Skip to main content
Agent commands manage autonomous agent definitions. Agents are separate from human-driven harness artifacts.

Commands

list

Lists agents visible to the authenticated identity. Usage
Searches every accessible workspace unless --workspace narrows the request. Usage

get

Shows one agent by its stable agent_... ID. Usage

validate

Validates one local YAML or JSON agent definition without writing to the Registry. Usage

create

Creates an agent from a local YAML or JSON definition. Usage

update

Updates one exact agent from a local definition. Names are not unique, so the stable ID is required. Usage

Examples

Output and behavior

  • list, search, and get return human output unless --json, --jq, or --template is supplied.
  • validate rejects unknown fields and invalid tool definitions locally.
  • create and update prompt before writing unless --auto-approve is used.
  • update reads the current content hash and refuses an unconditional update.
  • Agent creation does not print or store the Registry’s reveal-once machine credential.