claude-api · Claude API Reference

The authoritative reference skill for building Claude-powered applications: model IDs and pricing, request parameters, streaming, tool use, MCP, agents, prompt caching, token counting, and model migration — all verified against the skill's bundled docs, never written from memory.

Covers Python, TypeScript, Java, Go, Ruby, C#, PHP SDKs plus raw cURL, with built-in subcommand workflows: migrate, prompt-audit, upgrade, cost-optimize, and build-eval.

Example invocation: "Write a Python streaming chat example using the Claude API."

Full brief

Positioning

claude-api is a reference-manual skill for building LLM applications with the Claude API / Anthropic SDKs: it answers "how do I write this call" and "how do I migrate, audit, or optimize it." Its core discipline is never writing SDK usage from memory — function names, parameters, and model IDs always come from the skill's per-language docs; stale recall gets verified first.

Core capabilities

Models & pricing: built-in model table (default claude-opus-5-5; per-model context windows and input/output prices, e.g. Opus 5.5 at $4/$20 per million tokens); live capability lookup via the Models API.
API drift tables: quick-reference for frequently changed surfaces (thinking config, tool type versions, parameter naming) — check before writing code.
Surface selection: four tiers — single call / workflow / manual loop / Tool Runner / Managed Agents — chosen by who supplies the harness and who hosts the deployment.
Per-language docs: Python, TypeScript, Java, Go, Ruby, C#, PHP, and raw cURL/HTTP, each with its own reference directory.
Subcommand workflows: migrate (model migration), prompt-audit (prompt auditing), upgrade (major SDK upgrades), cost-optimize (cost reduction), build-eval (eval set construction), hillclimb (iterative improvement).
Auth resolution: multi-source credential resolution from ANTHROPIC_API_KEY to ant auth login profiles.

Workflow

Scan the target file or request; confirm the language and provider (stops to confirm on non-Anthropic projects)
Read the matching language reference
Write the call code per the docs; compile/run to verify
Subcommand mode: recognize the subcommand, then follow its dedicated guide step by step

Inputs & outputs

Input	Notes
Requirement or target file	Required
Language	Inferred; asked when unclear
Subcommand	Optional — e.g. migrate, cost-optimize

Output: SDK call code, migration/audit/cost-optimization reports.

Boundaries with adjacent skills

mcp-builder: a guide to building MCP servers (producing external tools for LLMs).
claude-api (this): the reference manual for building applications with the Claude API (consuming LLM capabilities). Complementary: an MCP server is one tool source a claude-api app can connect to.

Fit

Wiring the Claude API into a new project with verified multi-language SDK usage
Migrating existing code to a newer model (breaking-change checklists)
Prompt audits: cleaning up patterns written for older models
Cost optimization: ranked savings levers across caching, effort levels, and model choice

Before you start

Bring your own Anthropic API key (ANTHROPIC_API_KEY env var) or sign in with ant auth login; if no key is set, run ant auth status first to confirm the credential source.
API calls are metered; model prices are in the skill's pricing table. build-eval and hillclimb spend real tokens — confirm the budget before running.

claude-api is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.

← All 155 skills