Prompt Refiner
Turn rough prompts into production-ready briefs.
Structured prompt
# Role You are a senior editor with a strong, distinctive voice. You strip filler, fix flabby phrasing, and never use stock phrases. # Context The user is asking for: "write a tweet thread about how AI agents are changing software engineering". Treat the rest of their message as supporting context. # Task write a tweet thread about how AI agents are changing software engineering # Constraints - Do not invent facts. If you don't know something, say so. - Keep the output focused on what was asked — no preamble. - If the request is ambiguous, ask one clarifying question first. # Output format - Lead with the answer, not a recap of the question. - Use prose by default; use bullets only when a list genuinely beats a sentence. - Cite specific names, files or numbers wherever possible.
What this tool does
Applies the standard prompt-engineering pattern: role → context → task → constraints → output format. The result drops cleanly into Claude, GPT, Gemini or your own system prompt.
About Prompt Refiner
A heuristic prompt refactorer. It looks at what you wrote, applies the standard prompt-engineering patterns (role assignment, contextual framing, task definition, constraints, output spec, examples slot) and gives you a structured version you can paste into Claude, GPT-4, Gemini or your own system prompt.
Why we built it
We were copy-pasting the same prompt-engineering checklist over and over. So we turned it into a tool.
How to use
- 1Paste your rough prompt into the editor on the left.
- 2Pick a tone (assistant, expert, analyst, writer) and a target model.
- 3Copy the structured version — it's ready to drop into Claude or GPT.
Most prompts that fail in production fail for the same reason: missing structure. A good prompt assigns a role, frames the context, names the task, defines constraints, and specifies output format. A weak prompt does one or two of those and hopes for the best. This refiner walks any input through the same checklist a careful engineer would use and hands back a prompt that's actually production-ready — not flashy, just complete.
The five elements of a structured prompt
Role: tell the model who it's being. "You are a senior backend engineer reviewing pull requests for race conditions" outperforms "please review this code" by a wide margin. Context: what the model needs to know about the situation, the codebase, the user, the constraints. Task: a clear directive. Constraints: things the model must avoid, must include, or must obey ("never recommend mutation", "output must be valid JSON"). Output format: explicit — markdown sections, JSON schema, plain text only, list of N items. Models don't read minds. Each missing element is a place the output will drift.
Why this isn't just rephrasing
The refiner doesn't paraphrase your prompt to sound better. It restructures it into named sections, fills gaps, and surfaces constraints that were implicit in your draft. The improvement isn't in the prose; it's in the structure. Pasted into Claude, GPT-4, Gemini, or any other modern LLM, the structured version yields outputs that are more predictable, more on-spec, and easier to iterate on.
Frequently asked questions
Quick answers to the questions people actually ask about Prompt Refiner.
What does the refiner actually change?
What does the refiner actually change?
It restructures your prompt into the five-element pattern most modern LLMs reward: role, context, task, constraints, output format. If your draft is missing any of those, the refiner fills the gap with a sensible default and flags what it added. The wording is still yours — the structure is what changes.
Does it work for every model?
Does it work for every model?
Yes — the structured-prompt pattern works the same way across Claude, GPT-4, Gemini, Llama, and most open-weight models. The refiner offers per-model tone presets (Claude tends to like more explicit constraints; GPT-4 tolerates more compressed prompts), but the core structure is universal.
Is my prompt sent to a model?
Is my prompt sent to a model?
No. The refiner runs a rule-based transformer locally in your browser. Your prompt isn't sent to Claude or GPT or any other API. That's deliberate — we don't want sensitive prompt content (system prompts, customer data, internal tooling) crossing a network boundary for what's structurally a formatting task.
When should I not use the structured pattern?
When should I not use the structured pattern?
Conversational chat with an LLM — "What's a good restaurant near me?" — doesn't need a role/context/task structure. The pattern earns its keep when you're building something with the LLM: a feature, a workflow, a system prompt that gets reused thousands of times. The more an output gets generated, the more the structure pays off.