A prompt is an instruction given to a LLM for completion/inference. It can be anything from a direct question to a complex instruction to complete a complicated task.
Since this prompt is added to the Context along with any system prompts and Retrieval text, the contents and even the order of the contents can influence results substantially.
Some best practices around Prompt Structure have evolved over the years, with a whole discipline called Prompt Engineering.
When working in an Agent Harness like an IDE, TUI, or desktop application, it is common to wrap up commonly used prompts as Slash Commands.