Lean Canvas Template — AI skill for Claude Code

lean-canvas-template

Answer nine questions. Ship the smallest app that tests your riskiest assumption.

This is a template for starting apps with an AI coding agent (Cursor, Claude Code, Codex, whatever you use). Instead of writing a PRD, you fill in a Lean Canvas through a short interview with your agent. The agent then builds only what tests your riskiest belief, and the canvas stays the source of truth as you learn.

Why a canvas instead of a PRD

A PRD is pages of detail about features nobody has validated. It made sense when engineering was the bottleneck and specs had to survive a handoff. Your agent is not that bottleneck. It can figure out the technical details on its own; what it can't know is whether anyone wants the app, will open it twice, or will pay at the paywall. That part is yours, and it fits on one page.

A Lean Canvas (Ash Maurya's one-page adaptation of the Business Model Canvas) captures exactly that part: who it's for, what problem it solves, why you, how it makes money, and the one belief that kills the whole thing if you're wrong. You build to test that belief. Everything else waits.

Quickstart

1. Click "Use this template" above (or clone the repo).
2. Open the folder in your AI IDE or agent.
3. Say: "Read AGENTS.md and start."

Your agent interviews you, one question at a time, about customers, problem, pricing, and channels. Then it fills in `CANVAS.md`, names your riskiest assumption, proposes the smallest v0.1 that tests it, suggests a stack (override it with a sentence), and builds.

No CLI, no phases, no setup. The state of `CANVAS.md` is the state of the project: empty means interview, filled means build.

What's in the box

Four files.

File	What it is
CANVAS.md	The Lean Canvas. Single source of truth. Every feature must trace back to a box here.
AGENTS.md	Instructions your agent follows: interview rules, build rules, house rules.
LOG.md	The agent's memory: stack, decisions, exp