nidhinjs/prompt-master

This repo is a Claude Skill — a plugin you upload into Claude.ai or Claude Code — that acts as an expert prompt-writer. Instead of you crafting a prompt, you describe your goal in plain language ("write me a Cursor prompt to refactor my auth module") and the skill returns a ready-to-paste, purpose-built prompt for that specific tool.

Why it exists

The README's core complaint: most AI users waste money through trial-and-error prompting — vague prompt → bad output → re-prompt → closer → re-prompt again. It wants every word in a prompt to be "load-bearing," producing correct output on the first try.

How it works (8-step pipeline)

When you ask it for a prompt, the skill:

1. Detects the target tool (Claude, ChatGPT, Midjourney, Cursor, etc., from ~30 built-in profiles)
2. Extracts 9 dimensions of intent — task, input, output, constraints, context, audience, memory, success criteria, examples
3. Asks at most 3 clarifying questions if something critical is missing
4. Auto-selects a prompt framework from 14 templates (e.g., CO-STAR for business writing, Visual Descriptor for image tools, File-Scope for code editors)
5. Applies only "safe" prompt techniques — role assignment, few-shot examples, XML structure, grounding anchors, auditable reasoning. It deliberately excludes methods like Tree-of-Thought that it claims cause unpredictable output
6. Checks model recency against provider docs when the task depends on a "latest" model
7. Audits prompt length for token efficiency
8. Outputs the final prompt in one copyable block with a one-line strategy note

Notable contents

• Tool profiles: 30+ entries, each listing what the skill fixes for that tool — e.g., for Midjourney it enforces comma-separated descriptors plus --ar/version flags and negative prompts; for coding agents it enforces file paths, stop conditions, and "do-not-touch" lists.
• Universal Fingerprint: a 4-question fallback that writes prompts for any tool it has never seen.
• 37 anti-pattern catalog: a detailed table of common prompt mistakes, each with a before/after example (vague verbs, missing success criteria, no file scope, ungrounded factual requests, etc.).
• Memory Block system: extracts prior decisions from conversation history so the AI never contradicts earlier choices.

Example output (image tool)

For "realistic samurai standing in the rain at night": it produces a compact, comma-separated descriptor with camera framing, lighting, --ar 16:9 --v 6, and a negative prompt — roughly 60 tokens, versus a prose paragraph.

Installation

Download the repo as a ZIP and upload it via claude.ai → Sidebar → Customize → Skills, or clone it into ~/.claude/skills/prompt-master for Claude Code. It's invoked naturally ("Write me a prompt for...") or explicitly with /prompt-master.

All claims are drawn directly from the README; note that the README references model names like GPT-6 and Grok 4.6 that may not reflect the latest shipping models.