claude-agent-sdk
Claude Agent SDK — Skill Router
Build and deploy AI agents with the same tools, agent loop, and context management that power Claude Code.
Source: platform.claude.com/docs | Python: claude-agent-sdk v0.1.51 | TypeScript: @anthropic-ai/claude-agent-sdk v0.2.86
Reference Files
| Reference | File | Read When |
|---|---|---|
| Overview & Quickstart | references/00-overview.md |
Getting started, installation, architecture, authentication, first agent |
| Configuration | references/01-configuration.md |
ClaudeAgentOptions, model selection, effort, thinking, environment variables |
| Built-in Tools | references/02-built-in-tools.md |
Read, Write, Edit, Bash, Glob, Grep, WebSearch, tool availability |
| Custom Tools | references/03-custom-tools.md |
@tool decorator, tool() function, input schemas, return types, annotations |
| MCP Integration | references/04-mcp-integration.md |
MCP servers, stdio/HTTP/SSE/SDK transports, .mcp.json, tool naming |
| Hooks | references/05-hooks.md |
PreToolUse, PostToolUse, hook matchers, callbacks, async hooks |
| Permissions | references/06-permissions.md |
Permission modes, canUseTool, evaluation order, security |
| Sessions | references/07-sessions.md |
Multi-turn, ClaudeSDKClient, resume, fork, session management |
| Subagents | references/08-subagents.md |
AgentDefinition, spawning, model selection, tool scoping, limitations |
| Structured Outputs | references/09-structured-outputs.md |
output_format, JSON Schema, Pydantic, Zod, validation |
| Deployment | references/10-deployment.md |
System requirements, deployment patterns, sandboxes, cost, security |
| Streaming | references/11-streaming.md |
StreamEvent, partial messages, streaming vs single mode, building UIs |
| User Input | references/12-user-input.md |
AskUserQuestion, approval flows, canUseTool, option previews |
| Agent Loop & Errors | references/13-agent-loop.md |
Turns, compaction, stop reasons, error types, rate limits, cost tracking |
| System Prompts & Features | references/14-system-prompts-features.md |
CLAUDE.md, skills, plugins, slash commands, output styles |
| Secure Deployment | references/15-secure-deployment.md |
Docker hardening, credential management, proxies, isolation |
Installation
# Python
pip install claude-agent-sdk
# TypeScript
npm install @anthropic-ai/claude-agent-sdk
Quick Reference
- Docs: https://platform.claude.com/docs/en/agent-sdk/overview
- GitHub (Python): https://github.com/anthropics/claude-agent-sdk-python
- GitHub (TypeScript): https://github.com/anthropics/claude-agent-sdk-typescript
- PyPI: https://pypi.org/project/claude-agent-sdk/
- npm: https://www.npmjs.com/package/@anthropic-ai/claude-agent-sdk
More from abhisheksharma-17/skills-graph
agno
Build AI agents, multi-agent teams, and agentic workflows using the Agno framework. MANDATORY TRIGGERS: Agno, agno-agi, AgentOS, any mention of the Agno framework. Also trigger when the user wants to build AI agents with tools/memory/knowledge, create multi-agent systems, RAG pipelines, reasoning agents, agentic workflows, or deploy agents to production. Trigger even if the user just says 'build me an agent', 'create an AI assistant', or 'make a chatbot' — if Agno is anywhere in their stack or project dependencies. When in doubt about whether to use this skill for agent-building tasks, use it.
10remotion-prompt-generator
Generate detailed, production-ready prompts for the Remotion Dev skill to create programmatic React-based videos. MANDATORY TRIGGERS: remotion prompt, video prompt, generate video prompt, remotion video, create video prompt, video brief, video specification, make a video, create a video. Also trigger when user wants to create any kind of video using Remotion, needs help describing a video project, wants a prompt for a video generation tool, or asks for a video brief/spec. When in doubt, use it.
7design-engineering
Comprehensive frontend design engineering system covering the full design lifecycle — planning, building, styling, reviewing, refining, hardening, and upgrading interfaces. MANDATORY TRIGGERS: design, UI, UX, frontend, interface, component, layout, typography, color, animation, responsive, accessibility, critique, audit, polish, redesign, design system, onboarding, micro-interactions, visual design. Also trigger when user asks to build a page, create a component, review a design, improve typography, add animation, fix layout, audit accessibility, redesign an existing project, add delight, make it look professional, or any frontend visual/interaction task. When in doubt about whether to use this skill for any design or frontend task, use it.
2litellm
Unified LLM gateway and SDK for calling 100+ LLM providers (OpenAI, Anthropic, Bedrock, Vertex, Azure, Ollama, etc.) with a single OpenAI-compatible interface. MANDATORY TRIGGERS: litellm, litellm proxy, litellm router, litellm.completion, llm gateway, multi-provider llm, openai-compatible proxy, model fallback, llm cost tracking, llm load balancing. Also trigger when user wants to switch LLM providers without rewriting code, route requests across multiple models, add fallbacks/retries to LLM calls, track token costs, run a self-hosted OpenAI-compatible proxy, or unify observability across providers. When in doubt about whether to use this skill for any multi-provider LLM task, use it.
1livekit
Build realtime voice, video, and AI agent applications with LiveKit. MANDATORY TRIGGERS: livekit, live kit, voice agent, voice AI, realtime audio, realtime video, WebRTC agent, speech-to-speech, voice pipeline, voice assistant framework. Also trigger when user wants to build voice AI agents, realtime communication apps, telephony integrations, video conferencing, or multimodal AI agents with audio/video. When in doubt about whether to use this skill for realtime media or voice AI tasks, use it.
1