What it is. A capability listed as an agent template is not the agent itself — it's the scaffold: the system prompt, the list of tools the agent is allowed to call, retry and planning behavior, and a working example. Frameworks like AutoGen, OpenAI Agents SDK, LangGraph, and CrewAI all publish agent templates you can copy into your own project.
When a template fits. Reach for a template when the task has a well-known shape — research assistant, conversation agent, code reviewer — and you'd otherwise be reimplementing the same tool loop yourself. Templates let you skip the plumbing and start iterating on the task-specific bits.
What to check. Templates are opinionated. Read the default tool list, model choice, and permission scope before adopting — a research template that assumes web access won't fit an air-gapped environment. AgentMaps flags each template's runtime, protocol, and required permissions on the detail page so you can spot the mismatch before you install.