Master the terminology behind building custom agents with the Claude Agent SDK.
0 / 5 completed
1 / 5
At standup, a dev wants to build a custom autonomous agent in a Node/TypeScript backend using Claude, with tool use and multi-turn loops. Which fits?
The Claude Agent SDK provides TypeScript building blocks for constructing agents with tool use, multi-turn reasoning loops, and context management, rather than requiring a team to hand-roll that orchestration around raw API calls. It targets developers embedding agentic behavior into their own applications. This differs from a one-off prompt-response wrapper.
2 / 5
During a design review, the team wants the agent to call a custom internal function (e.g., looking up an order) mid-conversation. Which SDK concept enables this?
The SDK lets developers register a tool with a name, schema, and handler, which the agent can invoke when it decides that action serves the user's request. This is how the agent takes real actions beyond generating text. Well-defined tool schemas make the agent's capabilities predictable.
3 / 5
In a code review, a dev notices the agent kept prior tool results in context across several turns of a task. What supports this?
The SDK manages session state, carrying prior messages and tool results forward so the agent can reason over accumulated context across a multi-step task. Without this, each turn would start from a blank slate. Statefulness is essential for tasks that require several dependent steps.
4 / 5
An incident report shows an agent-triggered tool call ran an unintended write operation against a production system. What is the recommended mitigation?
Tools should be scoped to the minimum necessary permissions, and high-risk actions like writes to production should require explicit confirmation or human-in-the-loop review. Broad default access increases the blast radius of an incorrect agent decision. This mirrors least-privilege principles applied to agentic tool use.
5 / 5
During a PR review, a teammate asks how the Agent SDK differs from calling the raw Messages API directly. What is the distinction?
The Agent SDK builds a higher-level orchestration layer, handling tool-call loops and session state, on top of the same underlying Messages API a developer could call directly. Using the raw API means implementing that orchestration manually. The SDK exists to reduce that boilerplate for agentic use cases.