claude-api

claude-api is a Skill that Anthropic publishes in its official anthropics/skills repository. SKILL.md describes it as a "Reference for the Claude API / Anthropic SDK", covering model IDs, pricing, parameters, streaming, tool use, MCP, agents, prompt caching, token counting and model migration. When you build a Claude application, Claude first works out which language the project uses, then reads the documentation for that language; Python, TypeScript, Java, Go, Ruby, C#, PHP and cURL are covered. SKILL.md requires code to call Claude through the official SDK, and to write raw HTTP requests only when you explicitly ask, when the project is a shell or cURL project, or when the language has no official SDK; function names, class names and import paths must come from the Skill's documentation or the official SDK repositories, with an explicit "Never guess SDK usage." Before editing, it checks the target file for code from other providers such as OpenAI, and if it finds any, it stops and asks you instead of rewriting that code to call Claude. The Skill includes a list of official documentation URLs (live-sources.md), its built-in model and pricing table is marked as cached on 2026-10-06, and SKILL.md says to fetch the official pages again when you ask for "latest" or "current" information. What sets it apart is an API drift table near the top that sets common older patterns beside the current ones, warning the AI not to write code from outdated interfaces it remembers from training.

Who it's for:developers building applications or writing integration code with the Claude API or Anthropic SDKs. Readers who don't write code won't get much from it.

Before you start: the repository's plugin manifest lists it as its own claude-api plugin, which you can install in Claude Code after adding the anthropics/skills plugin marketplace. The Skill is reference material; actually calling the API needs your own Anthropic account and API key, and requests are billed at official rates, so check the official site for pricing. It triggers very broadly. Going by SKILL.md, it activates whenever a prompt mentions Claude, Anthropic or any Claude model name, whenever you ask about any LLM's pricing, model choice or limits, and for LLM-type tasks that don't name a provider (agents, MCP, RAG, generation, summarizing, classification and so on); it skips only when another provider such as OpenAI or Gemini is named in the prompt or already used in the project. SKILL.md also sets the current Opus model as the default unless you explicitly name another one, and says "Never downgrade for cost". That is the Skill's own default; choose a model based on your task and budget, and check current models and prices on the official pages before you start. The documentation also covers Managed Agents, computer use, the Admin API and credential vaults, which involve cloud environments, account administration and control of a computer; AI Inspiration World has not reviewed or recommended those parts. The README says the Skills in this repository are for demonstration and educational purposes and that what you get in Claude may differ from what the repository shows. The license is Apache-2.0.

Based on the project's public documentation. Not tested by AI Inspiration World.