Prompt Constructor
Answer a few questions and get a prompt assembled in the order that models read best: stable material first, task last, data delimited, format explicit. The prompt is assembled in your browser.
Your prompt
Improve with Claude sends the prompt above to this site's server, which passes it to Anthropic's API for one rewrite. Nothing is stored here. Leave out anything you would not send to an API.
Checklist
- Role and audience stated
- Goal is a single clear sentence
- Inputs delimited with tags
- Constraints are explicit
- Output format is specified
- At least one example (standard models)
- Task comes last, stable material first
Why this shape
Stable first, task last
Models attend most to the start and end of the context, and APIs can cache an unchanging prefix. Reference material goes at the top, the request at the bottom.
Delimit the data
Tags around pasted content keep instructions inside it from being followed and make long prompts navigable.
Different advice for reasoning models
Standard models benefit from being told to think first. Reasoning models already do, so the constructor leaves out the scaffolding and keeps the problem statement clean.
History
The 2023 version of this tool sent your text to a local Python service that called GPT-3.5. This version assembles the prompt in your browser with fixed rules. Model-assisted rewriting is back as an optional step, now through Claude.