OpenCode Coding Agent: terminal workflow, cheatsheet, and why it matters

OpenCode feels useful when I want the coding agent to stay close to my normal terminal workflow. I can ask it to inspect a repository, make a change, review a diff, or explain a bug without turning the whole process into a separate IDE experience. The nice part is that it is open-source and provider-friendly, so I am not locked into one model or one company.

Claude Code is still very strong for deep coding sessions and polished agent behavior. OpenCode is interesting for a different reason: it gives me more control over the stack. I can choose providers, experiment with models, keep the workflow terminal-first, and use it as a practical coding helper instead of a black-box product. For personal projects, that flexibility matters.

These are the commands I would keep in a small note when using OpenCode on a real project. The pattern is simple: authenticate once, run a specific request, check the diff, run the tests, then ask for a smaller follow-up if needed.

I would not use an agent by saying something vague like ‘make the app better’. I would give it one job at a time: inspect the repo, explain the issue, change the smallest useful piece, then verify with tests or a build. That keeps the output easier to review and avoids the agent rewriting too much code at once.

I would pick OpenCode when I want an open-source coding agent, when I want to try different models, or when I want the terminal to remain the main place where work happens. It is especially good for personal projects, experiments, and setups where local or self-hosted model support is part of the appeal.