Skip to main content

Type

agent_communication

Variants

Read Only, Limited, Full Orchestration
Purpose: The Agent Communication skill enables hierarchical agent orchestration, allowing agents to delegate tasks to specialist agents or coordinate multi-agent workflows.

​
Common Use Cases

Delegating to specialist agentsRoute tasks to agents with specific expertise or capabilities

Orchestrating multi-agent workflowsCoordinate complex workflows across multiple specialized agents

Calling team-based executionsExecute tasks using team agents with shared capabilities

Session continuationResume or extend existing agent sessions

​
Variants Overview

VariantSecurityKey PermissionsBest ForCreate Command
Read Only 🟢SafeStatus monitoring onlyObservability, audit--variant read_only
Limited 🟡RecommendedWhitelisted agents, depth=2Standard orchestration--variant limited
Full Orchestration 🔴UnrestrictedAny agent, any team, depth=5Complex workflows--variant full_orchestration
Choosing a variant: Use Limited for most scenarios with whitelist-based control. Only upgrade to Full Orchestration when unrestricted delegation is required. See Variant Configuration for detailed differences.

​
Configuration

Example Configuration: 
{
  "allowed_operations": ["execute_agent", "get_execution_status"],
  "allowed_agents": ["deploy-agent", "test-agent"],
  "max_execution_depth": 2,
  "timeout": 600,
  "max_concurrent_calls": 3
}
ParameterTypeDefaultDescription
allowed_operationsarrayvariant-specificPermitted operations
allowed_agentsarray[”*“]Whitelisted agent IDs
allowed_teamsarray[]Whitelisted team IDs
max_execution_depthnumbervariant-specificMaximum delegation depth
timeoutnumber600Execution timeout (seconds)
max_concurrent_callsnumber3Concurrent agent executions
enable_session_continuationbooleantrueAllow resuming sessions
session_timeoutnumber1800Session expiry (seconds)
Read Only:
  • allowed_operations: [“get_execution_status”] (locked)
  • No execution capability
Limited:
  • max_execution_depth: 2 (recommended)
  • Requires whitelist configuration
  • max_concurrent_calls: 3
Full Orchestration:
  • allowed_operations: [”*”] (all operations)
  • allowed_agents: [”*”] (all agents)
  • max_execution_depth: 5
  • max_concurrent_calls: 10
See: Variant Configuration Guide

​
Quick Start

# Create skill with variant
kubiya skill create --name "Orchestrator" --type agent_communication --variant limited --enabled

# Associate with agent
kubiya skill associate agent <agent-id> <skill-id>

View Complete Examples

See full orchestration patterns, deployment workflows, and multi-agent coordination examples

​
Agent Execution Examples

​
Single Agent Call

Orchestrator Request: 
"Deploy the backend service to production"
Orchestrator Action: 
result = execute_agent(
    agent_id="backend-deploy-agent",
    input={
        "environment": "production",
        "service": "api-backend",
        "version": "v1.2.3"
    }
)

​
Sequential Execution

Orchestrator Request: 
"Run tests, then deploy if they pass"
Orchestrator Actions: 
# Step 1: Run tests
test_result = execute_agent(agent_id="test-agent", input={"test_suite": "integration"})

# Step 2: Deploy if tests passed
if test_result["status"] == "success":
    execute_agent(agent_id="deploy-agent", input={"version": test_result["version"]})

​
Team Execution

Orchestrator Request: 
"Have the SRE team investigate the latency issue"
Orchestrator Action: 
result = execute_team(
    team_id="sre-team",
    task="Investigate API latency spike in production",
    context={"service": "payment-api", "latency_p99": "2500ms"}
)

​
Security Features

Circular Call DetectionPrevents infinite loops where agents call each other recursively

Execution Depth LimitingLimits how deep agent call chains can go

Whitelist-based AccessOnly allows calling explicitly permitted agents and teams

Concurrent LimitsPrevents resource exhaustion by limiting parallel calls

​
Best Practices

Always specify allowed_agents rather than using ”*” wildcards.
"allowed_agents": ["agent-1", "agent-2"]  # Explicit whitelist
# NOT: allowed_agents: ["*"]
Keep max_execution_depth low (2-3) to prevent complex call chains.
"max_execution_depth": 2  # Sufficient for most orchestration
Configure max_concurrent_calls to prevent resource exhaustion.
"max_concurrent_calls": 3  # Balance parallelism and resources
Best Practice: Set max_execution_depth to prevent infinite loops. A depth of 2-3 is sufficient for most orchestration scenarios.
Solutions:
  • Verify agent ID is in allowed_agents list
  • Check execute_agent is in allowed_operations
  • Confirm target agent exists and is enabled
Solutions:
  • Increase max_execution_depth if appropriate
  • Restructure orchestration to reduce call chain depth
  • Consider using workflow executor for complex chains
Solutions:
  • Review agent call chain for circular references
  • Restructure orchestration logic to avoid loops
  • Use different agents for different stages

Workflow Executor

Alternative for complex multi-step workflows

Contextual Awareness

Share context between orchestrated agents

Knowledge API

Agents can share knowledge base access

View All Skills

Return to built-in skills overview