I took the approach I've battle-tested across dozens of projects and wrapped it into a CLI tool that runs on Claude Code.
Mango Lollipop runs entirely through Claude Code locally.
You give it your product URL, it analyzes your business, and walks you through building a complete messaging system using the AARRR pirate metrics framework.
It will take about 15 minutes from init to a complete v0.1 messaging system ready to implement and test.
What it produces:
- A structured messaging matrix with triggers, guards, suppressions, and timing for every message - Full message copy (email, in-app, SMS, push) written in your brand voice - An Excel workbook for your team's source of truth - An interactive HTML dashboard with journey maps and message previews - Developer hand-off docs with event specs and code examples
Demo with sample outputs here https://sr-kai.github.io/mango-lollipop/
The whole thing is built as markdown skills (slash commands). Each step is independent so you can review, iterate, and course-correct before moving on.
I built it on Claude Code, but since the skills are just markdown files, it should work in any AI coding tool that supports them, like Cursor, OpenCode, Windsurf, etc.
Different models produce WILDLY different copy, I recommend trying a few to see what works best for your use-case.
npm install -g mango-lollipop
GitHub: https://github.com/sr-kai/mango-lollipop
I'd love feedback on the approach and how you build similar system.
The trigger/wait/guard/suppression model is not the most elegant solution, so if you have a better implementation please share your experience.