How to Add AI Skills & Automations

Go to chat.openai.com and click your profile → My GPTs → Create a GPT

In the 'Create' tab, describe what you want your GPT to do in natural language

Configure name, description, instructions, and conversation starters

Upload knowledge files (PDFs, docs) for your GPT to reference

Add Actions to connect external APIs (optional)

Set visibility: Only me, Anyone with a link, or Public

Click 'Create' and share the link

Click your profile icon → Settings → Personalization → Custom Instructions

Fill in 'What would you like ChatGPT to know about you?' with context about your role and needs

Fill in 'How would you like ChatGPT to respond?' with output format preferences, tone, and constraints

These instructions apply to all new conversations automatically

In GPT Builder, go to the 'Configure' tab → Actions

Click 'Create new action' and paste your OpenAPI schema

Set the authentication method (None, API Key, or OAuth)

Test the action with sample requests

The GPT can now call your API during conversations

Go to claude.ai and click 'Projects' in the sidebar

Click 'Create Project' and give it a name and description

Add custom instructions that define how Claude should behave in this project

Upload files (PDFs, code, docs) to the project knowledge base — Claude can reference up to 200K tokens

Start conversations within the project — Claude will use your instructions and files automatically

When using the Claude API, include a 'system' message at the beginning of your conversation

Define Claude's role, behavior, output format, and constraints in the system prompt

Include few-shot examples of desired inputs and outputs

Use XML tags like <instructions>, <context>, <output_format> for clear structure

Install Claude Code: npm install -g @anthropic-ai/claude-code

Create a CLAUDE.md file in your project root with instructions and context

Add custom slash commands in .claude/commands/ directory as markdown files

Each .md file becomes a /command — the file content is the prompt template

Use $ARGUMENTS placeholder for dynamic input

Claude Desktop and Claude Code support Model Context Protocol (MCP) tools

Install MCP servers: npx @anthropic-ai/create-mcp-server or use existing ones from npm

Configure in claude_desktop_config.json or .claude/settings.json

Claude can now use the MCP tools during conversations to access databases, APIs, files, etc.

Go to aistudio.google.com and click 'Create new prompt'

Choose 'Structured Prompt' for consistent, template-based outputs

Define input columns (variables) and output columns (expected results)

Add example rows showing input-output pairs for few-shot learning

Test with new inputs and iterate on your examples

Save and share via the 'Get Code' button (Python, JavaScript, REST)

Click 'Create new prompt' → 'Chat Prompt'

Write system instructions defining the AI's role and behavior

Add tools (function declarations) if your prompt needs to call external APIs

Configure model settings: temperature, top-p, max output tokens

Test in the playground and export code when ready

Get your API key from aistudio.google.com → 'Get API Key'

Install the SDK: pip install google-genai (Python) or npm install @google/genai

Initialize the client with your API key

Use generateContent() for text, image, audio, and video inputs

Add tool declarations for function calling capabilities

For production, use Vertex AI in Google Cloud Console

Create an endpoint and deploy your tuned model or use Gemini directly

Set up authentication with service accounts

Configure autoscaling, monitoring, and logging

Use Vertex AI Pipelines for automated ML workflows

Open the Antigravity dashboard and click 'New Workflow'

Choose a trigger: Schedule, Webhook, Form Submission, or Event

Drag and drop action blocks from the sidebar: Send Email, API Call, Database Query, etc.

Connect blocks with arrows to define the flow

Add conditional branches (If/Else) for decision logic

Configure each block's settings (API endpoints, email templates, etc.)

Test the workflow with sample data before activating

Toggle the workflow to 'Active' to start running

Browse the Template Gallery for pre-built workflows

Click 'Use Template' to create a copy in your workspace

Customize the template: change trigger settings, update API keys, modify conditions

Connect your accounts (Slack, Gmail, CRM, etc.) when prompted

Test and activate

Go to Settings → Integrations → Add New

Choose 'Custom API' for services not in the built-in list

Enter the API base URL and authentication details

Define the available actions (endpoints) with request/response schemas

Your custom integration now appears in the workflow builder sidebar

Find MCP servers on npm, GitHub, or this directory

Install: npm install -g @modelcontextprotocol/server-name

Configure in your AI client's settings file (e.g., claude_desktop_config.json)

Add the server command and any required environment variables

Restart your AI client — the tools become available automatically

The AI can now use the MCP tools during conversations

Initialize: npx @anthropic-ai/create-mcp-server my-server

Define tools in src/index.ts using the MCP SDK

Each tool needs: name, description, input schema (JSON Schema), and a handler function

Add resources for data the AI can read (files, database records, etc.)

Add prompts for reusable prompt templates

Test locally: npx tsx src/index.ts

Publish to npm for others to use: npm publish

For Claude Desktop: Edit ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%/Claude/claude_desktop_config.json (Windows)

For Claude Code: Edit .claude/settings.json in your project or ~/.claude/settings.json globally

Add your server to the 'mcpServers' object with command and args

Set environment variables with the 'env' field for API keys

Use 'stdio' transport for local servers, 'sse' for remote ones

MCP is an open standard — it works with any compatible AI client

For custom integrations, use the MCP client SDK (@modelcontextprotocol/sdk)

Connect to MCP servers programmatically in your own applications

Bridge MCP to OpenAI/Gemini by translating MCP tool schemas to their function calling format

Use MCP proxy servers to expose MCP tools via REST APIs