Claude Platform Development Skill

A comprehensive Claude Code skill for building applications on the Anthropic Claude platform, covering API integration, prompt engineering, tool use, streaming, and production deployment patterns.

When to Use This Skill

Choose this skill when:

• Building applications that integrate with the Claude API
• Designing multi-turn conversational systems with Claude
• Implementing tool use (function calling) with Claude models
• Setting up streaming responses for real-time user experiences
• Optimizing prompts and managing context windows effectively
• Deploying Claude-powered applications to production

Consider alternatives when:

• You need to use a different LLM provider (use a provider-specific skill)
• You need fine-tuning capabilities (Claude uses prompt engineering, not fine-tuning)
• You need local model inference (Claude is a cloud API service)

Quick Start

Copy
# Install the Anthropic SDK
npm install @anthropic-ai/sdk

# Set your API key
export ANTHROPIC_API_KEY="your-key-here"

# Add the skill to your project
claude mcp add claude-platform

Copy
import Anthropic from '@anthropic-ai/sdk';

const anthropic = new Anthropic();

// Basic message
const response = await anthropic.messages.create({
  model: 'claude-sonnet-4-20250514',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello, Claude!' }]
});

console.log(response.content[0].text);

Core Concepts

Model Selection Guide

Model	Best For	Max Tokens	Relative Speed
claude-opus-4-20250514	Complex reasoning, code generation, analysis	32K output	Slower
claude-sonnet-4-20250514	Balanced performance, most use cases	16K output	Medium
claude-haiku-3-5-20241022	Fast responses, simple tasks, high volume	8K output	Fastest

Tool Use Pattern

Copy
const response = await anthropic.messages.create({
  model: 'claude-sonnet-4-20250514',
  max_tokens: 1024,
  tools: [{
    name: 'get_weather',
    description: 'Get current weather for a location',
    input_schema: {
      type: 'object',
      properties: {
        location: { type: 'string', description: 'City name' },
        units: { type: 'string', enum: ['celsius', 'fahrenheit'] }
      },
      required: ['location']
    }
  }],
  messages: [{ role: 'user', content: 'What is the weather in London?' }]
});

// Handle tool use response
for (const block of response.content) {
  if (block.type === 'tool_use') {
    const result = await callWeatherAPI(block.input);
    // Send tool result back to Claude
  }
}

Streaming Responses

Copy
const stream = await anthropic.messages.stream({
  model: 'claude-sonnet-4-20250514',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Write a story' }]
});

for await (const event of stream) {
  if (event.type === 'content_block_delta' &&
      event.delta.type === 'text_delta') {
    process.stdout.write(event.delta.text);
  }
}

Configuration

Parameter	Type	Default	Description
model	string	"claude-sonnet-4-20250514"	Claude model to use for generation
max_tokens	number	1024	Maximum tokens in the response
temperature	number	1.0	Randomness of output (0.0-1.0)
system	string	""	System prompt for behavior configuration
stop_sequences	array	[]	Sequences that stop generation
top_p	number	1.0	Nucleus sampling threshold
top_k	number	—	Top-k sampling limit
stream	boolean	false	Enable streaming responses

Best Practices

1. Use system prompts for consistent behavior — define the assistant's role, constraints, and output format in the system message rather than repeating instructions in every user message.

2. Implement exponential backoff for rate limits — the Claude API returns 429 status codes when rate limited; implement retry logic with exponential backoff and jitter to handle traffic spikes gracefully.

3. Structure tool definitions with detailed descriptions — Claude performs better with tool use when each tool and parameter has a clear, specific description explaining what it does and when to use it.

4. Manage context windows proactively — track token usage in responses and implement conversation summarization or sliding window strategies before hitting context limits.

5. Cache prompt prefixes for cost savings — when sending the same system prompt or document prefix repeatedly, use prompt caching to reduce costs and latency on subsequent requests.

Common Issues

Responses cut off mid-sentence — This happens when max_tokens is too low for the expected response length. Increase max_tokens or check the stop_reason field — if it says max_tokens, the response was truncated and you need a higher limit.

Tool use loop without final answer — Claude may repeatedly call tools without producing a text response. Set a maximum tool-use iteration count (typically 5-10) and include instructions in the system prompt telling Claude to synthesize a final answer after gathering enough information.

High latency on complex requests — For long prompts or complex reasoning, response times can increase significantly. Use streaming to display partial results immediately, and consider breaking complex tasks into smaller sequential requests.