Agents and teams

From one assistant to a team. An agent is Claude solving a task on its own, step by step. Here you get specialised agents, fixed workflows and control that does not depend on luck.

Track
2 · Work
Duration
about 5 hours
Modules
5 modules
Status
In development 1 of 5 modules ready

Who it is for

Developers, power users and IT who know sessions and the project's memory from course 2.

Modules

  1. Module 3.1Planned

    Subagents, skills and hooks

    What you learn
    The three building blocks: a subagent is a specialist with its own task, a skill is a fixed recipe, and a hook is a small program that runs automatically on certain events, for example to stop a dangerous command.
    Exercise
    Make one subagent, one skill and one hook in the practice project.
    What you have afterwards
    Three building blocks that work, and you know when to use each.
  2. Module 3.2Ready

    The AI team

    What you learn
    A finished team: verifier, root-cause analyst and user tester, the /hold command and a block against dangerous commands.
    Exercise
    Install the team with the three small programs and run the team on a real task.
    What you have afterwards
    A team in your own project with automatic checks on GitHub (the service where the code is stored and checked).
    1. Step 1Presentation
    2. Step 2Before you start
    3. Step 3Install the team
    4. Step 4Connect GitHub
    5. Step 5Playbook
    6. Step 6System map

    Open in the client portal The material is in the client portal for clients with access.

  3. Module 3.3Planned

    Connections to your systems

    What you learn
    MCP is the standard that lets Claude use your systems (cases, customers, stock). A plugin packages agents, skills and connections so they can be shared.
    Exercise
    Connect a practice database and package the setup as a plugin.
    What you have afterwards
    A connection and a package your colleagues can install.
  4. Module 3.4Planned

    Parallel work

    What you learn
    Several agents on the same project without getting in each other's way: separate working copies and a clear division of tasks.
    Exercise
    Let three agents solve three parts of the same task and merge the result.
    What you have afterwards
    A division of work you can repeat.
  5. Module 3.5Planned

    Automation without a screen

    What you learn
    Run Claude from a script (a file of commands), in GitHub Actions (GitHub's automation that runs on every change) and as code review.
    Exercise
    Set up automatic review on the practice project.
    What you have afterwards
    An automatic check on every change.

    Note: Automatic review on GitHub (Code Review) requires Team or Enterprise. Review on your own computer and GitHub Actions work on every plan with Claude Code.

Want to use the academy?

Tell me which roles you have and where you stand today. I will suggest a path through the courses: self-study, a workshop or a complete programme.