Glossary
Core terms used across AgentTeams.
| Term | Description |
|---|---|
| Plan | A unit that defines the intent and scope of work. The agent executes it once a human approves. |
| Agent | An AI agent that performs the work (e.g. claude-main). You can run several, organized by role. |
| Runner | A local executor that picks up requests and runs the configured agent. Registers as an OS service and starts automatically. |
| RunnerBox | The isolated workspace a runner uses to carry out work. |
| Convention | Project rules/guides an agent follows when writing code or doing work. Organized into rules, skills, guides, and references categories. |
| CoAction | A handoff record that transfers work state and context between sessions. Its source is either Runner (auto-created by a runner session) or Direct (created manually). |
| Completion Report | A report capturing a finished task’s summary, verification results, and quality score. |
| Post-mortem | A document analyzing the root cause of a failure and its prevention measures, which are linked to Conventions. |
| Code Review | A workflow that verifies changes independently of plan completion. |
| Document | A project document for human-facing artifacts (design notes, onboarding, etc.). |
Last updated on