Claude Agent SDK
The Claude Agent SDK is a software development kit provided by Anthropic to help developers build custom AI agents based on the Claude Code Agent Harness framework. The SDK supports both Python and TypeScript, providing standardized interfaces and toolchains that enable developers to extend and customize AI agent behaviors and capabilities. It is applicable to scenarios such as automated task execution and workflow orchestration.
Core Facts
Agent Skills是Anthropic发布的能力封装标准，定义为一个包含Skill.md文件的目录，可包含说明、脚本和其他资源
Claude Agent SDK supports multi-step reasoning, tool calls, and environment interaction, including capabilities such as code writing, file operations, and command execution.
The Claude Agent SDK has built-in tools such as Web Search and Web Fetch, giving the Agent the ability to search the internet and scrape web pages.
Timeline (last 90 days)
The Claude Agent SDK's permission defaults follow the principle of least privilege: newly created agents by default have only minimal tool access, and teams must explicitly declare the capabilities they need.
The Claude Agent SDK introduces structured output, enabling agents to return data with fixed fields according to a predefined schema (such as JSON Schema), making it easy for downstream programs to deserialize and use directly.
In the Claude Agent SDK, skills are positioned as capability units that encapsulate reuse of an agent's internal logic, and plugins are positioned as the channel for integrating external tools or services into the agent system.
The Claude Agent SDK lists SessionStore as a standalone capability, moving session state from memory to a manageable storage layer to support state persistence and replayable traces for long tasks.
The Claude Agent SDK's nested sub-agent mechanism allows an agent to spawn lower-level agents during execution, but the official documentation explicitly sets limits on the number of levels and quantity.
The updates to the Claude Agent SDK in the second half of 2026 focus on five areas: nested sub-agents (with quantity limits), SessionStore, skills, plugins, and structured output plus permission defaults.
Changes to permission defaults usually affect newly created configurations, and existing projects should verify that explicit permission declarations are complete when upgrading.
SessionStore is not mandatory in the Claude Agent SDK and can be omitted for short tasks; it is recommended to enable it for long tasks or scenarios requiring recovery and auditing.
A reasonable onboarding path for adopting the Claude Agent SDK is: first get a single agent plus structured output working, then introduce SessionStore for state persistence, and only then split out sub-agents; skills and plugins should be abstracted only when a second reuse need arises.
The official descriptions of the Claude Agent SDK's limits on nested sub-agents and the specific rules for permission defaults are relatively general; before adopting them in production, you should consult the detailed documentation for the corresponding version and rely on actual test results.
38 more timeline events
All Facts (20)
Agent Skills是Anthropic发布的能力封装标准，定义为一个包含Skill.md文件的目录，可包含说明、脚本和其他资源
80%VerifiedClaude Agent SDK supports multi-step reasoning, tool calls, and environment interaction, including capabilities such as code writing, file operations, and command execution.
65%VerifiedThe Claude Agent SDK has built-in tools such as Web Search and Web Fetch, giving the Agent the ability to search the internet and scrape web pages.
65%Unverifiedclaude-memory-compiler采用三层自动化流水线架构：会话捕获层（Hooks机制）、知识提取层（Claude Agent SDK）和知识编译层（LLM Compiler）
90%Unverified该方案的所有数据和推理都在本地完成，不会外传到第三方服务器，适合对数据隐私有严格要求的企业场景
90%Unverified通过设置环境变量ANTHROPIC_BASE_URL指向LiteLLM Proxy的本地端口，可以将Claude Agent SDK的请求重定向到本地模型，无需修改SDK源码
90%Unverified2024-2025年，Anthropic Claude Agent SDK、OpenAI Agents SDK、Google ADK等Agent框架集中涌现
85%UnverifiedClaude Agent SDK提供工具注册与调用、多轮对话上下文管理、智能体间消息传递与协调机制以及结构化输出控制等能力
85%Unverified框架能力与推理能力解耦是AI工程领域的重要趋势，Claude Agent SDK提供Agent编排框架，实际推理可由任何兼容模型完成
75%Unverified该方案的整体架构分为三层：Agent Framework层（任务管理和调度）、Claude Agent SDK层（核心Worker）、LiteLLM Proxy层（请求代理转发）
60%UnverifiedThe solution of connecting the Claude Agent SDK to a local large model via LiteLLM Proxy can reduce API call costs from nearly $1 per simple task to almost zero.
60%UnverifiedThe official descriptions of the Claude Agent SDK's limits on nested sub-agents and the specific rules for permission defaults are relatively general; before adopting them in production, you should consult the detailed documentation for the corresponding version and rely on actual test results.
50%UnverifiedA reasonable onboarding path for adopting the Claude Agent SDK is: first get a single agent plus structured output working, then introduce SessionStore for state persistence, and only then split out sub-agents; skills and plugins should be abstracted only when a second reuse need arises.
50%UnverifiedThe Claude Agent SDK's nested sub-agent mechanism allows an agent to spawn lower-level agents during execution, but the official documentation explicitly sets limits on the number of levels and quantity.
50%UnverifiedThe Claude Agent SDK introduces structured output, enabling agents to return data with fixed fields according to a predefined schema (such as JSON Schema), making it easy for downstream programs to deserialize and use directly.
50%UnverifiedThe updates to the Claude Agent SDK in the second half of 2026 focus on five areas: nested sub-agents (with quantity limits), SessionStore, skills, plugins, and structured output plus permission defaults.
50%UnverifiedThe Claude Agent SDK lists SessionStore as a standalone capability, moving session state from memory to a manageable storage layer to support state persistence and replayable traces for long tasks.
50%UnverifiedIn the Claude Agent SDK, skills are positioned as capability units that encapsulate reuse of an agent's internal logic, and plugins are positioned as the channel for integrating external tools or services into the agent system.
50%UnverifiedThe Claude Agent SDK's permission defaults follow the principle of least privilege: newly created agents by default have only minimal tool access, and teams must explicitly declare the capabilities they need.
50%UnverifiedSessionStore is not mandatory in the Claude Agent SDK and can be omitted for short tasks; it is recommended to enable it for long tasks or scenarios requiring recovery and auditing.
50%