Runprompt allows users to execute .prompt files from their shell using a simple Python script. The .prompt files contain prompts and metadata, enabling structured responses from large language models (LLMs). Users can quickly get started by downloading the script or running it directly with tools like uvx. Installation via pip is also an option. A sample prompt setup shows how to define a model and create a prompt that personalizes greetings based on input. The article explains various ways to run prompts, including passing data through standard input or command-line arguments. It highlights special variables like {{STDIN}} and {{ARGS}} to capture user input effectively. For more structured tasks, users can define schemas for extracting information, such as names and ages, and even process outputs between multiple prompts. This feature allows for a seamless flow of data, where the output of one prompt can feed directly into another. A key aspect of Runprompt is its ability to execute shell commands before sending prompts. This feature can gather dynamic context, with the results available as variables in the prompt. Users can also make .prompt files executable directly. Customization options let users override frontmatter values from the command line, attach local files, and define tools as Python functions that LLMs can call during execution. Safety measures are in place for tool interactions, allowing users to set certain functions as "safe" to bypass confirmation prompts. Overall, Runprompt combines flexibility with user control, enabling efficient interaction with LLMs in a shell environment. The system supports detailed configurations, file handling, and dynamic context gathering, making it a powerful tool for developers and data scientists looking to leverage LLMs in their workflows.