Start a loop with a defined outcome
Running the orbit means keeping the question, the context, and the intended result connected. Begin with a problem that can be described and an outcome that can be recognized. If the aim is to fix a layout, identify the page and the screen where it fails. If the aim is to shape a story, identify the audience and what the reader should understand. Without that center, the work can expand indefinitely while missing the point. A clear outcome gives every later step something concrete to return to.
Clarify before making a destructive change
The next stage is to check what is known and what is missing. Protect approved images, copy, structure, and identity unless the request specifically unlocks them. Ask a question when an essential detail cannot be recovered, but do not ask people to re-upload a file that is already available. The purpose of clarification is to prevent unnecessary damage, not to postpone useful work. A small, well-scoped change is often more effective than replacing the entire system. Where a larger change is necessary, make its scope explicit before treating it as the new baseline.
Act, then test the result
A task is not complete because an edit was made. The result has to be checked in the environment that matters. For a site, that includes the relevant page, the viewport, the controls, and the actual media. For other work, the equivalent might be a factual check, a read-through, or an agreed acceptance test. Keep track of what was tested and what was not. A preview, a packaged build, and a live deployment are different stages. Passing one stage does not silently certify the next. Each handoff should preserve that distinction.
Close the loop without losing the thread
After verification, deliver the usable output and explain any meaningful limitation. If feedback reveals another problem, return to the same source and change only what is necessary. This is what keeps the work in one orbit rather than scattering it across disconnected versions. In the current beta, the public site provides the request entry and explanation; it does not itself guarantee an autonomous background agent or automatic completion of every stage. The operating idea is still practical: fewer repeated instructions, clearer responsibility, visible progress, and a final result that matches the original goal.
Question → context → operator → action → verification → useful result.