Agents and teams

Agents

An agent is a named role with a system prompt, the skills and playbooks it reads, the sources and record types it may use, a model, a budget and a place in the hierarchy (parent). Agents answer in chat, run missions, file asks and propose actions.

The workspace lead (lead: in workspace.yaml) answers anything not routed elsewhere.
Routing: a message goes to the agent that handles it; the reply says which agent answered.
Every agent gets built-in tools (web search, fetch, crawl, image generation, code execution) plus tools for the sources in its scope. harness.grantTools adds powerful ones; harness.excludeTools removes.
Where the agent loop runs is set by harness.runsOn — in the app, in a container on Amazon Bedrock AgentCore, or as an external worker. See Models and agent execution.

Teams

A team groups agents under a lead, with a goal, the measures it is judged on and an accountable person. The team report is at /dashboard/team-report.

Skills and playbooks

A skill (skills/<slug>/SKILL.md) is a procedure in markdown with frontmatter. The agent reads it when its judgment says the task calls for it.
A playbook (playbooks/<slug>/SKILL.md, same shape) is standing context attached to an agent or skill by name.

Skills replaced the old typed "operations"; a workspace that still has an operations/ folder fails to load with a message telling you to convert it.

Missions

A mission is a standing responsibility owned by one agent: a goal, success criteria and an autonomy level. An automation decides when it is checked. See Missions, workflows and automations.

The base pack

Core ships a base pack (core@2.1.0: nine agents, thirteen skills) you can pin with extends: and activate with use:. Plugins ship their own agents the same way.