Claude API skill (claude-api)

A model writing code against its own API can still reach for old model names and parameters. The claude-api skill fixes that by loading current documentation when you build with Anthropic's SDKs. Anthropic's docs describe it as open source; it ships with Claude Code and is also published in the anthropics/skills repository under Apache 2.0. For how skills load, see how Claude Skills work.

What the skill does

It covers two surfaces: the Messages API and Claude Managed Agents, which is in beta. For each, it holds language-specific docs for Python, TypeScript, C#, Go, Java, PHP, Ruby and cURL, and loads only the parts your project needs. Anthropic's page lists what it brings:

• SDK setup, error handling, streaming, tool use, batches and prompt caching.
• Current model IDs, context window sizes and pricing.
• Model migration with /claude-api migrate: it swaps model IDs, removes parameters the new model no longer accepts, cleans up beta headers, explains each change inline and ends with a checklist of what to verify by hand.
• A guided setup for a new Managed Agent with /claude-api managed-agents-onboard.

It activates when your code imports an Anthropic SDK or you ask to build with the Claude API. It stays out of general programming work and of code that uses other AI SDKs, such as OpenAI's.

When to use it

• Starting an app or script that calls Claude.
• Moving a project to a newer Claude model without hunting for breaking changes.
• Checking prompt caching or batch settings in existing code.

Choosing a model first? Read our Claude Opus vs Sonnet comparison. Our Claude Code cheat sheet and Claude Code prompts cover the rest of the workflow.

How to install it

Claude Code: nothing to install. Type /claude-api or just ask for API help.

Other agents that read skills:

npx skills add https://github.com/anthropics/skills --skill claude-api

Or as a Claude Code plugin from the repository:

/plugin marketplace add anthropics/skills
/plugin install claude-api@anthropic-agent-skills

Example request

/claude-api migrate [folder or files] to [target model ID]. Explain each change, keep our existing prompts, and give me the list of things I need to test afterward.

Limits

• It is for code that calls Anthropic's API. It does not help with other providers' SDKs.
• Managed Agents runs only on the Claude API and Claude Platform on AWS, so the skill routes other cloud setups to the Messages API.
• Some checks send real API requests, billed at normal rates. The docs say the skill states the cost and asks first.
• A migration still needs your tests. The skill's own checklist names integration tests and cost re-baselining.