Playbook + AI agents
Your AI can build anything.
It just can’t feel anything.
Three skills give your agent the playbook’s judgement. One to build, one to plan, one to review. Works with Claude, Cursor and any AI that reads instructions.
Claude
Codex
Cursor
Replit
/playbook-build
Design the moment before you write the code
Ask an AI for a screen and you get the default: “No items found”, a button that says “Continue”, a paywall on launch. Competent, and forgettable.
This skill picks two to four named plays before your agent writes a line, so every moment is designed, not defaulted.
Say something like
- “Design the onboarding for my meditation app”
- “Write the empty state for the saved items tab”
- “What should this button say?”
What comes back
A designed moment, not a competent one: the plays it used, why they fit, and copy that sounds like a person wrote it.
/playbook-plan
Turn a stuck metric into a plan
“Improve retention” is not a plan. Ten generic tactics, none sequenced, none measured, is not one either.
Give it the stuck number and it diagnoses the leak, then hands back a prioritised play stack with a measurement plan.
Say something like
- “My D7 retention is stuck at 12%”
- “How do I get more free users to upgrade?”
- “What should we build next to improve activation?”
What comes back
A strategy, not a feature: which plays, in what order, and the numbers that will tell you it is working.
/playbook-review
A critique that earns its approval
Ask an AI to review your flow and it tells you it looks great. Flattery is the cheapest output a model can produce.
This skill reviews at a high craft bar: what fails, which play fixes it, what the fix looks like. Approval is earned.
Say something like
- “Would you ship this paywall?”
- “Review my signup flow”
- “What’s wrong with this notification copy?”
What comes back
An honest read: the failures named, the fixes shown, and a bar your product has to clear rather than a compliment.
Built from the playbook’s 175+ cards
Your agent stops shipping defaults and starts shipping moments users feel. Seven years of product craft, applied while you build.