Takes a rough prompt and makes it more useful by applying LLM best practices. It pushes you to add verifiability (like test cases or screenshots), specify file paths and constraints, and separate exploration from execution when a task is complex. The workflow is methodical: it classifies your task type and complexity, flags anti-patterns like vague instructions or missing verification commands, then rewrites your prompt with concrete examples and questions about what's missing. Honestly most valuable when you're about to ask Claude to "fix the login bug" and realize you should probably mention the error message and which files are involved. Includes task-specific templates for bugfixes, refactors, research, and ops work.
npx skills add https://github.com/dotneet/claude-code-marketplace --skill prompt-improver