One-shot

One-shotting a task means the first prompt produces the finished thing: no clarifying questions, no second turn, no 'actually, also…'. It is a claim about the prompt and the task together. The prompt was complete enough, and the task small and well-trodden enough, that the agent could go from description to working result without you in the loop.

Don't confuse it with the prompt-engineering meaning, where one-shot means a prompt that includes exactly one worked example (next to zero-shot and few-shot). In agentic coding the word is about the number of rounds, not the number of examples. When someone says a feature 'one-shotted', they mean it landed on the first try.

What makes a prompt one-shottable is that it reads like a spec: the stack, the files to touch, the acceptance criteria, the constraints, the things not to do. Vague prompts get a plausible guess; complete prompts get the thing. That makes one-shot a useful benchmark for your prompting (the 'Can I Vibe Code It?' prompts on this site are written to be one-shottable) and a poor default for real work, where iterating with human review between rounds catches what a single pass can't. One-shot the small, boring pieces; iterate on the ones that matter.