What is prompt-improver?

prompt-improver is a skill for AI coding assistants that turns a vague, quickly typed request into a precise, structured specification before any work is done. Instead of the main assistant grinding through your rough instructions directly, it hands the request off to a separate model call that rewrites it into a verifiable XML spec, then either runs that spec right away or shows it to you first for review. The generator only rewrites your request, it never implements the feature itself. You can run it with no flags to improve and execute immediately, or add plan to have it show you the rewritten spec first and wait before doing anything. You can also force a specific model to do the rewriting for a single run. It supports several different assistant CLIs as the generator, including Claude, Grok, Gemini, and Codex, each with its own default model and a fallback chain if that model is unavailable or rate limited. Because of this, you can mix and match: for example run the tool inside Claude Code but have Grok do the rewriting, as long as that other CLI is installed on your machine. If every generator option is limited, the tool falls back to letting the assistant you are already talking to finish the request itself. The process behind the scenes has four steps: it first checks whether your input is already a solid spec and skips rewriting if so, then generates the improved version, validates its structure, and finally either executes it or shows it to you depending on the mode you chose. Installation is done through the skills.sh tool, the Agent Skills format, or the Claude Code plugin marketplace, and there is also a standalone shell script version that works without any of those installers. Settings such as which model to use by default, custom fallback chains, and custom CLI commands can be configured in a local settings file. The project is released under the MIT license.