AI GitHub
OpenCode for provider-flexible coding sessions
Work with a coding agent in the terminal using a configured model provider.
Key Takeaways
- OpenCode gives the model tools for navigating and changing a checkout.
- Configure the provider and project access before trying a write.
- Start with a read-only investigation of a failing test.
- Tool permissions and model capabilities vary with configuration.
How it works
OpenCode gives the model tools for navigating and changing a checkout. Its session interface shows the conversation and tool activity, while provider configuration controls where prompts go.
What you need
Configure the provider and project access before trying a write. Use a test checkout with the same dependency layout you expect in day-to-day use.
What to try first
Start with a read-only investigation of a failing test. Check which files the agent inspected, then allow a small edit in a separate branch and validate the result.
Try a workspace with two packages that have similar names. A focused investigation should identify which package owns the failing command before edits begin.
Where to look for failures
Tool permissions and model capabilities vary with configuration. Review shell access, credential exposure and any persistent session state before opening a sensitive repository.
Related Franklin pages
Use the same clean checkout and failing test to compare terminal agents. Look for correct file selection, a minimal diff and evidence from the test run.
- Use OpenAI's terminal coding agent to inspect, edit and test a project. Codex CLI for repository work from the terminal
- Make model-assisted edits in a Git checkout and review the result as a diff. Aider for terminal-based pair programming
For a library of development workflows that can complement a coding agent, see Franklin’s Superpowers profile. Superpowers
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗