TechSetupGuides
Beginneraiprompt-engineeringclaudeskillsllmproductivity

Prompt Master: AI prompt engineering skill

Generate optimized prompts for any AI tool using Claude skills. Zero wasted tokens, full context retention, works with 20+ AI tools.

  1. Step 1

    What is Prompt Master?

    Prompt Master is a Claude skill that generates optimized, production-ready prompts for any AI tool. It writes accurate prompts on the first try — zero tokens or credits wasted on re-prompting. The skill includes support for 20+ AI tools including Claude, ChatGPT, Gemini, o1/o3, Midjourney, Stable Diffusion, Cursor, Claude Code, and many more.

    The key insight: 'The best prompt is not the longest. It's the one where every word is load-bearing.' Unlike typical prompt generators that add verbosity, Prompt Master makes prompts sharper through tool-specific optimization, pattern detection, and token efficiency.

  2. Step 2

    Technology stack

    Prompt Master is built as a Claude AI skill — a pure markdown-based architecture with no dependencies:

    Core Architecture:

    • Type: Claude Skill (text-based skill file)
    • Format: SKILL.md with YAML frontmatter + markdown body
    • Structure: PAC2026 positional structure (30/55/15 attention zones)
    • Version: 1.6.0 (latest)

    Components:

    • SKILL.md - Main skill definition with identity, hard rules, and output format
    • references/templates.md - 12 prompt templates for different use cases
    • references/patterns.md - 35 credit-killing patterns with before/after fixes

    Supported AI Tool Categories:

    • Reasoning LLMs: Claude, ChatGPT/GPT-5.x, Gemini 2.x, o3/o4-mini, Qwen 2.5/3, DeepSeek-R1, MiniMax
    • Agentic AI: Claude Code, Cursor, Windsurf, Cline, GitHub Copilot, Devin, Manus
    • Image AI: Midjourney, DALL-E 3, Stable Diffusion, ComfyUI, SeeDream
    • Video AI: Sora, Runway, LTX Video, Dream Machine, Kling
    • 3D AI: Meshy, Tripo, Rodin, Unity AI, BlenderGPT
    • Voice AI: ElevenLabs
    • Workflow AI: Zapier, Make, n8n
    • Research AI: Perplexity, SearchGPT

    Key Technologies:

    • No dependencies or installations required
    • Pure markdown-based knowledge system
    • YAML frontmatter for metadata (name, version, description, activation triggers)
    • XML structural tags for complex multi-section prompts
    • PAC2026 attention architecture for optimal model parsing
    prompt-master/
├── SKILL.md (main skill definition)
│   ├── PRIMACY ZONE (30%): Identity, hard rules, output format
│   ├── MIDDLE ZONE (55%): Execution logic, tool routing, diagnostics
│   └── RECENCY ZONE (15%): Verification, success criteria
├── references/
│   ├── templates.md (12 prompt templates)
│   └── patterns.md (35 credit-killing patterns)
├── README.md (documentation)
└── LICENSE (MIT)

Templates:
├── RTF (Role, Task, Format)
├── CO-STAR (Context, Objective, Style, Tone, Audience, Response)
├── RISEN (Role, Instructions, Steps, End Goal, Narrowing)
├── CRISPE (Capacity, Role, Insight, Statement, Personality, Experiment)
├── Chain of Thought (logic/math/debugging)
├── Few-Shot (pattern replication)
├── File-Scope (Cursor, Windsurf, Copilot)
├── ReAct + Stop Conditions (Claude Code, Devin)
├── Visual Descriptor (image generation)
├── Reference Image Editing
├── ComfyUI (node-based workflows)
├── Prompt Decompiler
└── Opus 4.7 Task Brief (complex agentic)
  3. Step 3

    Prerequisites

    Before using Prompt Master, you need:

    • Access to Claude.ai (free tier works) or Claude Code application
    • Understanding of basic prompt concepts (optional but helpful)
    • Knowledge of which AI tool you want to generate prompts for
  4. Step 4

    Installation - Claude.ai (Recommended)

    The recommended way to install Prompt Master is through Claude.ai browser interface. This activates the skill within your Claude conversations.

    # 1. Download the repository as a ZIP
# Go to: https://github.com/nidhinjs/prompt-master
# Click the green "Code" button and download ZIP

# 2. Upload to Claude.ai
# Navigate to claude.ai
# Click the sidebar menu → Customize → Skills → Upload a Skill
# Upload the SKILL.md file from the downloaded ZIP

# The skill will activate automatically when you mention prompt generation
  5. Step 5

    Installation - Claude Code (Alternative)

    For advanced users, you can clone Prompt Master directly into the Claude Code skills directory. This is not recommended for most users as it may not trigger properly.

    # Clone directly into Claude Code skills directory
mkdir -p ~/.claude/skills
git clone https://github.com/nidhinjs/prompt-master.git ~/.claude/skills/prompt-master

# Verify installation
ls ~/.claude/skills/prompt-master/

# Files you should see:
# SKILL.md    README.md    LICENSE    references/
  6. Step 6

    Basic usage

    Once installed, invoke Prompt Master naturally in conversation. The skill activates silently and generates optimized prompts based on your request:

    The skill follows a 7-step pipeline:

    1. Detects the target AI tool from your request
    2. Extracts 9 dimensions of intent (task, input, output, constraints, context, audience, memory, success criteria, examples)
    3. Asks targeted clarifying questions if critical info is missing (max 3)
    4. Routes to the correct prompt framework automatically
    5. Applies safe techniques only (role assignment, few-shot, XML structure, grounding anchors)
    6. Runs a token efficiency audit to remove non-essential words
    7. Delivers one clean copyable prompt with a one-line strategy note
    # Simple examples:

"Write me a prompt for Cursor to refactor my auth module"

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

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

"Generate a Midjourney prompt for a cyberpunk city at night"

"I have a reference image — help me write a prompt to edit just the head angle"

"Break this prompt down and adapt it for Stable Diffusion"

Or explicitly invoke:

"/prompt-master

I want to ask Claude Code to build a todo app with React and Supabase"
  7. Step 7

    Example: Midjourney prompt generation

    Here's how Prompt Master transforms a simple request into an optimized Midjourney prompt:

    User input: "Write me a midjourney prompt for a realistic samurai standing in the rain at night"

    Generated output: The skill converts natural language into comma-separated descriptors with parameters, negative prompts, and aspect ratios — Midjourney's native format. It avoids prose descriptions which perform poorly in Midjourney.

    Input:
"Write me a midjourney prompt for a realistic samurai standing in the rain at night"

Generated Prompt:
lone samurai standing in heavy rain at night, traditional armor,
neon reflections on wet cobblestone street, cinematic lighting,
dramatic shadows, fog, ultra detailed, photorealistic,
shallow depth of field --ar 16:9 --v 6 --style raw

negative: blurry, low quality, watermark, cartoon, anime, extra limbs

Target: Midjourney
Strategy: Comma-separated descriptors over prose, 
lighting and mood anchored early, aspect ratio and version locked, 
negative prompt prevents style drift.
  8. Step 8

    Example: Claude Code prompt generation

    For coding agents like Claude Code, Prompt Master generates structured prompts with explicit file scopes, stop conditions, and technical specifications. This prevents runaway agents and credit waste.

    Input:
"Build a claude code prompt for a landing page for a business dashboard 
that looks and feels exactly like notion - smooth animations, clean ui"

Generated Prompt:
Objective:
Build a fully detailed, production-quality static landing page for a 
business dashboard product. Design must match Notion's exact aesthetic —
clean, minimal, typographic, with smooth animations throughout.

Stack:
Single HTML file with embedded CSS and vanilla JS. No frameworks.
No external dependencies except Google Fonts (Inter).

Design Spec — match these exactly:
- Background: #ffffff
- Primary text: #1a1a1a
- Secondary text: #6b7280
- Border color: #e5e7eb
- Accent: #000000
- Font: Inter, weights 400 / 500 / 600 / 700
- Base spacing unit: 8px
- Border radius: 6px on cards, 4px on buttons

Strategy: Every vague Notion aesthetic cue translated into exact hex values 
and pixel specs — Claude Code cannot guess wrong. Animations defined with 
exact timing, method, and trigger so there is no interpretation needed.
  9. Step 9

    Tool-specific routing

    Prompt Master auto-routes to different frameworks based on the target tool:

    Reasoning LLMs (Claude, ChatGPT, Gemini):

    • Explicit output contracts
    • XML structural tags for complex prompts
    • Verbosity control and completion criteria

    Thinking Models (o1/o3/o4-mini, DeepSeek-R1, Qwen3 thinking):

    • Short clean instructions only
    • NEVER adds Chain of Thought (they think internally)
    • Zero-shot preferred, few-shot only when strictly needed

    IDE Agents (Cursor, Windsurf, Copilot):

    • File-Scope Template (file path + function name + current behavior + desired change)
    • Do-not-touch lists
    • Sequential prompt guidance for complex tasks

    Autonomous Agents (Claude Code, Devin, Cline):

    • ReAct + Stop Conditions Template
    • Starting state + target state + allowed/forbidden actions
    • Human review triggers for destructive operations

    Image AI (Midjourney, DALL-E, Stable Diffusion):

    • Visual Descriptor format
    • Positive/negative prompt split
    • Tool-specific parameters (--ar, --v for Midjourney, weights for SD)
    Tool Routing Examples:

Claude / Claude 4.x:
├── Be explicit and specific
├── XML tags help for complex prompts
├── Provide context and reasoning WHY
└── Specify output format and length

o3 / o4-mini (reasoning models):
├── SHORT clean instructions ONLY
├── NEVER add CoT instructions
├── Prefer zero-shot first
└── Keep system prompts under 200 words

Cursor / Windsurf (IDE AI):
├── File path required
├── Function/component name
├── Current behavior description
├── Desired change specification
└── Do-not-touch list

Midjourney (image):
├── Comma-separated descriptors
├── Subject → style → mood → lighting
├── Parameters at end: --ar --v --style
└── Negative prompts via --no
  10. Step 10

    Credit-killing patterns detection

    Prompt Master detects and fixes 35 common patterns that waste tokens and credits:

    Task Patterns (7): Vague verbs, two tasks in one prompt, no success criteria, over-permissive agents, emotional descriptions, build-the-whole-thing, implicit references

    Context Patterns (6): Assumed prior knowledge, no project context, forgotten stack, hallucination invites, undefined audience, no mention of prior failures

    Format Patterns (6): Missing output format, implicit length, no role assignment, vague aesthetic adjectives, no negative prompts (image AI), prose for Midjourney

    Scope Patterns (6): No scope boundary, no stack constraints, no stop conditions, no file path (IDE AI), wrong template for tool, pasting entire codebase

    Reasoning Patterns (5): No CoT for logic tasks, adding CoT to reasoning models, no self-check, expecting inter-session memory, contradicting prior decisions

    Agentic Patterns (5): No starting state, no target state, silent agent, unlocked filesystem, no human review trigger

    Pattern Examples:

Vague task verb:
Before: "help me with my code"
After:  "Refactor `getUserData()` to use async/await and handle null returns"

Two tasks in one prompt:
Before: "explain AND rewrite this function"
After:  Split into two separate prompts: explain first, then rewrite

No success criteria:
Before: "make it better"
After:  "Done when function passes existing unit tests and handles null input"

Prose prompt for Midjourney:
Before: "A beautiful portrait of a woman standing in the rain"
After:  "woman portrait, rain, wet hair, dramatic lighting, cinematic, 
        ultra detailed, photorealistic --ar 4:5 --v 6 --style raw"

Adding CoT to reasoning models:
Before: "Think step by step" (sent to o3/o4-mini)
After:  Removed entirely — o3 reasons internally across thousands of tokens
  11. Step 11

    Memory Block system

    For conversations with history, Prompt Master includes a Memory Block to maintain context across prompts. This is critical — most wasted re-prompts come from AI forgetting prior decisions.

    The Memory Block captures:

    • Stack and technology decisions
    • Architecture choices locked in
    • Constraints from prior turns
    • What was attempted and failed
    • Component naming conventions
    • Design system specifications
    Memory Block Example:

## Memory (Carry Forward from Previous Context)
- Stack: React 18 + TypeScript + Supabase
- Auth uses JWT stored in httpOnly cookies, not localStorage
- Component naming convention: PascalCase, no default exports
- Design system: Tailwind only, no custom CSS files
- Architecture: no Redux, context API only
- Database: Supabase PostgreSQL with Row Level Security
- API pattern: Server Components with fetch caching

This Memory Block is prepended to all subsequent prompts in the session,
ensuring the AI never contradicts earlier decisions.
  12. Step 12

    12 prompt templates

    Prompt Master includes 12 specialized templates for different scenarios. The skill auto-selects the appropriate template based on your task:

    | Template | Best For | |----------|----------| | RTF (Role, Task, Format) | Simple one-shot tasks | | CO-STAR | Professional documents, business writing | | RISEN | Complex multi-step projects | | CRISPE | Creative work, brand voice | | Chain of Thought | Math, logic, debugging | | Few-Shot | Consistent structured output | | File-Scope | Cursor, Windsurf, Copilot | | ReAct + Stop Conditions | Claude Code, Devin | | Visual Descriptor | Midjourney, DALL-E, Stable Diffusion | | Reference Image Editing | Editing existing images | | ComfyUI | Node-based image workflows | | Prompt Decompiler | Breaking down existing prompts | | Opus 4.7 Task Brief | Complex multi-step on Claude Opus 4.7 |

    Template A - RTF (Role, Task, Format):
Role: You are a senior technical writer.
Task: Write a one-paragraph description of what a REST API is.
Format: Plain prose, 3 sentences maximum, no jargon.

Template B - CO-STAR:
Context: I'm a founder pitching a B2B SaaS tool.
Objective: Write a cold email that gets a reply from a CFO.
Style: Direct and conversational, not salesy.
Tone: Confident but not pushy.
Audience: CFO at a 200-person company, busy, skeptical.
Response: 5 sentences max. Subject line included. No bullet points.

Template G - File-Scope (IDE AI):
File: src/auth/login.tsx
Function: handleLogin

Current Behavior:
Accepts email/password, calls /api/login, stores JWT in localStorage

Desired Change:
Use httpOnly cookies instead of localStorage. Add CSRF token validation.

Do Not Touch:
- Any files outside src/auth/
- The database schema
- Existing tests

Done When:
Returns 401 on invalid credentials without exposing error details.
  13. Step 13

    Safe techniques applied

    Prompt Master only uses techniques with reliable, bounded effects. It explicitly excludes methods known to produce hallucinations or unpredictable output (Tree of Thought, Graph of Thought, Universal Self-Consistency, prompt chaining).

    Safe techniques applied when needed:

    • Role Assignment: Assigns specific expert identity to calibrate depth
    • Few-Shot Examples: Adds 2-5 examples when format consistency matters
    • XML Structural Tags: Wraps sections for Claude-based tools
    • Grounding Anchors: Adds anti-hallucination rules for factual tasks
    • Chain of Thought: Forces step-by-step reasoning for logic tasks (never on o1/o3/DeepSeek-R1/Qwen3-thinking)
    Technique Application:

Role Assignment (for specialized tasks):
Weak: "You are a helpful assistant"
Strong: "You are a senior backend engineer specializing in 
        distributed systems who prioritizes correctness over cleverness"

Grounding Anchors (for factual tasks):
"Use only information you are highly confident is accurate. 
If uncertain, write [uncertain] next to the claim. 
Do not fabricate citations or statistics."

Few-Shot Examples (2-5 examples for format lock):
When format is easier to show than describe, examples 
outperform written instructions every time.
  14. Step 14

    Advanced usage patterns

    Beyond simple prompt generation, Prompt Master supports several advanced patterns:

    Prompt Decompiler: Paste an existing prompt and have it broken down, simplified, or adapted for a different tool.

    Reference Image Editing: When you have an existing image, Prompt Master detects this and builds prompts focused on delta-only changes.

    Cross-Tool Adaptation: Take a prompt designed for one tool (e.g., Midjourney) and adapt it for another (e.g., Stable Diffusion) while maintaining intent.

    Session-Aware Prompting: The Memory Block ensures prompts build on prior decisions without re-providing context each time.

    Advanced Examples:

# Prompt Decompiler
"Here's a complex prompt I found online. Break it down, explain each component,
and adapt it for Claude Code instead of Midjourney: [paste prompt]"

# Reference Image Editing
"I have a portrait photo. Write a prompt to change just the background
to a beach at sunset while keeping the person exactly the same."

# Cross-Tool Adaptation
"I have this Midjourney prompt that works great. Adapt it for Stable
diffusion 1.5 with proper weight syntax and negative prompt format: [prompt]"

# Multi-Prompt Complex Task
"Break this large project into sequential prompts I can feed to Cursor:
Build a full-stack authentication system with JWT, email verification,
password reset, and OAuth with Google."
  15. Step 15

    Version history and updates

    Prompt Master is actively maintained with regular updates:

    1.6.0 (latest) — Opus 4.7 update, added Template M for Opus 4.7 Task Brief, updated Claude routing for literalism and session hygiene, added patterns 36-37

    1.5.0 — Added more tool routing, new Agentic AI and 3D Model AI support, removed token estimates from output, added instruction layer

    1.4.0 — Added reference image editing detection, ComfyUI support, Prompt Decompiler mode, 3 new templates

    1.3.0 — Rebuilt around PAC2026 positional structure, silent routing replaces user-facing framework selection, references folder introduced

    Version History:

v1.6.0 (latest)
├── Opus 4.7 support
├── Template M (Opus 4.7 Task Brief)
├── Updated Claude Code routing
├── Patterns 36-37 added
└── Session hygiene improvements

v1.5.0
├── Agentic AI routing expansion
├── 3D Model AI support
├── Token estimate removal
└── Copywriting placeholders

v1.4.0
├── Reference image editing
├── ComfyUI templates
├── Prompt Decompiler mode
└── 3 new templates
  16. Step 16

    Resources

    Additional resources for using and understanding Prompt Master:

    GitHub Repository: https://github.com/nidhinjs/prompt-master

    Documentation: README.md in the repository contains full documentation

    Skill Definition: SKILL.md contains the complete skill logic

    Template Reference: references/templates.md has all 12 templates

    Pattern Reference: references/patterns.md documents all 35 patterns

    Claude Documentation: Learn about skills at https://docs.anthropic.com/en/docs/clients/skills

    Links:
GitHub: https://github.com/nidhinjs/prompt-master
Stars: 8.2k as of May 2025
License: MIT
Author: nidhinjs

Documentation:
README: Overview and examples in the repo
SKILL.md: Complete skill implementation
references/templates.md: All 12 prompt templates
references/patterns.md: All 35 credit-killing patterns

Claude Skills Guide: https://docs.anthropic.com/en/docs/clients/skills

Community:
Discussions: GitHub repo Discussions tab
Issues: Report bugs or request tool additions

Feature requests

Sign in to suggest features or vote on existing ones.

No feature requests yet.

Discussion

0 people marked this as worked·Sign in to mark your own.

Sign in to join the discussion.

No comments yet.