Ad Brief

This process is the same regardless of ad platform. Run it before any campaign work.

How to interact

• Reply in the same language as the user.
• Use AskUserQuestion / ask_user_question when asking about experience level, goals, or decisions. Present 2-4 selectable options - avoid making them type long answers. For questions where multiple answers are valid (e.g. target audiences, customer types), allow selecting multiple or typing a custom answer. Never force picking one when the user may have several.
• Exception: financial and numerical inputs (budget, CPA, LTV, revenue) must be open-ended — never pre-select dollar amounts. Pre-selected options anchor the user on arbitrary values. Ask them to type their number, then validate it against the formulas in the template.
• Ask for reference materials early. Ask if the user has existing documents you should read: brand guidelines, ICP/persona descriptions, competitor lists, landing page URLs, or previous ad performance data. If they provide files, read them and incorporate into your recommendations.

Check for existing brief

First, check if .agents/ad-brief.md exists in the user's project. If it does, read it and skip this process.

If not, build one using the template at template.md. Two phases:

Phase 1: Research (you do the work)

1. Ask the user for their product name and URL (or detect from the project: package.json, README, etc.)
2. Research proactively: visit their website, infer the audience, look up competitors, figure out the offer
3. Fill in as much of the brief as you can from what you found

Phase 2: Confirm and complete (user fills gaps)

1. Present your filled brief to the user: "Here's what I found. Correct anything that's wrong and fill in what's missing."
2. For any remaining gaps, ask one at a time using AskUserQuestion / ask_user_question
3. Save the result as .agents/ad-brief.md in the user's project root. This persists for future sessions.