BuildLLMPrompt: A Prompt Builder for AI Coding Assistants
I’ve been using AI coding assistants like Claude Code and OpenAI’s Codex CLI daily for the past several months. One thing I noticed: the quality of results varies dramatically based on how you frame your request.
A simple “fix this bug” gets mediocre results. But a structured prompt that specifies scope, tool preferences, and behavioral constraints? That’s when the magic happens.
So I built buildllmprompt.com - a simple, privacy-first prompt builder for software engineering tasks.
The Key Insight: AskUser Changes Everything
The most impactful setting isn’t exploration scope or tool preferences—it’s encouraging the AI to ask clarifying questions.
When you crank up “Ask Questions” and set AskUser to “Prefer” or “Must Use”, something interesting happens: the AI converges on the right solution faster, using fewer tokens. Instead of guessing wrong and backtracking, it asks upfront. A 10-second clarification saves 10 minutes of wrong implementation.
I use this systematically when designing new features. Before any code gets written, I want the AI to understand:
- What exactly are we building?
- What are the constraints?
- Which of the three possible approaches do I prefer?
This front-loaded clarification produces better results than letting the AI guess and iterate.
The Problem
AI coding assistants are powerful but sensitive to prompt engineering. The same task can produce vastly different results depending on:
- Scope: Should the AI explore the codebase or stay focused?
- Process: Should it ask clarifying questions or execute immediately?
- Tool usage: Which tools should it prefer or avoid?
- Behavior: Should it complete the task in one go or pause for confirmation?
Manually writing these structured prompts every time is tedious. I wanted a simple UI to generate them consistently.
The Solution
BuildLLMPrompt gives you sliders and checkboxes to configure:
Exploration Scope (1-10)
- Low: Stay focused, don’t explore
- High: Explore codebase thoroughly first
Ask Questions (1-10)
- Low: Execute immediately
- High: Plan first, ask clarifying questions
Tool Preferences
Each tool can be set to Avoid, Normal, Prefer, or Must Use. Tools vary by platform:
- Claude Code (18 tools): Read, Edit, Write, Bash, Glob, Grep, Task, TaskOutput, WebSearch, WebFetch, Todo, AskUser, Notebook, KillShell, Skill, SlashCmd, EnterPlan, ExitPlan
- Codex (12 tools): Shell, ShellCmd, ExecCmd, Patch, ReadFile, ListDir, GrepFiles, ViewImage, Plan, ListMCP, ReadMCP, WebSearch
Behavior Checkboxes (8 options)
- No Stalling: Complete without pausing
- Ask First: Clarify before coding
- Tight Scope: Minimal changes only
- Explore First: Understand codebase
- Commit Often: Small atomic commits
- Test First: TDD approach
- No Comments: Skip documentation
- Dry Run: Explain without doing
Why Keep It Simple?
I considered adding more options: programming language, framework, tech stack, coding style preferences. But I decided against it.
The tool stays high-level intentionally. Language and stack details belong in your actual prompt or CLAUDE.md file, not in behavioral configuration. This builder focuses on how the AI should work, not what it should build.
A Moving Target
One caveat: the tool list and system prompts change over time. Claude Code and Codex both evolve—new tools get added, old ones get renamed or deprecated. BuildLLMPrompt may not always be perfectly up to date with the latest changes.
I update it when I notice drift, but if you see a tool that doesn’t exist anymore or a new one that’s missing, that’s why.
Privacy First
This is a fully client-side application. Nothing leaves your browser. No analytics, no tracking, no server-side processing.
The entire app is a single HTML file with inline CSS and JavaScript. You can view the source code to verify.
Try It
Visit buildllmprompt.com and give it a spin.
The source code is available on GitHub.