Skip to content

Agent Capabilities

Capabilities control what "special" operations an agent is allowed to perform. You can think of them as a special kind of tools with "extra superpowers", having access to the internals. They are defined in the agent's configuration:

agents:
  my_agent:
    capabilities:
      # Agent Discovery & Delegation
      can_list_agents: true        # Discover other available agents
      can_delegate_tasks: true     # Delegate tasks to other agents
      can_observe_agents: false    # Monitor other agents' activities
      can_ask_agents: true    # Can ask other agents in the pool

      # History & Statistics Access
      history_access: "own"        # "none" | "own" | "all"
      stats_access: "none"         # "none" | "own" | "all"

      # Resource Management
      can_load_resources: true     # Load and access resource content
      can_list_resources: true     # Discover available resources

      # Tool Management
      can_register_tools: false    # Register importable functions as tools
      can_register_code: false     # Create new tools from provided code
      can_install_packages: false  # Install Python packages for tools
      can_chain_tools: true       # Chain multiple tool calls into one

      # Code Execution (Use with caution)
      can_execute_code: false     # Execute Python code (no sandbox)
      can_execute_commands: false # Execute CLI commands

      # Agent Creation
      can_create_workers: true    # Create worker agents as tools
      can_create_delegates: true  # Spawn temporary delegate agents
      can_add_agents: true  # Create new persistent agents in the pool

Capability Details

Agent Discovery & Delegation

  • can_list_agents: Allows discovering other agents in the pool
  • can_delegate_tasks: Enables sending tasks to other agents
  • can_observe_agents: Permits monitoring of other agents' activities
  • can_ask_agents: Can ask other agents in the pool

History & Statistics Access

  • history_access: Controls access to conversation history
  • "none": No access to history
  • "own": Can access own conversation history

  • "all": Can access all agents' history

  • stats_access: Controls access to usage statistics

  • "none": No access to statistics
  • "own": Can view own usage statistics
  • "all": Can view all agents' statistics

Resource Management

  • can_load_resources: Allows loading content from resources
  • can_list_resources: Permits discovery of available resources

Tool Management

  • can_register_tools: Allows registering Python functions as tools
  • can_register_code: Permits creating new tools from provided code
  • can_install_packages: Enables installing Python packages
  • can_chain_tools: Allows chaining multiple tool calls together

Execution Capabilities

Security Warning

These capabilities provide direct code execution and should be used with caution: - can_execute_code: Enables Python code execution (no sandbox) - can_execute_commands: Allows CLI command execution

Agent Creation

  • can_create_workers: Allows creating worker agents as tools
  • can_create_delegates: Enables spawning temporary delegate agents
  • can_add_agents: Allows creating new persistent agents in the pool

Default Configuration

By default, all capabilities are set to false or "none" for security. Enable only the capabilities that your agent needs.

Example Use Cases

Read-Only Agent

agents:
  reader:
    capabilities:
      can_load_resources: true
      history_access: "own"
      # All other capabilities: false

Task Coordinator

agents:
  coordinator:
    capabilities:
      can_list_agents: true
      can_delegate_tasks: true
      can_observe_agents: true
      history_access: "all"
      stats_access: "all"

Tool Power User

agents:
  power_user:
    capabilities:
      can_register_tools: true
      can_chain_tools: true
      can_load_resources: true
      can_list_resources: true