Humanizer
Multi-pass editing to remove AI-writing tells: em-dash overuse, throat-clearing openers, rule of three, synonym cycling, promotional inflation, and 24+ documented patterns (Wikipedia AI-writing guide).
No curl, terminal, Node.js, or Git required! Download the official SKILL.md directly into your Downloads folder.
Your browser immediately downloads the authentic file (9.4 KB) directly into your local Downloads folder.
Place the downloaded SKILL.md into the skills path for your chosen AI assistant:
<project>/.agents/skills/ultimate-prompt-builder/SKILL.md (or global in ~/.gemini/antigravity/skills/ultimate-prompt-builder/SKILL.md)~/.claude/skills/ultimate-prompt-builder/SKILL.md<project>/.cursor/skills/ultimate-prompt-builder/SKILL.md or ~/.cursor/skills/ultimate-prompt-builder/SKILL.md<project>/.agents/skills/ultimate-prompt-builder/SKILL.mdYour agent automatically discovers the skill. You don't need any restart—it's immediately active!
---
name: prompt-builder
description: Build high-quality, production-ready prompts for any use case using structured prompt engineering frameworks. Trigger this skill whenever the user says "build a prompt for", "create a prompt that", "write a prompt to", "make a prompt for", "generate a prompt", "help me prompt", "I need a prompt for", or drops any task and wants a ready-to-use AI instruction. Covers every domain — writing, coding, marketing, design, image generation (Nano Banana, GPT-4o, Seedream, Midjourney, Flux), vibe coding tools (Cursor, Lovable, Bolt, v0), business, finance, education, healthcare, legal, HR, e-commerce, and more. If someone needs a prompt built, use this skill — every single time, automatically.
---
# Prompt Builder Skill
## Purpose
This skill builds production-ready prompts for any domain and any AI tool. It uses structured framework selection, smart clarifying questions, and a mandatory humanization layer so every output prompt feels written by a thoughtful human — not generated by a machine.
---
## Step 1 — Show the Popup First
The moment a user asks to build a prompt, before doing or writing anything else, use `ask_user_input_v0` to show two questions:
**Q1:** "How do you want to build this?"
- Auto mode — you pick the best frameworks and ask smart questions
- Custom flow — walk me through the options step by step
**Q2:** "Want me to ask clarifying questions before building?"
- Yes, ask me questions first
- No, build it with smart assumptions
Then proceed based on their answers.
---
## Step 2 — Clarifying Questions (if selected or if task is vague)
Ask only what matters for this specific request. Use `ask_user_input_v0`. Pick 2–3 max — do not ask everything.
**Always relevant:**
- What is the target audience or end-reader of this output?
- What tone is needed — formal, conversational, aggressive, friendly, technical?
- Which AI tool or platform will this prompt be used in?
**For writing tasks:**
- What is the primary goal — clicks, replies, signups, trust-building, conversion?
- Any brand voice rules or phrases to avoid?
**For code/build tasks:**
- Which platform — Cursor, Lovable, Bolt, v0, standard editor?
- Tech stack or constraints?
**For image generation tasks:**
- Which model — Nano Banana, GPT-4o, Seedream, Midjourney, Flux, Stable Diffusion, other?
- Output format — portrait, landscape, square, specific aspect ratio?
- Style direction — photorealistic, illustration, cinematic, 3D, anime, product photography?
**For unknown or niche use cases:**
Run a web search before asking the user anything. Come back with what you learned, then ask targeted questions if still needed.
---
## Step 3 — Use Case Detection
Check if the request matches a specialized file in `/use-cases/`. If it does, load that file and follow its templates and instructions.
### Specialized Use Cases Available
| Use Case | File | Triggers |
|---|---|---|
| Ad Copy | `ad-copy.md` | Meta ad, Facebook ad, Google ad, Instagram ad, YouTube pre-roll, PPC, paid ad |
| Email Writing | `email.md` | Cold email, marketing email, drip, nurture, follow-up, outreach, newsletter |
| Build Website | `build-website.md` | Build a site, landing page, web app, Lovable, Cursor, Bolt, v0 |
| Refactor Website | `refactor-website.md` | Refactor, redesign, improve code, clean up, fix UI, code quality |
| Image Generation | `image-generation.md` | AI image, Nano Banana, GPT-4o, Seedream, Midjourney, Flux, DALL-E, generate image |
| YouTube Script | `youtube-script.md` | YouTube script, video script, shorts script, hook, video outline, talking head |
| Social Media | `social-media.md` | Instagram caption, LinkedIn post, Twitter thread, X thread, community post, social post |
| Sales Page Copy | `sales-page.md` | Sales page, VSL script, sales letter, offer page, landing page copy, webinar page |
| SEO Blog Post | `seo-blog.md` | Blog post, SEO article, how-to guide, listicle, comparison article, long-form content |
| Product Description | `product-description.md` | Product listing, Amazon listing, Shopify description, e-commerce copy, product page |
| n8n / Automation | `n8n-automation.md` | n8n prompt, system prompt, AI agent, chatbot instructions, LLM node, automation prompt |
| Industry-Specific | `industry-specific.md` | Finance, legal, healthcare, HR, recruitment, real estate, education, course content |
If the request does NOT match any file — proceed with general framework selection from `references/frameworks.md`.
---
## Step 4 — Web Search Fallback (ALWAYS trigger for unknown platforms or industries)
If the user mentions any platform, tool, model, or domain that is not in the use case files above, or that you have limited specific knowledge about:
1. Do NOT guess or invent platform-specific syntax or norms
2. Run a web search: `[platform or topic] prompt guide best practices [current year]`
3. Read results, extract actual documented requirements
4. Build the prompt using those real requirements
5. Add to the framework breakdown: "Web search used to retrieve [platform/domain]'s current documented requirements"
**Always web search for these when no file exists:**
- Any new or updated image generation model
- Any vibe coding tool or IDE not listed
- Finance, legal, healthcare, HR content — industry-specific norms matter
- Any platform whose character limits or prompt syntax changes frequently (Meta, Google Ads, LinkedIn)
---
## Step 5 — Framework Selection
Read `references/frameworks.md` for full framework definitions.
**Always select exactly 3 frameworks** — the most relevant for this specific request. Never fewer, never more.
### Quick Selection Table
| Task Type | Recommended 3 Frameworks |
|---|---|
| Ad Copy | Role Assignment + Constraints + Context & Background |
| Email (cold outreach) | Role Assignment + Context & Background + Constraints |
| Email (marketing) | Role Assignment + Few-Shot + Constraints |
| Website Build | Decomposition + Clear Instruction + Chain-of-Thought |
| Website Refactor | Context & Background + Constraints + Decomposition |
| Image Generation | Clear Instruction + Constraints + Context & Background |
| Blog / Long-form | Role Assignment + Context & Background + Clear Instruction |
| YouTube Script | Role Assignment + Decomposition + Constraints |
| Social Media Post | Role Assignment + Few-Shot + Constraints |
| Product Description | Context & Background + Constraints + Clear Instruction |
| Sales / Landing Page Copy | Role Assignment + Chain-of-Thought + Constraints |
| Finance / Business Analysis | Chain-of-Thought + Constraints + Structured Output |
| Research / Fact-heavy | Chain-of-Thought + Safety/Hallucination + Structured Output |
| HR / Legal / Healthcare | Role Assignment + Safety/Hallucination + Constraints |
| Automation / Agent | Clear Instruction + Decomposition + Constraints |
| Education / Course | Role Assignment + Decomposition + Context & Background |
In **Auto mode**: select silently, build immediately.
In **Custom mode**: show the 3 selected frameworks to the user and let them swap before building.
---
## Step 6 — Humanization Layer (MANDATORY IN EVERY PROMPT EXCEPT IMAGE GEN)
Every prompt this skill produces must include a humanization instruction baked directly into the prompt text itself.
This tells the AI receiving the prompt to:
- Write like a real human, not an AI assistant trying to sound professional
- Avoid filler phrases, robotic transitions, and corporate-speak
- Use natural sentence rhythm — short punchy lines mixed with longer ones
- Sound like someone who genuinely understands this topic
- Avoid these words: "leverage", "utilize", "in today's world", "it's important to note", "as an AI", "I hope this finds you well", "touch base", "circle back", "synergize", "paradigm shift", "cutting-edge", "seamless", "empower"
**Adapt phrasing to task tone:**
- Formal/business → formal humanization: "Write with the precision of a senior professional — no filler, no buzzwords, no AI-sounding transitions."
- Casual/creative → conversational: "Sound like a real person wrote this — not an AI trying to seem helpful."
- Technical/code → clarity-focused: "Keep explanations clear and direct. No robotic phrasing. Write comments like a thoughtful developer, not documentation software."
- Image generation → skip humanization entirely (not applicable to visual models)
---
## Step 7 — Output Format
ALWAYS output in this exact structure:
````
### Built Prompt
```
[THE COMPLETE PROMPT — copy-paste ready]
```
### Framework Breakdown
**[Framework 1 Name]:** One sentence — why this was chosen for this specific task.
**[Framework 2 Name]:** One sentence — why this was chosen for this specific task.
**[Framework 3 Name]:** One sentence — why this was chosen for this specific task.
````
No weak-to-strong comparison. No long preamble. No excessive explanation. The code block is the deliverable.
---
## Edge Cases
**"Make it more human" after seeing output:** Humanization instruction wasn't specific enough. Strengthen it and rebuild.
**Very little info given in Auto mode:** Make smart assumptions, state them in one line before the code block ("Assuming B2B SaaS audience, formal tone, lead generation goal"), then build.
**Very specific industry (finance, legal, medical):** Web search for that industry's content conventions, load the right framework bias, then build.
**Platform specs may have changed:** Always web search current character limits and syntax before building. Platforms update frequently.
In Google Antigravity, skills reside in your project workspace under .agents/skills/ultimate-prompt-builder/SKILL.md or globally in ~/.gemini/antigravity/skills/.
npx skills add https://github.com/vivekmishraishere/ultimate-prompt-builder-skill --skill ultimate-prompt-builder -y
# Workspace path: .agents/skills/ultimate-prompt-builder/SKILL.md
# Global path: ~/.gemini/antigravity/skills/ultimate-prompt-builder/SKILL.mdIn Claude Code, install globally using the official skills CLI or add via the Claude Code plugin marketplace.
npx skills add https://github.com/vivekmishraishere/ultimate-prompt-builder-skill --skill ultimate-prompt-builder -g -y
# Or inside Claude Code session:
/plugin install vivekmishraishere/ultimate-prompt-builder-skillIn Cursor, place the skill folder in .cursor/skills/ultimate-prompt-builder/SKILL.md or at the workspace level in .agents/skills/.
npx skills add https://github.com/vivekmishraishere/ultimate-prompt-builder-skill --skill ultimate-prompt-builder -y
# Or place directly into:
# ~/.cursor/skills/ultimate-prompt-builder/SKILL.mdIn Windsurf / VS Code, agent skills can be installed targeting the windsurf profile or loaded as project rules.
npx skills add https://github.com/vivekmishraishere/ultimate-prompt-builder-skill --skill ultimate-prompt-builder -a windsurf -yBuild production-ready prompts for any use case using structured prompt engineering frameworks. Covers writing, coding, marketing, image generation (Flux, Midjourney, GPT-4o), and vibe coding tools (Cursor, Lovable, Bolt, v0).
This skill coordinates companion sub-skills, domain templates, or interactive sub-modules from its repository. The downloaded SKILL.md contains the master coordination instructions. To ensure your AI agent has access to all companion templates and sub-skills, we recommend installing via CLI or using the Ask-AI prompt above with the GitHub repo link.
Whenever you need to create, optimize, or debug AI prompts, system instructions, or agent personas with structured frameworks and humanization.
Uses framework selection, clarifying questions, domain templates (writing, coding, image gen, vibe coding), and a humanization layer to produce prompts that feel thoughtful and human-crafted.
Copy and paste this starter prompt into your AI assistant (Antigravity, Claude, Cursor, ChatGPT) to activate this skill:
Act as an expert using the "Ultimate Prompt Builder" agent skill. Guide me through constructing a high-impact, production-ready prompt using the best prompt engineering frameworks.Multi-pass editing to remove AI-writing tells: em-dash overuse, throat-clearing openers, rule of three, synonym cycling, promotional inflation, and 24+ documented patterns (Wikipedia AI-writing guide).
Hard-edged writing style contract for timeless, forceful English prose—concrete, verb-forward, no therapeutic tone or marketing cadence.