

The core logic uses a "Chain of Thought" approach in the backend:

Intent Analysis: It breaks down your request into a goal.
Tool Mapping: It decides if the agent needs web search, file access, or code execution.
Constraint Generation: It adds safety rails to prevent hallucinations.

The Result

I've opened it up for everyone to use for free. You can generate prompts for Gemini Gems, ChatGPT, Claude, Replit Agents, and N8N Workflows.

You can try it here: openprompt.sprktac.in

Update: I just shipped the "Save to Workspace" feature! 🚀
You can now save your generated agents and version control them just like a GitHub repo.

Let me know what you think! 👇

DEV Community: Jenil Saija

How I built a meta-agent to write my prompts for me

The Problem: "Blank Page Syndrome" for Agents

The Solution: A "Meta-Agent" 🤖

How it works (The Stack)

The Result