What You'll Learn Along the Way

Bash commands. This really clicks once you’ve been running the same process over and over. Take a changelog workflow, for example. After doing it manually enough times, the “workflow” becomes obvious. That’s when you can get the agent to create a slash command that runs a series of bash commands and prompts the model to do specific things: reading through GitHub diffs, checking what’s behind a feature flag, sorting items into the right sections of new features and bug fixes.

From there, it’s natural to get more into bash and CLIs. A common pattern is to stop using MCPs (Model Context Protocol servers) and use CLI versions of tools instead. Yes, MCPs take up context, but mostly it’s simpler when you only need a few of the tools an MCP would include. For tools like Supabase, Vercel, and GitHub, the CLIs often work better than the MCPs.

Building custom CLIs is also a great practice. For example, you might build a Linear CLI to query issues and run everything from the terminal instead of opening the desktop or web interface.

VPS (Virtual Private Server). You might abstractly know what a VPS is: it’s like another computer that’s always on, somewhere else. But until you truly need one, the practical details stay fuzzy. A common use case: running something like a data tracker that pulls information every single minute and needs to always stay on.

A VPS is also handy for running bots and using tools like SyncThing to sync local repos, so your code is always up to date and in the same state you left it. That means you can pick up your work on the go.

Skills. These are reusable knowledge files and commands that your agent can use across projects. Think of them as portable toolkits. For example, a Gmail CLI that lives at your root directory and can be pulled into any project. Anytime you need Gmail functionality (like an email triage system), it’s just there, ready to go.