In a similar vein, I use Storybook to build my presentational components, and run each story in a try/catch to check for "explosions".

If you add some dependency injection to look for `console.warn` or `console.error` calls, you can catch failed PropTypes or similar.

Today you can even combine all of these with Jest's new "Tree Snapshot Testing" to compare the actual output of `render` with an expected/approved output.

Finally, you can use your stories in combination with a service like Browserstack or SauceLabs to run these tests across platforms, and even take screenshots of each story and compare them to test for styling regressions and cross-platform differences.

Sources:

Storybook - https://github.com/kadirahq/react-storybook

Jest's "Tree Snapshot" testing - https://facebook.github.io/jest/blog/2016/07/27/jest-14.html

Hmmm... this is good but presumably react components are actually going to do things. But I suppose you're saying that the code that actually does things should be separated from the react code and should be tested separately from the react components.