Packaging and Hosting: Agent Skills and Managed Agents
Explore the concepts of packaging reusable agent skills and deploying AI agents with Claude Managed Agents. Understand how to design and document workflows that can be reused across customers and decide between Anthropic-hosted or self-hosted agent runtimes. This lesson helps you develop deployment strategies that balance customer needs, infrastructure management, and compliance considerations.
Anthropic's FDE job posting lists agent skills among the technical deliverables and emphasizes documenting reusable deployment patterns. These responsibilities raise two architectural questions beyond implementing a single agent. The first concerns reuse: "How would you package this workflow for another customer?" The second concerns deployment: "Where should the agent run?"
Claude provides options for addressing both concerns. Agent Skills package reusable instructions, workflows, and supporting resources. Claude Managed Agents provides an Anthropic-hosted agent runtime, reducing the need to operate the agent loop and its supporting infrastructure.
The following examples continue the expense audit scenario. First, we package the audit checklist as a reusable skill. Then, we deploy an expense audit agent using Claude Managed Agents.
What an agent skill is
An agent skill is a folder that packages instructions and supporting resources for one kind of task. Its core is a SKILL.md file with two parts.
Frontmatter: A short YAML header with the skill’s
nameand adescriptionof when to use it.Instructions: Markdown text that tells Claude how to do the task, such as a checklist and an output format.
A skill folder can also hold scripts, templates, and reference files that the instructions point to.
Claude loads skills in stages. At the start of a session, it sees only each skill’s name and ...