oneshotprompts

For AI agents

If you use an AI agent that works in your browser, it can use the editor directly: read and write your prompt, look up techniques and change the settings, instead of clicking around the page.

How it works

The editor offers a set of tools through WebMCP, a new web standard from Google and Microsoft that Chrome is trying out. An agent working in a Chrome tab with the editor open can see the tools and use them, the way it would use any other tool. It gets the editor’s own guide to how the text works (scenes, steps, [blanks] and linked techniques), so its changes look like yours.

Everything happens in your browser. Each change the agent makes shows a short note, and the ↶ button undoes it like your own typing. Nothing you write is sent to us.

Copy this into your agent

Open the editor in Chrome, then give your agent this, with what you want at the end:

I have oneshotprompts.com open in Chrome. It’s a prompt editor for motion design, and it has WebMCP tools. Call get_writing_guide first, then read_prompt to see what I’ve written so far. Use the tools instead of clicking: edit_prompt or write_prompt to change the text, search_techniques to find the right technique ids, and change_setting for the shape, length, look and so on. Keep my own wording unless I ask you to rewrite it, and leave details you don’t know as [blanks]. When it reads right, call get_final_prompt and show me the finished prompt.

What I want:

Which agents can use it

Any agent that supports WebMCP in Chrome. Support is new and still arriving, so if yours doesn’t see the tools yet, it can read the page’s notes for agents instead, or you can write in the editor yourself and copy the prompt as usual.

The tools

  • read_prompt: the prompt as written so far, with its scenes, blanks, linked techniques and settings
  • get_writing_guide: how the text works: ## scenes, - steps, [blanks] and [words]{technique} links
  • write_prompt, edit_prompt: replace the whole prompt, or change one exact piece of it
  • search_techniques, get_technique: find techniques among all 923 by name or plain description
  • get_settings, change_setting: shape, length, style, look, sound, attachments, how far to take it, what to deliver, and the model
  • list_ideas, start_from, add_scene: ready-made starting points and kinds of scene
  • get_final_prompt: the finished prompt to paste into Claude, ChatGPT or Codex

Developers: the same in plain text is in /llms.txt.