Skip to content

Agents

Named personalities, a system prompt, a set of specialties and the tools they carry, that you can mention in a chat or give recurring work to.

Where
AgentsIn the app (sign-in required), opens in a new tab.
Reading time
2 min read

How it works

An agent is a personality: a name and handle, a system prompt, a set of specialties, and the skills and MCP tools it is given. Mention one in a workspace's chat to put it to work, or give it recurring work on the Schedule.

Hire from library is a catalogue of ready-made personalities, project manager, team lead, full-stack engineer, sales, growth, SEO, GEO, content, design, QA, security, DevOps, data. Hiring one copies it into your organization, so tuning its prompt afterwards is yours to do; an improved version of a persona reaches new organizations rather than overwriting an agent you have already adjusted.

MCP servers are registered separately, on Tools → MCP servers, and picked per piece of work rather than attached to the agent. That is deliberate: the same SEO specialist can have web search on the weekly crawl and nothing on the monthly summary, without keeping two copies of the agent. Only the tools a schedule lists exist for that execution, the restriction is real, not advisory.

Runs as picks the coding agent a persona works through, Claude Code, Codex or OpenCode: the CLI in its own container and in the boxes its scheduled tasks get, unless a task picks another. It starts on the organization's default agent.

When a scheduled task runs as an agent, the box is booted as that persona: its prompt is written as the box's CLAUDE.md (which Codex reads as its AGENTS.md), its MCP tools into the agent's own configuration, and its skills into the box's skills directory. Note what that is not: the work never goes to the agent's own long-lived container. That container accumulates state from everything the agent has ever done, and a recurring task's continuity is supposed to come from its briefing, where it can be read and switched off, not from a leftover working directory.

Steps

Following along? Tick each number off as you go.

  1. Open Agents from Automate in the top bar (or the search: Ctrl Ctrl, ⌘ ⌘ on a Mac).

  2. Press Hire from library and pick the personalities your team needs.

    The counter is how many are still un-hired.

  3. Or press New agent and write the prompt and specialties yourself.

  4. Pick Runs as: Claude Code, Codex or OpenCode.

    Its own container and its scheduled tasks' boxes run that agent, unless a task picks another.

  5. Press Edit on an agent to tune its prompt, specialties and skills.

    Yours to change, the library never overwrites it.

  6. Register the MCP servers your agents need on Tools, under MCP servers.

  7. Put an agent to work: mention its handle in a workspace's chat, or give it a recurring task on the Schedule.

    On a schedule you pick the tools that task may use.

  8. Use Pause to take an agent out of service without deleting it.

Worth knowing

  • An agent task cannot name a workspace. Booting a persona inside a workspace's box would overwrite that workspace's own CLAUDE.md for everyone who opens it afterwards.

Read next

AutomationScheduled tasksRecurring work on a calendar, each execution in a fresh box, with a memory of what previous runs found.AutomationShared toolsSkills every workspace's agent can run, with their own environment variables and named credential profiles.Getting startedWorkspace chatThe conversation that lives with the workspace, where agents post, where the team replies, and where @agent puts the workspace's agent to work.
Report a problem