Describe what you want

What a good request includes

OpenAI’s Codex guide suggests four parts. They work for every agent:

• Goal: what you want to change or build.
• Context: what matters for this task, such as a page, a doc, an error, or an example you like.
• Constraints: what to stick to or stay away from.
• Done when: what should be true at the end, “such as tests passing, behavior changing, or a bug no longer reproducing.”

You don’t need to know the code. Describe what people see and do, and let the agent find where that lives. Claude Code’s guide suggests treating the agent like a capable colleague: “Give context and direction, then trust Claude to figure out the details.”

How to add context in Enjoy

• Show, don’t describe. Paste a screenshot into the message box, drag an image in, or select Message options, the plus button, then Add images. A message can include up to 8 images. Claude Code’s guide recommends screenshots “when text descriptions would be unclear or cumbersome.”
• Point to your docs, files, and workflows. Type @, or select Files and workflows in an empty message box, and choose from Docs, Workflows, or Files. You can add up to 8 of each.
• Attach a file. In the computer app, Message options also has Attach file, for files up to 20 MB each.
• Write as much as you need. Press Shift+Enter for a new line. A message can be up to 100,000 characters.

In the iPhone app, use the + button (Add image) and the @ button (Reference files and workflows) under the text you type.

An example

Instead of:

“Make the homepage better.”

Try:

“The homepage should make people want to book a class. Right now the schedule is hidden at the bottom (screenshot attached). Move the schedule up, right below the welcome message, and make the ‘Book’ buttons stand out. Keep our colors and the existing photos. It’s done when the schedule is visible without scrolling on a phone, and you’ve checked that in the preview.”

Together, the sentences cover all four parts, and the last one also tells the agent how to check its work.

Common pitfalls

• Several unrelated requests in one message. Send them as separate tasks. See Use one conversation per task.
• Leaving out what done looks like. Without it, the agent stops when the work “looks done.” See Check the agent’s work.
• Describing a bug without the symptom. Say what you did, what you expected, and what happened instead. Paste the exact error text.
• Assuming the agent remembers other conversations. A new conversation starts fresh. Mention the doc or decision it needs.
• Big, fuzzy ideas. For those, ask the agent to interview you and propose a plan first. See Plan before you build.