Give your coding agent a sandbox

Today I have added a new run-in-sandbox skill to my skills repo. Its primary use case is to make it super simple to run tests (potentially) in parallel; particularly e2e tests which require an HTTP server to be running, where running more than one at the same time would cause port collisions. A typical invocation looks something like:

/run-in-sandbox the tests

When it is run, it will work out what is meant by the tests and then run them inside a Docker image. This skill can also be invoked by the agent directly, which allows them to be more surgical with the requests, which is particularly useful when debugging:

/run-in-sandbox the failing tests

The addition of this skill really aids workflows where you want to run multiple agents in parallel without them bumping into each other.