Orchestrator

Orchestrator Agent - Skill Definition

NOTE: Cursor uses the canonical Skills in .claude/skills/.
This file is framework documentation; prefer .claude/skills/orchestrator/SKILL.md for Cursor.

Purpose

The Orchestrator Agent coordinates YAML-defined workflows and makes gate decisions. It is responsible for:

• Loading and executing workflows from YAML definitions
• Making gate decisions based on scoring and conditions
• Routing to appropriate workflows (Greenfield/Brownfield)
• Tracking workflow state and progress
• Coordinating agent execution within workflows

Permissions

• Read: ✅
• Write: ❌
• Edit: ❌
• Grep: ✅
• Glob: ✅
• Bash: ❌

Type: Read-only agent (coordination only, no code modification)

Commands

*workflow-list

List all available workflows in the workflows/ directory.

Example:

tapps-agents orchestrator *workflow-list

Returns:

{
  "workflows": [
    {
      "id": "example-feature-development",
      "name": "Example Feature Development Workflow",
      "description": "Standard workflow for new feature implementation",
      "version": "1.0.0",
      "type": "greenfield",
      "file": "workflows/example-feature-development.yaml"
    }
  ]
}

*workflow-start {workflow_id}

Start a workflow by ID.

Example:

tapps-agents orchestrator *workflow-start example-feature-development

Returns:

{
  "success": true,
  "workflow_id": "example-feature-development",
  "workflow_name": "Example Feature Development Workflow",
  "status": "running",
  "current_step": "requirements",
  "message": "Workflow 'Example Feature Development Workflow' started"
}

*workflow-status

Get the current workflow execution status.

Example:

tapps-agents orchestrator *workflow-status

Returns:

{
  "workflow_id": "example-feature-development",
  "status": "running",
  "current_step": "planning",
  "current_step_details": {
    "id": "planning",
    "agent": "planner",
    "action": "create_stories"
  },
  "completed_steps": ["requirements"],
  "skipped_steps": [],
  "artifacts_count": 1,
  "can_proceed": true
}

*workflow-next

Get information about the next step in the workflow.

Example:

tapps-agents orchestrator *workflow-next

Returns:

{
  "next_step": {
    "id": "design",
    "agent": "architect",
    "action": "design_system",
    "context_tier": 2,
    "requires": ["requirements.md", "stories/"],
    "creates": ["architecture.md"]
  }
}

*workflow-skip {step_id}

Skip an optional step in the workflow.

Example:

tapps-agents orchestrator *workflow-skip review

Returns:

{
  "success": true,
  "message": "Step 'review' skipped",
  "current_step": "testing"
}

*workflow-resume

Resume an interrupted workflow (loads state from persistence).

Example:

tapps-agents orchestrator *workflow-resume

*gate {condition}

Make a gate decision based on condition and scoring data.

Example:

tapps-agents orchestrator *gate --condition "scoring.passed == true" --scoring-data '{"passed": true, "overall_score": 85}'

Returns:

{
  "passed": true,
  "condition": "scoring.passed == true",
  "scoring": {
    "passed": true,
    "overall_score": 85
  },
  "message": "Gate passed"
}

*help

Show help for orchestrator commands.

Workflow Integration

The Orchestrator Agent integrates with the Workflow Engine to:

1. Load Workflows: Parse YAML workflow definitions
2. Execute Steps: Coordinate agent execution for each step
3. Track Artifacts: Monitor artifact creation and dependencies
4. Make Gates: Evaluate conditions and scoring to determine workflow progression
5. Handle Branching: Route to different steps based on gate decisions

Gate Decision Logic

The orchestrator evaluates gate conditions using:

• Scoring Data: Results from reviewer agent (overall_score, passed, etc.)
• Conditions: String expressions like "scoring.passed == true" or "overall_score >= 70"
• Thresholds: Minimum scores for passing gates

Gate Outcomes:

• Pass: Workflow proceeds to on_pass step
• Fail: Workflow loops back to on_fail step (typically for retry)

Context Tier Usage

The orchestrator uses Tier 1 context (minimal) since it only coordinates workflows and doesn't analyze code.

Integration with Other Agents

The orchestrator coordinates:

• Planner: Creates stories and plans
• Architect: Designs system architecture
• Implementer: Generates code
• Reviewer: Reviews code and provides scoring
• Tester: Generates and runs tests
• Debugger: Analyzes errors
• Documenter: Generates documentation

Example Workflow Execution

# workflows/example-feature-development.yaml
workflow:
  id: example-feature-development
  steps:
    - id: review
      agent: reviewer
      action: review_code
      gate:
        condition: "scoring.passed == true"
        on_pass: testing
        on_fail: implementation

Execution Flow:

1. Orchestrator starts workflow
2. Reviewer agent reviews code
3. Orchestrator evaluates gate: scoring.passed == true
4. If passed → proceed to testing
5. If failed → loop back to implementation