Prompting best practices

​

The C.L.E.A.R. framework

C.L.E.A.R. is a simple structure for writing effective prompts. Each letter stands for a principle that improves clarity and output quality.

​

Concise

Keep prompts focused. Avoid filler words and tangents. Get to the point. Do this:

“Add a booking form with fields for name, email, and preferred time. Include a submit button.”

Not this:

“I was thinking we could really use a form that collects user details like their name and email address, and also asks them when they’d like to book something. It would be great if there was a button to submit it too.”

​

Logical

Order your request in a way that makes sense. Put steps in sequence, and structure complex asks clearly. Do this:

“Create a form with a heading at the top, then fields for name, email, and time slot, and a submit button at the bottom.”

Not this:

“Put the email field first, then add a heading somewhere, include a name field, and a button.”

​

Explicit

Be clear and specific. Don’t rely on the agent to guess. Do this:

“Add placeholder text to each input field. Use a dropdown for time selection. Show a success message when the form is submitted.”

Not this:

“Make a form with some fields and make sure it works properly.”

​

Adaptive

Adjust your style to the task. Different goals need different prompts. For design:

“Create a centered booking form with plenty of white space and a clean, minimal look.”

For functionality:

“When the form is submitted, validate all fields, save the data to the bookings table, and display a confirmation message.”

For fixing issues:

“The submit button isn’t working. Check if it’s disabled or missing the click handler, and make sure it actually submits the form.”

​

Reflective

Review the output and refine your prompt. Learn what works and what doesn’t. If the result is off, try rephrasing, adding examples, or splitting the request.

​

Agent-specific tips

​

Brain (general)

• Good for open-ended and exploratory tasks
• Provide context about your domain and audience
• Use follow-ups to narrow or expand the response
• Attach files when they add useful context

​

Luma (social)

• Specify platform: LinkedIn, Instagram, etc.
• Mention tone: professional, casual, playful
• Include length: character count for Twitter/X, word count for LinkedIn
• Add brand voice via Memory or examples in the prompt

​

Optimus (SEO)

• Provide your website URL when relevant
• Specify scope: full audit, specific page, keywords only
• Mention competitors if you want comparison
• Ask for actionable recommendations, not just data

​

Atlas (data)

• Attach the data file or describe its structure
• Specify output: summary, charts, tables
• Mention the question you want answered
• For charts, specify type: bar, line, pie, etc.

​

Ledger (finance)

• Provide context: business type, time period
• Specify output: P&L, cash flow, expense summary
• Mention any categories or accounts to focus on
• Clarify if this is for internal use or external reporting

​

Justice (legal)

• Attach the document when possible
• Specify what you need: risk review, clause summary, redlining
• Mention jurisdiction if relevant
• Clarify if you need plain-language explanations

​

Polyglot (translation)

• Specify source and target languages
• Mention tone: formal, casual, technical
• Include domain: legal, medical, marketing
• Ask to preserve formatting if needed

​

Handling errors and refinements

When the response is wrong or incomplete:

1. Identify what’s off - Tone? Length? Missing sections? Wrong facts?
2. Give targeted feedback - “The tone is too casual. Make it more formal.”
3. Provide examples - “Here’s the style I want: [paste example].”
4. Simplify - Break the request into smaller steps.
5. Switch agents - If a specialized agent isn’t working, try Brain or another agent.

​

Using Memory with prompts

Memories give agents lasting context. Reference them when helpful:

“Using my brand guidelines in Memory, draft a LinkedIn post about [topic].”

“Reference the product specs in Memory when answering questions about features.”

You don’t have to paste Memory content into every prompt - agents access it when relevant. Mentioning it explicitly can help when you want to ensure it’s used.

​

What’s next?

​

Need help?