# AI System Prompts

A collection of production-ready system prompts and instructions for the AI tools developers actually use. Every prompt is copy-paste ready and includes explanation, usage tips, and customization guidance.

---

## What's in here

### `chatgpt/` — OpenAI ChatGPT

Prompts for the ChatGPT API, Custom GPTs, and the ChatGPT settings panel.

| File | What it does |
|------|-------------|
| `README.md` | How system prompts work across the API, Custom GPTs, and ChatGPT settings |
| `software-architect.md` | Thinks in trade-offs, asks about scale first, draws ASCII diagrams |
| `pair-programmer.md` | Thinks out loud, writes incrementally, flags wrong approaches |
| `data-analyst.md` | Checks data quality before analyzing, pandas-first, honest about uncertainty |

### `gemini/` — Google Gemini

Prompts for Google AI Studio, the Gemini API, and Vertex AI.

| File | What it does |
|------|-------------|
| `README.md` | How to use system prompts across AI Studio, REST API, Python SDK, and Vertex AI |
| `research-assistant.md` | Labels every claim with a confidence level, structures findings clearly |
| `multimodal-analyst.md` | Systematic analysis for charts, code screenshots, UI mockups, and diagrams |

### `github-copilot/` — GitHub Copilot

Instructions files for `.github/copilot-instructions.md` and VS Code settings.

| File | What it does |
|------|-------------|
| `README.md` | How Copilot instructions work (repo file, VS Code settings, prompt files) |
| `project-conventions.md` | Naming, TypeScript rules, React patterns, testing, error handling, imports |
| `security-first.md` | Input validation, SQL safety, secrets handling, XSS prevention, least privilege |

### `v0-bolt/` — v0 (Vercel) and Bolt (StackBlitz)

Prompts for AI UI generators — structured differently from chat prompts.

| File | What it does |
|------|-------------|
| `README.md` | How v0 and Bolt work, how prompting differs from chat AI, tips for iteration |
| `dashboard-ui.md` | SaaS analytics dashboard — dark mode, shadcn/ui, Recharts, realistic data |
| `landing-page.md` | High-converting landing page — hero, social proof, features, pricing, CTA |

---

## How to use this repo

**For ChatGPT and Gemini:** Open the relevant `.md` file, copy the text inside the code block, and paste it into the system prompt field of your tool.

**For GitHub Copilot:** Copy the contents of the instructions file into `.github/copilot-instructions.md` in your project. Customize the stack and conventions for your specific setup.

**For v0 and Bolt:** Copy the prompt from the code block and paste it directly into v0.dev or bolt.new. Iterate with follow-up messages.

---

## Principles behind these prompts

**Specificity beats generality.** "Write clean code" does nothing. "Use `const` by default, never `var`, and explicit return types on all exported functions" does something.

**Structure the output.** Prompts that specify format (headers, confidence levels, step-by-step) produce more consistent results than prompts that just describe a persona.

**Define failure behavior.** Every prompt here tells the model what to do when it doesn't know: ask for context, say "I'm uncertain," flag the limitation. Without this, models fake confidence.

**Prompts drift in long conversations.** For chat-based tools, expect behavior to drift after many exchanges. Starting a fresh conversation with the same system prompt resets it.

---

## Contributing

If you improve a prompt or write a new one that fits the pattern, follow the file structure: prompt in a code block, explanation of what it does, how to use it, and customization tips.