Skip to content

Workflow Orchestration Component

Purpose

Executes multi-step workflows with advanced templating, state tracking, and dynamic tool generation. The workflow engine transforms complex, multi-step platform operations into simple, single-command executions.

Key Responsibilities

  • Workflow Definition Management: Parse and validate YAML workflow definitions
  • Dynamic Tool Generation: Generate workflow_<name> execution tools for each defined workflow
  • Step Execution: Execute workflow steps with dependency management and error handling
  • Template Processing: Advanced argument templating with variable substitution
  • Execution Tracking: Persistent execution state and comprehensive history
  • Cross-Tool Integration: Orchestrate core tools, external MCP tools, and services

Component Architecture

Key Files

  • executor.go: Workflow execution engine with step orchestration
  • execution_tracker.go: Real-time execution state management
  • execution_storage.go: Persistent execution history and results
  • api_adapter.go: API integration for tool access

Tool Generation Pattern

Each workflow definition automatically generates a corresponding execution tool:

.muster/workflows/connect-monitoring.yaml → workflow_connect-monitoring
.muster/workflows/check-cilium-health.yaml → workflow_check-cilium-health
.muster/workflows/login-management-cluster.yaml → workflow_login-management-cluster

Advanced Features

Template Processing

  • Input Variables: {{ .input.argName }} for workflow arguments
  • Step Results: {{ .results.stepId.fieldName }} for step result chaining — every step's result is referenceable by later steps regardless of any flag
  • One Expression Language: step args, condition templates, and condition jsonPath expectations all use the same Go-template/sprig engine and the same path navigator (dotted paths with array indexing, e.g. items[0].name)
  • Default Values: Built-in support for argument defaults
  • Type Validation: Schema validation for all arguments

Result Shaping

  • output flag: marks the steps whose results are included in the returned document (the deprecated store flag is a backwards-compatible alias)
  • output template: a workflow-level templated object rendered against .input / .results / .vars after all steps complete, returned in place of the default response to produce a small, structured result (JSON types are preserved)

Execution Management

  • Execution History: Full audit trail with core_workflow_execution_list
  • State Tracking: Real-time execution progress monitoring
  • Error Recovery: Configurable retry logic and error handling
  • Resource Cleanup: Automatic cleanup of temporary resources

Tool Integration Patterns

Workflows seamlessly orchestrate across tool types:

steps:
  # External MCP tool usage
  - id: "kubernetes-action"
    tool: "x_kubernetes_port_forward"

  # Conditional execution
  - id: "health-check"
    tool: "core_service_status"
    args:
      name: "monitoring-service"

Integration with muster Ecosystem

Service Management Integration

Workflows orchestrate services: - Monitor health with core_service_status - Manage lifecycle with start/stop/restart operations

MCP Server Coordination

Workflows can coordinate multiple external MCP servers: - Infrastructure management (Kubernetes) - Monitoring setup (Prometheus, Grafana) - GitOps operations (Flux)

Configuration Persistence

All workflow definitions persist in .muster/workflows/ enabling: - Version control of operational procedures - Team sharing of standardized workflows - Consistent execution across environments

This architecture transforms complex, error-prone manual procedures into reliable, repeatable, single-command operations that AI agents can execute consistently.