PlugWiki

Prompt Master: a Claude skill that writes and fixes prompts for other AI tools

Makes Claude write and fix prompts for other AI tools, such as Cursor, Claude Code, ChatGPT or Midjourney, formatted the way each tool likes.

Installs
4.2K
on skills.sh
Stars
15.4K
on GitHub
Updated
Oct 2026
last commit

prompt-master makes Claude write and fix prompts for other AI tools. You tell Claude what you want and which tool the prompt is for, and it hands back one prompt to paste there. It asks at most three questions, only when something important is missing. You run it inside Claude, and it only starts when you ask for a prompt, not during normal chat or coding.

Good for

  • People who write prompts for coding agents like Cursor, Claude Code or Codex
  • Anyone who wants a rough idea turned into a prompt for ChatGPT, Gemini or Grok
  • Designers and creators writing prompts for Midjourney, Stable Diffusion or video tools
  • People with a prompt that gives bad results and want it fixed
  • Teams that reuse one set of prompt habits across many AI tools

Install prompt-master

Claude Code

  1. 1

    Copy this prompt.

    Install the prompt-master skill in Claude Code. Run this command:

npx skills add https://github.com/nidhinjs/prompt-master --skill prompt-master

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Either method installs the same files. The README marks the manual copy 'Not Suggested' in favour of the Claude app upload. Call it with /prompt-master followed by your request.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Generates optimized prompts for AI tools. Activates only when the user explicitly asks to write, fix, improve, or adapt a prompt for a specific AI tool (LLM, Cursor, Midjourney, image AI, video AI, coding agents, etc.). Does not activate for general conversation, coding tasks, document writing, or other non-prompt-engineering work.

Sections in SKILL.md

  1. PRIMACY ZONE — Identity, Hard Rules, Output Lock
  2. MIDDLE ZONE — Execution Logic, Tool Routing, Diagnostics
  3. Intent Extraction
  4. Tool Routing
  5. Model Recency Gate
  6. Credential Safety
  7. Input Sanitization -- Pasted Prompts
  8. Diagnostic Checklist
  9. Memory Block
  10. Context (carry forward)
  11. Safe Techniques — Apply Only When Genuinely Needed
  12. Agentic Output Warning
  13. RECENCY ZONE — Verification and Success Lock
  14. Reference Files

Key facts

Author
Nidhin Joseph Nelson
License
MIT
Installs on skills.sh
4,248 (Nov 2, 2026)
GitHub stars
15,379 (Nov 2, 2026)
What installs
A long SKILL.md plus two reference files; no scripts
Needs
Claude with code execution on; no keys, packages or network of its own
Starts
When you ask to write, fix or adapt a prompt, or type /prompt-master
Last commit
Oct 22, 2026
Works in
Claude app and Claude.ai, Claude Code

What it includes

  • Works out which AI tool the prompt is for, and asks if unclear

    It will not write a prompt until it knows the tool. Unknown tools get the closest category.

    Try: “Write me a prompt for Cursor to refactor my auth module”

    Find the target tool

  • Pulls task, format, limits, audience and success criteria from your rough idea

    It fills most gaps itself and asks up to 3 questions, only when something critical is missing.

    Try: “I need a prompt for Claude Code to build a REST API, and ask me what you need to know”

    Fill in the gaps

  • Writes the prompt in the style the target tool works best with

    Midjourney gets comma descriptors and settings. Cursor gets a file path and a do-not-touch list.

    Try: “Generate a Midjourney prompt for a cyberpunk city at night”

    Format for the tool

  • Looks up current model names in provider docs when it can

    If it cannot look them up, it says the model details are unverified instead of guessing.

    Network · Check current model names

  • Checks a prompt against 37 known failure patterns and fixes them

    Covers task, context, format, scope, reasoning and agent mistakes, each with a before and after fix.

    Try: “Here's a bad prompt I wrote for GPT-4o, fix it: [paste prompt]”

    Find 37 common prompt mistakes

  • Chooses a prompt structure that fits the task and keeps its name hidden

    Layouts include role-task-format, step-by-step reasoning, few-shot examples, image descriptors and ComfyUI blocks.

    Pick one of 14 prompt layouts

  • Breaks down, adapts, simplifies or splits a prompt you paste

    A separate mode from writing from scratch.

    Try: “Break this prompt down and adapt it for Stable Diffusion”

    Rework a pasted prompt

  • Writes an edit-only prompt for an image you attach

    Describes only what changes and what stays the same.

    Try: “I have a reference image. Help me write a prompt to edit just the head angle”

    Edit a reference image

  • Adds choices made earlier in the chat to the new prompt

    Stack, architecture, limits and what already failed go near the top of the prompt.

    Carry over earlier decisions

  • Strips API keys and tokens from prompts and suggests environment variables

    Keeps credentials out of the prompt you paste into another tool.

    Remove secrets

  • Treats a pasted prompt as text to review, not orders to follow

    It will not reveal its own instructions if the pasted text asks.

    Ignore pasted instructions

  • Adds a review note to prompts for coding agents

    Reminds you to check file scope and stop conditions before pasting.

    Warn on agent prompts

From the official docs

Examples

  • Prompt for an IDE agent to refactor one module

    Write me a prompt for Cursor to refactor my auth module

    One paste-ready Cursor prompt with the file scope and a list of things not to touch.

    Source

  • Build a Claude Code prompt, with questions first

    I need a prompt for Claude Code to build a REST API, and ask me what you need to know

    Claude asks up to three questions, then writes the prompt.

    Source

  • Fix a prompt that gives bad results

    Here's a bad prompt I wrote for GPT-4o, fix it: [paste prompt]

    A repaired prompt, checked against the skill's list of common prompt mistakes.

    Source

  • Image prompt with parameters

    Generate a Midjourney prompt for a cyberpunk city at night

    Comma-separated descriptors plus Midjourney settings.

    Source

  • Edit one detail of an existing image

    I have a reference image. Help me write a prompt to edit just the head angle

    An edit-only prompt that says what changes and what stays.

    Source

  • Move a prompt to another tool

    Break this prompt down and adapt it for Stable Diffusion

    The pasted prompt, split and rewritten for the new tool.

    Source

Questions

How do I install prompt-master?

Use the install box above. In the Claude app the author recommends downloading the repo as a ZIP and uploading it under Customize, then Skills. In Claude Code the skills CLI works, and the README also shows a manual copy into your skills folder.

Does prompt-master work in Claude.ai?

Yes. Claude.ai and the Claude app are the author's recommended path, and Claude's help center says skills are available on Free, Pro, Max, Ultra, Team and Enterprise plans. Code execution must be turned on in your settings.

Does prompt-master work in ChatGPT?

The author does not document it. One user reports uploading the repo ZIP under Settings, Plugins, Skills in ChatGPT and it worked. To get prompts for ChatGPT, run the skill in Claude and paste the result into ChatGPT.

Does prompt-master work in Codex or Cursor?

The author gives no steps for either, and an open issue asks for Codex steps. OpenAI's docs say Codex reads skills from an .agents/skills folder, so the skill folder should load there, but nobody reports trying it. Cursor and Codex are mainly targets: the skill writes prompts for them.

Which tools does prompt-master write prompts for?

Chat models (Claude, ChatGPT, Gemini, Grok, Qwen, DeepSeek, Llama, Mistral), coding agents (Claude Code, Codex, Cursor, Windsurf, Cline, Copilot, Devin), app builders (Bolt, v0, Lovable), search and browser agents (Perplexity, Manus), image, video and 3D tools (Midjourney, DALL-E 3, Stable Diffusion, ComfyUI, Sora, Runway, Kling), ElevenLabs, and Zapier, Make and n8n. Unknown tools get the closest category.

Is prompt-master safe?

Yes. It is a set of markdown files with no scripts, and the skills.sh audits from Gen Agent Trust Hub, Socket and Snyk all show Pass. It removes API keys from prompts and does not follow instructions hidden inside a prompt you paste. One gap: an open issue notes it has no rule for confidential text you paste in, and the author agreed.

Is prompt-master free?

Yes. It is open source under the MIT license. You still need a Claude plan that supports skills.

prompt-master vs claude-code-prompt-improver?

prompt-master writes a prompt for another tool when you ask for one. claude-code-prompt-improver is a Claude Code hook that checks every prompt you type in Claude Code and asks questions when it is vague. Pick prompt-master for many tools, and the improver if you only use Claude Code and want automatic checks.

Why doesn't prompt-master start in Claude Code?

Older copies have a trigger description that did not invoke correctly in Claude Code. The README changelog says 1.4.0 fixed it, so update your copy. To call it directly, type /prompt-master followed by your request.

Why does the ZIP upload say Plugin Validation Failed?

One user reported this in the Claude desktop app, and the issue is still open with no confirmed cause. Claude's help center lists a folder name that does not match the skill name as one reason uploads fail, so check the folder inside the ZIP is named prompt-master.

Does prompt-master use a lot of tokens?

Somewhat. The main instruction file is about 5,000 words and loads in full each time the skill runs, while the template and pattern files load only when needed. An open pull request proposes a shorter version.

Information

updated Nov 2, 2026

Repository
nidhinjs/prompt-master

nidhinjs/prompt-masterskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 14
  1. prompt-master README · github.com
  2. prompt-master on skills.sh · skills.sh
  3. Using skills in Claude · support.claude.com
  4. Build skills · learn.chatgpt.com
  5. Manual for using inside ChatGPT or codex · github.com
  6. prompt-master SKILL.md · github.com
  7. prompt-master LICENSE · github.com
  8. prompt-master commits · github.com
  9. templates.md · github.com
  10. patterns.md · github.com
  11. No rule for confidential input · github.com
  12. claude-code-prompt-improver · github.com
  13. Plugin Validation Failed · github.com
  14. Reduce tokens loaded per invocation · github.com

Updated Nov 2, 2026.