Agent
Design agent behavior, lifecycle hooks, context, tools, and runtime policies.
Agent docs cover the server-side behavior that belongs to an agent definition: model behavior, instructions, tools, lifecycle hooks, context, and runtime policies.
Start with Context when you need runtime-only dependencies for models, instructions, tools, and lifecycle hooks.
Use Skills for reusable instruction packages that load progressively through hidden backend tools.
Use Tools And Human Input for backend tools, typed suspend/resume wait points, client-provided capabilities, and tool state.
Use Calling Agents when a tool should run another registered agent as part of the current run — your code picks the agent.
Use Subagents when the model should pick the specialist, by name, from a roster you declare on one delegation tool.
Use Commands for typed, thread-scoped server actions that clients can call through the agent contract.
Use Lifecycle Hooks when you need to customize model steps, tool calls, or persisted output.