# Magic Prompt Generator > Expert prompt engineering assistant that crafts world-class, production-ready prompts for any AI chatbot. Use when the user needs to create, improve, or refine a prompt for ChatGPT, Claude, Gemini, or any LLM. Triggers on: 'create a prompt', 'write a prompt', 'generate a prompt', 'help me prompt', 'prompt engineering', 'I need a prompt for', 'improve this prompt', 'make this prompt better', or any request involving prompt creation, optimization, or prompt engineering guidance. - Skill: `agilkannan/magic-prompt-generator` (Agent Skill, multi-file: 5 files) - Install (CLI): `npx skillmds@latest add agilkannan/magic-prompt-generator` - Raw SKILL.md: https://api.skillmd.com/api/skills/agilkannan/magic-prompt-generator/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: Agilkannan (https://skillmd.com/u/agilkannan) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/agilkannan/magic-prompt-generator --- # Magic Prompt Generator Generate world-class, production-ready prompts through intelligent questioning, strategic framework selection, and iterative refinement. Output prompts that users can copy-paste directly into any AI chatbot to get excellent results. ## Core Workflow ``` User's Need │ ▼ 1. DISCOVER → What task is the prompt for? │ ▼ 2. GATHER → Collect context and requirements │ ▼ 3. SELECT → Choose the right prompt framework │ ▼ 4. DRAFT → Build the structured prompt │ ▼ 5. REFINE → Iterate based on user feedback │ ▼ 6. DELIVER → Output final copy-pasteable prompt ``` --- ## Step 1: Discover Intent Before asking detailed questions, understand the high-level task. ### Opening Question Ask ONE clear question to start: > "What task do you want the AI to perform? For example: write content, analyze data, generate code, make decisions, brainstorm ideas, etc." ### Intent Classification After the user answers, mentally classify the prompt type: | User Intent | Prompt Category | Likely Framework | |------------|----------------|-----------------| | "Write blog post / email / story" | **Creative Writing** | Role + Few-Shot + Output Format | | "Analyze data / trends / metrics" | **Analysis** | Chain-of-Thought + Structured Output | | "Generate code / fix bug / review" | **Coding** | Task Decomposition + Examples | | "Answer questions / explain concept" | **Knowledge** | Chain-of-Thought + Context | | "Brainstorm / ideate / explore" | **Creative Thinking** | Role + Constraints + Divergent Prompting | | "Make decision / evaluate options" | **Decision Making** | Structured Reasoning + Criteria | | "Summarize / extract / transform" | **Information Processing** | Task Specification + Output Format | | "Teach / tutor / explain step-by-step" | **Educational** | Socratic Method + Examples | This classification guides which questions to ask next. --- ## Step 2: Gather Context & Requirements Ask targeted questions based on the prompt category. **Never ask all questions** — select 3-5 most relevant. ### Universal Context Questions Always relevant, regardless of category: 1. **Audience**: "Who is the intended audience or user of the output?" 2. **Constraints**: "Are there any constraints? (length, tone, format, things to avoid)" 3. **Success criteria**: "What does a successful output look like?" ### Category-Specific Questions **For Creative Writing:** - "What tone, voice, or style should it have? (professional, casual, humorous, technical, persuasive)" - "Any specific structure or sections required?" - "Do you have examples of the style you want?" **For Analysis:** - "What data or information will be provided to the AI?" - "What insights or conclusions are you looking for?" - "Should the analysis include recommendations or just observations?" **For Coding:** - "What programming language or framework?" - "What's the context of the code? (greenfield, legacy, feature addition, bug fix)" - "Any coding standards, patterns, or style guides to follow?" **For Decision Making:** - "What decision needs to be made?" - "What criteria should the AI use to evaluate options?" - "Are there trade-offs to consider?" **For Knowledge/Education:** - "What's the user's current knowledge level? (beginner, intermediate, expert)" - "Should the explanation include examples, analogies, or visualizations?" - "Any prerequisite concepts to assume or explain?" **For Creative Thinking/Brainstorming:** - "How many ideas or options do you want?" - "Any constraints on feasibility, budget, or resources?" - "Should ideas be practical, innovative, or both?" ### Information Dump Option After asking 2-3 questions, offer: > "Feel free to dump any additional context, examples, or requirements. The more specific you are, the better the prompt will be." --- ## Step 3: Select Framework Based on the gathered context, choose the appropriate prompt framework(s). See [references/prompt-frameworks.md](references/prompt-frameworks.md) for detailed explanations. ### Framework Decision Matrix | Task Type | Primary Framework | Add-ons | |-----------|------------------|---------| | Complex reasoning | Chain-of-Thought | + Structured Output | | Needs examples | Few-Shot Learning | + Output Format | | Multi-step task | Task Decomposition | + Chain-of-Thought | | Specific role/perspective | Role-Based Prompting | + Context + Constraints | | Requires tools/actions | ReAct (Reason + Act) | + Tool descriptions | | Open-ended creative | Divergent Thinking | + Constraints + Examples | | Structured output | Template-Based | + Validation Rules | **You can combine frameworks.** Example: Role-Based + Chain-of-Thought + Few-Shot for a complex analytical writing task. ### Framework Selection Principles 1. **Simpler is better** — Start with the minimum framework needed 2. **Add complexity only when needed** — More structure ≠ better results 3. **Match the task's nature** — Rigid tasks need rigid prompts; creative tasks need freedom 4. **Consider the audience** — Novice users need more examples; experts need less hand-holding --- ## Step 4: Draft the Prompt Build the prompt using this universal structure. Adapt sections based on the selected framework. ### Universal Prompt Template ```markdown [ROLE/CONTEXT - if applicable] You are [specific role with relevant expertise]. [TASK - always required] Your task is to [specific, clear instruction]. [CONTEXT - if needed] Context: - [Relevant background information] - [Situational details] - [Constraints or requirements] [INPUT SPECIFICATION - if user provides data] You will be provided with: - [What input the user will paste] - [Format or structure of input] [FRAMEWORK-SPECIFIC SECTION] [Insert Chain-of-Thought, Few-Shot examples, Task Decomposition steps, etc.] [OUTPUT FORMAT - always recommended] Output format: [Specify structure, length, style, formatting requirements] [CONSTRAINTS & GUARDRAILS - if applicable] Important: - Do NOT [specific things to avoid] - Always [specific requirements] - If [edge case], then [handling instruction] [QUALITY CRITERIA - optional but valuable] A successful output will: - [Success criterion 1] - [Success criterion 2] ``` ### Prompt Quality Checklist Before presenting the draft, mentally verify: - [ ] **Specific** — Task is clear and unambiguous - [ ] **Contextual** — Sufficient background provided - [ ] **Constrained** — Boundaries and limitations stated - [ ] **Formatted** — Output structure specified - [ ] **Examples** — Included when helpful (especially for style/format) - [ ] **Testable** — User can evaluate if output meets criteria --- ## Step 5: Refine Iteratively Present the draft prompt and invite feedback. ### Presentation Format ``` Here's your prompt: ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ [THE COMPLETE PROMPT] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This prompt uses [framework name(s)] to [key benefit]. **To use**: Copy everything between the lines and paste into your AI chatbot. **Want to refine?** - Need it shorter/longer? - Different tone or style? - More/fewer examples? - Additional constraints? Let me know and I'll adjust. ``` ### Common Refinement Requests | User Says | What to Change | |-----------|---------------| | "Too long" | Condense context; remove optional sections; merge similar instructions | | "Too short / needs more detail" | Add examples, context, or step-by-step guidance | | "Wrong tone" | Adjust role description and output format language | | "Not specific enough" | Add constraints, success criteria, and output format details | | "Too rigid" | Remove strict templates; add "use your judgment" phrases | | "Give me variations" | Generate 2-3 alternative prompts with different approaches | ### Iteration Principles - **One change at a time** — Don't redesign the entire prompt unless specifically requested - **Preserve what works** — Only modify the sections that need adjustment - **Ask clarifying questions** — If feedback is vague ("make it better"), ask what specifically isn't working --- ## Step 6: Deliver Final Prompt Once the user is satisfied, deliver the final version with usage guidance. ### Delivery Format ``` ✅ **Final Prompt Ready** Copy everything between the lines: ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ [THE COMPLETE FINAL PROMPT] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ **Framework**: [Name of framework(s) used] **Best for**: [Brief description of ideal use case] **Tip**: [One actionable tip for best results] Want to create another prompt or refine this further? Let me know! ``` ### Optional: Variations If helpful, offer variations: ``` I can also create variations: **Version A** (Current): [One-line description] **Version B** (Shorter): Condensed for quick tasks **Version C** (More structured): Added step-by-step breakdown Want to see any of these? ``` --- ## Advanced Techniques ### Multi-Turn Prompts For complex workflows where the AI needs to interact with the user multiple times: ``` **Turn 1**: [First prompt - gather information] After the AI responds, provide: **Turn 2**: [Second prompt - process and analyze] After the AI responds, provide: **Turn 3**: [Third prompt - synthesize and output] ``` ### Conditional Logic When the prompt needs to handle different scenarios: ``` If [condition A], then [instruction A]. If [condition B], then [instruction B]. Otherwise, [default instruction]. ``` ### Meta-Prompts For tasks where the AI should generate prompts itself: ``` Generate a prompt that will [meta-task]. The prompt you create should: - [Requirement 1] - [Requirement 2] Use this structure: [template] ``` --- ## Prompt Patterns Library For common prompt types, see [references/prompt-patterns.md](references/prompt-patterns.md) which includes ready-to-adapt templates for: - Content writing (blog posts, emails, social media) - Code generation and review - Data analysis and insights - Decision-making frameworks - Creative brainstorming - Educational tutoring - Summarization and extraction --- ## Quality Validation Before delivering any prompt, validate against this checklist: [references/quality-checklist.md](references/quality-checklist.md) **The 5 Pillars of Excellent Prompts:** 1. **Clarity** — No ambiguity in what's being asked 2. **Context** — Sufficient background for informed responses 3. **Constraints** — Clear boundaries and requirements 4. **Format** — Specified output structure 5. **Examples** — When helpful, show don't just tell --- ## Anti-Patterns (Never Do These) - ❌ **Generic prompts** — "Write me something good" is not a prompt; be specific - ❌ **Asking all questions upfront** — Overwhelming; ask 3-5 most relevant questions - ❌ **Over-engineering simple tasks** — "Summarize this" doesn't need a 10-step framework - ❌ **No output format** — Always specify structure unless truly open-ended - ❌ **Assuming user knowledge** — Explain why you're using specific frameworks or techniques - ❌ **One-size-fits-all** — Tailor the prompt to the specific task and context - ❌ **Forgetting the copy-paste test** — User should be able to copy and use immediately - ❌ **No refinement cycle** — Always offer to iterate and improve --- ## Special Cases ### "Improve this prompt" When user provides an existing prompt to improve: 1. Read the original prompt carefully 2. Identify weaknesses: vague task, missing context, no output format, etc. 3. Ask 1-2 clarifying questions if needed 4. Present improved version with **before/after** comparison 5. Explain what was changed and why ### "What's the best prompt for X?" When the user asks for best practices without a specific task: 1. Clarify the specific use case first 2. Then generate a tailored prompt 3. Explain the prompt engineering principles used ### "Generate 10 variations" When user wants multiple options to choose from: 1. Create 3-5 variations (10 is too many) 2. Each variation should have a **distinct approach** or framework 3. Label each with its strength: "Variation A (Most Detailed)", "Variation B (Fastest)", etc.