Prompt Master: prompts that land on the first attempt
Prompt Master targets the usual loop of a vague request, a wrong answer and several rounds of corrections. It first identifies which AI tool the prompt is for, extracts nine dimensions of intent, asks at most three clarifying questions when something critical is missing, and returns one copyable block with a one-line strategy note.
Detects the target tool and adapts to it: stop conditions and file scope for Claude Code, comma-separated descriptors and parameters for Midjourney.
Silently picks one of 14 templates, including RTF, CO-STAR, RISEN and Few-Shot, without showing the framework name.
Applies five bounded techniques: role assignment, few-shot examples, XML tags, grounding anchors and auditable reasoning.
Decompiles an existing prompt and adapts it for another model, for instance moving a GPT request over to Stable Diffusion.
For a tool it has no profile for, asks the four Universal Fingerprint questions and builds the prompt from the answers.
When to use it
You need to hand Claude Code or Cursor a refactor and keep the agent away from files it should not touch.
You want an image from Midjourney or ComfyUI, but plain-language descriptions keep producing the wrong style.
A ChatGPT prompt underperforms and you want to see which words are dead weight and what is missing.
Ask your agent
Write me a prompt for Cursor to refactor my auth module without changing public functions.
Here is my GPT prompt that gives vague answers. Fix it.
Generate a Midjourney prompt for a realistic samurai in the rain at night, 16:9.
Questions
How do I install Prompt Master?
The author recommends downloading the repo as a ZIP and uploading it in claude.ai under Skills. For Claude Code use npx skills add nidhinjs/prompt-master@prompt-master or git clone into ~/.claude/skills/prompt-master.
Which AI tools can Prompt Master write for?
The README lists profiles for Claude, ChatGPT, Codex, Gemini, Cursor, GitHub Copilot, Midjourney, Stable Diffusion, Sora, ElevenLabs, Zapier and more. Unknown tools get a four-question universal fallback.
How is it different from other prompt generators?
Instead of making prompts longer, it cuts words that do not change the output. Techniques such as Tree of Thought are excluded by default because they raise the risk of fabricated answers.
