> ## Documentation Index > Fetch the complete documentation index at: https://evalgate.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Create a prompt > Author, version, and test a prompt before it ships. # Create a prompt A prompt in EvalGate is a versioned, content-hashed message template. You author drafts, test them, send them for review, and publish an approved version to a protected environment. ## Step 1 — Open Prompt Hub In the sidebar, click **Prompts**, then **New prompt**. A screenshot of the Prompt Hub registry with the New prompt button. ## Step 2 — Author the first version Give the prompt a stable **key** (for example, `checkout-assistant-system`). The key is the public identity; versions are immutable and content-hashed under it. Write the message content. You can include tool definitions and variables. Save the draft. A screenshot of the prompt workbench with a system message and a variable. ## Step 3 — Test against the Gateway Before review, run a **Gateway test** to confirm the prompt compiles and the model accepts it. This needs a [connected provider key](../evalgate-101/connect-provider-key). ## Step 4 — Submit for review Click **Submit for review**. A reviewer with the `prompt:publish` permission can approve the version. Until then, it is a draft and cannot be served by a protected alias. ## Step 5 — Publish to an environment After approval, publish the version to a protected environment alias (for example, `production`). The alias resolves that exact version — never a draft — and every consumer retains the exact `versionId` and `contentHash`. ## Up next [Publish and roll back](./publish-prompt) a version safely.