Zero dependencies · Works in any LLM

Raw prompts →
Structured JSON

One meta-prompt that extracts role, goal, audience, constraints, and format from any natural language request. Build reusable prompt libraries. Ship consistent AI outputs.

View on GitHub Try Baxter GPT

What it does

Five primitives that make prompt engineering systematic.

🔍

Structure Extraction

Pulls out Role, Goal, Audience, Context, and Constraints from any freeform input.

📐

Format Standardization

Enforces a consistent JSON schema — versionable, auditable, API-ready.

🧠

Intelligent Inference

Fills missing fields with safe, sensible defaults when your prompt is sparse.

🛡️

Execution Prevention

Transforms only — never fulfills the request. Separation of concerns built in.

⚡

Zero Dependencies

Pure text. Copy-paste into ChatGPT, Claude, Gemini, or any LLM chat.

Get started in 3 steps

No setup. No API key. No install.

1

Copy the transformer template

Grab the JSON template from Metaprompt.md. It defines the transformation contract — what to extract and where to put it.

2

Replace <USER_INPUT_HERE> with your prompt

Drop in any raw natural language request — "write me a performance review", "summarize this paper", "generate a SQL migration". Messy is fine.

3

Paste into your LLM. Get structured JSON back.

The model returns a fully populated template. Review, tweak, version it. Reuse it across chats, teams, or APIs. One transformation — infinite reuse.

See the transformation

One casual sentence in. Production-ready JSON template out.

INPUT

raw_prompt.txt 
"Write an email to my manager
asking for a day off
next Friday."
LLM

OUTPUT

template.json 
{
  "Role": "Email Assistant",
  "Goal": "Draft a professional leave request",
  "Audience": "Workplace manager",
  "OutputFormat": "text",
  "CreativityLevel": "Low",
  "Style": "polite, professional, concise",
  "Constraints": {
    "Positive": ["Use formal tone", "Be brief"],
    "Negative": ["No casual slang"]
  }
}
💡

The transformer separates transformation from execution — it never writes the email. Take the JSON, paste it into a new chat as your structured prompt, and get consistent, predictable output every time.

Ready to build your prompt library?

Star the repo, copy the template, or jump straight into the Baxter GPT to try it live.

⭐ Star on GitHub Try Baxter GPT →