Prompt

The built-in prompt tag — a copyable AI-prompt block: the prompt text with a copy button plus 'Open in ChatGPT / Claude / Cursor' deep-link buttons that pre-fill the prompt in that tool. Usage, actions, and a live example.

A built-in tag that renders a copyable AI-prompt block: your prompt text in a tidy code surface with a copy button and one-click “Open in …” deep links that launch ChatGPT, Claude, or Cursor with the prompt already filled in. It’s ideal for docs that hand the reader a ready-to-run prompt — a “fix this with AI”, a scaffolding request, a review checklist — so they can send it to their assistant without copy-pasting by hand.

The body is the literal prompt text: it’s kept verbatim — never run through Markdown — so line breaks, blank lines, and any code or angle brackets survive exactly as written (that’s the text that gets copied and encoded into each deep link). The title is an optional heading, and actions chooses which buttons appear. Close it with {% endPrompt %}.

Use it as {% prompt %} in Markdown, or call it from Python logic (loops, snippets) via component('aardvark', 'prompt', …).

Usage

Put the prompt between {% prompt %} and {% endPrompt %}:

You are a senior engineer reviewing a pull request. For each file in the diff:

1. Summarize what changed in one sentence.
2. Flag any correctness bugs, with the line and a suggested fix.
3. Note missing tests or docs.

Be concise and specific. Do not restate the diff.

The Copy button copies the whole prompt to the clipboard; each Open in … button URL-encodes the prompt into that tool’s “new chat” link — https://chatgpt.com/?q=…, https://claude.ai/new?q=…, and Cursor’s cursor://…?text=… app deep link.

Choosing the buttons

actions is a comma-separated list of the buttons to show, in order. Valid values are copy, chatgpt, claude, and cursor; the default is all four (copy, chatgpt, claude, cursor). Pass a shorter list to trim or reorder them — here, just Copy and Claude:

Add a new REST endpoint POST /widgets to this codebase.

- Validate the request body against the existing schema conventions.
- Persist with the current ORM/data layer.
- Return 201 with the created resource, 400 on validation errors.
- Include a unit test that covers the happy path and one validation failure.

Without a title

title is optional — omit it and the block shows a plain Prompt label above the text:

Summarize the following text in three bullet points, then give it a one-line title.

Verbatim body

The body is not Markdown — it’s copied exactly as written, so **stars**, # hashes, <angle brackets>, and fenced code all stay literal. If your prompt contains aardvark template syntax (a literal {% … %} or {{ … }}), wrap that part in a raw block so the template engine leaves it alone — the same {% raw %} … raw-guard convention used everywhere else in these docs.

Options

CSS Selectors

Each prompt carries data-aardvark-island="Prompt" on its wrapper and renders its own class names — target the block, the header, the buttons, or the prompt text.

[data-aardvark-island="Prompt"]              /* the island wrapper */
[data-aardvark-island="Prompt"] .aardvark-prompt        /* the bordered block */
[data-aardvark-island="Prompt"] .aardvark-prompt-head   /* the title / buttons row */
[data-aardvark-island="Prompt"] .aardvark-prompt-action /* a copy / open-in button */
[data-aardvark-island="Prompt"] .aardvark-prompt-text   /* the verbatim prompt <pre> */

Injecting Attributes

attr={…} forwards raw HTML attributes — id, data-*, ARIA, analytics hooks — onto the rendered block root.

Explain the following error message in plain language, then list the two most likely causes and how to check each.