Task picker
An editable worksheet for choosing a recurring check with a clear input and a result you can verify.
AI Marketing Agent Starter Kit · Version 1.0
Build a weekly marketing review that tells you what changed, what needs checking, and what to look at next.
Start with fictional data and a worked answer key. Then adapt the brief to a report you already understand.
Download the free kit ↓ZIP · 11 editable Markdown and CSV files · No email required
“Review my marketing” leaves too much open. “Compare these two complete weeks and flag the declines I define” gives you a result you can check.
The example includes a genuine zero, a missing value, and a new channel. Your assistant needs to handle each differently. The answer key shows you how to spot the mistakes.
You do not need three separate agents to begin. Start with one chat. Add the optional reviewer and planner roles when the first check works.
An editable worksheet for choosing a recurring check with a clear input and a result you can verify.
Define the analyst's task, have a reviewer check its evidence, and ask a planner for one proposed next step.
Two fictional lead reports, a ready-to-paste prompt, an answer key, and five boundary and failure cases.
Check the source rows, record corrections and time spent, then hand off accepted findings with unresolved questions.
Download and unzip the kit. Open 00-start-here.md in a text editor. The worksheets open in a spreadsheet app.
Paste the example prompt into your AI chat and attach the two CSVs. Keep the answer key out of that chat until you have saved its first answer.
Use the answer key and source rows. Record mistakes in the run log. Revise one rule and retry before moving to your own report.
No. It gives you editable briefs, worksheets, and an example to run in your own AI chat. You pass the files between roles. Nothing connects to accounts, changes campaigns, or sends messages.
The kit explains what a funnel review needs, but the included lead counts cannot show stage conversion. Define one transition and supply the matching stage counts, timing, and tracking gaps before asking for a drop-off analysis.
Use an app that accepts text prompts and can read CSV attachments or pasted tables. Output quality varies. The answer key lets you check the sample result; no model has been certified by this kit.
Use the fictional example first. For your own report, use a permitted copy and remove private customer details. Keep a person responsible for reviewing the output.
Start with the included example.
Download the free kit ↓Questions after your first run? Bring them to Cosmic AI on Eden.