What happens when you point a coding agent at a real, messy codebase and ask it to sort out the dev environment?
Justin (Our DevRel Lead <3) put our `floxify` skill to the test on Mastodon, a challenging repo (with Ruby, Node, PostgreSQL, Redis, native dependencies, multiple services) to build from source.
The skill went through the repo, figured out what it needed, built the Flox environment, checked it, hit a few problems, fixed them, and kept going until `flox activate` worked.
All in just a couple of minutes.
The result: a declarative, reproducible Mastodon environment without using containers, and without Justin (or an agent) having to reverse-engineer the setup from scratch.
He also tried the `flox` skill for a new project, starting from what he wanted to build rather than an existing codebase.
Two different starting points, same idea: Give the agent what it needs to quickly set up the runtime environment. This way it spends tokens running and testing code instead of debugging setup.
Justin wrote up the whole experiment, including what worked, what broke, and what the skills did about it 👇
