hubberdubber/reword-prompt-skill

{"i": 0, "repo": "hubberdubber/reword-prompt-skill", "stars": 2, "lang": null, "ok": true, "char_count": 1850, "text": "This repository is a Claude Skill called reword-prompt. A skill is an add-on that changes how Claude, the AI assistant from Anthropic, responds to your requests. This particular skill takes a rough, incomplete prompt and turns it into a single polished prompt you can paste and run immediately.\n\nMost tools that improve prompts give you multiple options, commentary, and explanations. This one does not. It gives you exactly one finished prompt with nothing before or after it. It detects what kind of AI system the prompt is meant for, such as a reasoning model, a coding agent, an image generator, or a research tool, then structures the prompt to match that target system.\n\nThe skill tries to keep the important words from your original input exactly as you wrote them. It fills in gaps in method, scope, and structure with sensible defaults, but it does not invent facts. If your rough prompt suggests a weak approach and a stronger one exists, it swaps in the better approach while keeping your original goal intact. It will ask you at most one clarifying question, and only when there is a genuine fork in what you want.\n\nTo install it on the Claude website or desktop and mobile apps, you download a .skill file from the Releases page and upload it through your Settings. To use it with Claude Code, you clone the repository and copy its folder into your local skills directory. You can customize a user profile file inside the skill by filling in bracketed placeholders, though the skill works fine even if you leave them blank.\n\nIn practice, you paste a rough prompt and tell Claude to reword it, sharpen it, or turn it into a real prompt. You can add hints like specifying a target system, requesting a shorter or longer output, asking for citations, or telling it to dry-run the prompt first. The skill is specifically for generating prompts aimed at AI systems, not for editing emails, prose, summaries, or translations meant for human readers.", "error": null}