What programming languages does this skill support?

It supports Python, TypeScript/JavaScript, Java, Kotlin, Scala, Go, Ruby, C#, and PHP with language-specific documentation. For unsupported languages like Rust, Swift, and C++, it provides cURL/raw HTTP examples and notes that community SDKs may exist.

What are the default settings this skill applies when building with Claude?

It defaults to model `claude-opus-4-6` (Claude Opus 4.6), adaptive thinking (`thinking: {type: "adaptive"}`) for complex tasks, and streaming for requests with long input/output or high `max_tokens` to avoid timeouts.

When should I use the Agent SDK versus Claude API with tool use?

Use the Agent SDK when Claude itself needs to discover and access files, the web, or a shell as part of its work, since it provides built-in tools, permissions, and MCP support. Use Claude API with tool use for multi-step, code-orchestrated workflows where your app controls the tools.

How does this skill handle projects with multiple programming languages?

It checks which language the current file or question relates to, or asks you explicitly. If no language can be inferred, it asks via a user question or falls back to Python examples with a note.

Skill

Build Claude-Powered Apps with SDKs

Name: Building LLM-Powered Applications with Claude
Availability: OnlineOnly
Author: Antigravity

Routes Claude integration work to the right surface (API, tool use, or Agent SDK) and language-specific docs, with model defaults.

Get skill

Works with anthropic

Antigravity

Own this? Claim it

Spark score

out of 100

Updated last month

Version 13.1.1

Models

claudeclaude 3 opus

Add to Favorites

Why it matters

Develop sophisticated applications leveraging the Claude API and SDKs. Automate complex tasks, integrate AI capabilities, and build intelligent agents with robust language understanding and generation.

Outcomes

What it gets done

Integrate Claude API for single LLM calls or multi-step workflows.

Build custom agents with your own tools or utilize the Agent SDK for built-in functionalities.

Detect project language and provide relevant code examples for Python, TypeScript, Java, Go, Ruby, C#, and PHP.

Implement structured outputs and tool use for controlled and predictable AI responses.

Install

Add it to your toolbox

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/ag-claude-api | bash

Overview

Building LLM-Powered Applications with Claude

Guides building LLM-powered applications with Claude: detects the project's programming language to pull the right SDK docs, decides which surface fits the task (a single Claude API call, a code-orchestrated tool-use workflow, or the Agent SDK for built-in file/web/terminal tools and MCP), and applies current defaults for model version, thinking mode, and streaming. Use it when building with the Claude API, Anthropic SDKs, or the Agent SDK - i.e. code imports anthropic, @anthropic-ai/sdk, or related packages. Not for general coding work unrelated to Claude integrations.

What it does

This skill helps build LLM-powered applications with Claude by first detecting the project's language from its files (Python, TypeScript/JS, Java/Kotlin/Scala, Go, Ruby, C#, PHP, or falling back to cURL/raw HTTP), then reading the matching language-specific documentation rather than guessing at SDK syntax. It also applies current defaults unless the user specifies otherwise: model string claude-opus-4-6 (Claude Opus 4.6), adaptive thinking (thinking: {type: "adaptive"}) for anything remotely complicated, and streaming for any request with long input/output or high max_tokens to avoid timeouts, using .get_final_message()/.finalMessage() when individual stream events don't need to be handled.

When to use - and when NOT to

Use it when building with the Claude API, Anthropic SDKs, or the Agent SDK - specifically when code imports anthropic, @anthropic-ai/sdk, or a related Claude SDK package. Do not use it for general coding work unrelated to Claude integrations. Its core judgment call is choosing the right surface for the task, defaulting to the simplest tier: a single Claude API call for classification/summarization/extraction/Q&A or batch/embeddings work; Claude API with tool use for multi-step, code-orchestrated workflows or a custom agent with your own tools; and the Agent SDK specifically when Claude itself needs to discover and access files, the web, or a shell as part of its work (not when your app just hands Claude a file) - because the Agent SDK provides built-in tools, permissions, and MCP support rather than requiring you to reimplement them.

Inputs and outputs

A language-support matrix documents which languages have beta tool-runner support (Python via @beta_tool, TypeScript via betaZodTool+Zod, Java, Go via BetaToolRunner, Ruby via BaseTool+tool_runner) and which have Agent SDK support (Python and TypeScript only, per the cached model list). When multiple languages are detected in a project, the skill checks which one the current file/question relates to, or asks explicitly; when no language can be inferred, it asks via AskUserQuestion or falls back to Python examples with a note. For unsupported languages (Rust, Swift, C++, Elixir), it suggests cURL/raw HTTP examples and notes that community SDKs may exist. A decision tree walks through: single LLM call -> Claude API; Claude itself needs file/web/shell access -> Agent SDK; multi-step workflow with your own tools -> Claude API with tool use.

Integrations

The skill bundles per-language reference files (python/, typescript/, java/, go/, ruby/, csharp/, php/, curl/) plus quick-reference sections on current models, thinking/effort settings, context compaction, and when to use WebFetch versus the Agent SDK's own web access.

Who it's for

Developers integrating Claude into an application who need to pick the right architecture tier (API call, workflow, or agent) and get accurate, language-correct SDK usage and current model defaults instead of outdated or mismatched examples.

FAQ

Common questions

Discussion