Dry run

A dry run has an agent report exactly what it would do without doing it, so you can inspect the planned changes, catch mistakes, and approve before any real action or side effect happens.

A dry run has an agent report exactly what it would do without doing it, so you can inspect the planned changes, catch mistakes, and approve before any real action or side effect happens. You tell the agent to plan the work and print the plan rather than execute it. Say you ask an assistant to clean up old files in a shared folder. In dry-run mode it lists every file it intends to delete and every one it would keep, and stops there. You read the list, notice it flagged a document you still need, correct the instruction, and only then let it run for real. The planned actions stay on paper until you say go.

Builder example

A dry run is your cheapest review step for any agent that changes things people depend on, such as a folder cleaner, a bulk email sender, or a deploy script. Have the agent describe its intended actions as a plain list first: which records it would touch, which it would skip, and what each change does. You catch the wrong-folder mistake or the duplicate-email mistake while it is still a printout, before a single account, file, or recipient is affected.

Common confusion: A dry run is a preview of intended actions, not a partial run that does a little and stops. Nothing is changed and no side effect happens; the agent reports the full plan so you can approve it before any real execution begins.