Deploying Agents on Claude Code

Most people using Claude work in one big context window and hope the model figures it out. That works for simple tasks. It breaks down when you want Claude to review code, query a database, and coordinate a deployment without those concerns bleeding into each other. Subagents solve this by giving each job its own isolated context, its own tools, and its own permissions. This course teaches you how to build them correctly.

You will start by understanding what a subagent actually is and what isolation buys you, then move into the specifics: writing valid .md agent files, scoping tool permissions to the minimum a job needs, wiring MCP servers to specific agents, and writing description fields that make Claude route reliably instead of guessing. These details look small on paper. In production, they are the difference between an agent that works and one that silently takes a wrong path.

The last part of the course covers shipping. A subagent sitting in .claude/agents/ on your laptop helps only you. This unit walks through converting that folder into a plugin, publishing it to a private marketplace, and letting your whole team install it with one command. You will also write CLAUDE.md routing rules that tell the main agent when to delegate and when to stay in the main thread, and learn how to recognize when a task has grown beyond subagents entirely.