Handing a coding agent a big task on day one makes it hard to tell what it did or whether it worked. This guides your coding agent through one bounded change in your own project, from prerequisites to a checked result, so you see the whole loop once.
It covers choosing the workspace in the CLI or editor integration, having Codex find the repository instructions, scripts and current changes, and giving it one visible outcome with constraints and a way to verify it. The worked task improves the empty state on a main list: it explains what belongs there and adds a link to create the first item.
You then review the evidence: the diff, the running result, and which checks actually ran or were blocked. A check that could not run is resolved, not treated as a pass. You keep a small checkpoint to return to before moving to a larger task.
You need a local project and access to Codex through your chosen interface. Provider accounts, model weights, GPU compute and hosting are not included.