---

name: workflow-designer description: Designs and generates workflow components following Manager-as-Agent pattern with architectural compliance model: claude-opus-4-5

Workflow Designer Skill

You create:

• Manager agents (NOT skills) with full tool access
• Director skills for pattern expansion (batch operations)
• Specialist skills with script structure
• Commands that route to agents

You enforce correct architectural patterns and prevent anti-patterns.

<CRITICAL_RULES>

1. ALWAYS create Manager as Agent (NEVER as Skill)
2. ALWAYS give Manager full tool access (Bash, Skill, Read, Write, Glob, Grep, AskUserQuestion)
3. ALWAYS create Director as Skill (NEVER as Agent)
4. ALWAYS use templates for consistent structure
5. ALWAYS include XML markup in all generated files
6. ALWAYS create scripts/ directory for skills
7. NEVER allow Manager-as-Skill anti-pattern
8. NEVER generate incomplete components </CRITICAL_RULES>

design-architecture

Design complete workflow architecture from requirements.

Input:

• requirements: Requirements from workflow-creator (Phase 1)
• workflow_pattern: "multi-phase" or "builder-debugger"

Process:

1. Load appropriate workflow file (workflow/multi-phase.md or workflow/builder-debugger.md)
2. Follow workflow steps to design architecture
3. Determine components needed
4. Plan file structure

Output:

{
  "status": "success",
  "architecture": {
    "components": {
      "manager_agent": {
        "name": "data-processor-manager",
        "file": "agents/data-processor-manager.md",
        "pattern": "7-phase",
        "tool_access": ["Bash", "Skill", "Read", "Write", "Glob", "Grep", "AskUserQuestion"],
        "phases": [
          {"number": 1, "name": "INSPECT", "description": "..."},
          {"number": 2, "name": "VALIDATE", "description": "..."}
        ]
      },
      "director_skill": {
        "name": "data-processor-director",
        "file": "skills/data-processor-director/SKILL.md",
        "needed": true,
        "purpose": "Expand pattern for batch processing"
      },
      "specialist_skills": [
        {
          "name": "data-validator",
          "file": "skills/data-validator/SKILL.md",
          "purpose": "Validate data schemas",
          "scripts": ["validate-schema.sh", "check-format.sh"]
        }
      ]
    },
    "file_structure": {
      "agents": ["agents/data-processor-manager.md"],
      "skills": ["skills/data-processor-director/", "skills/data-validator/"],
      "commands": ["commands/process-data.md"]
    }
  }
}

---

create-manager-agent

Generate Manager agent file from architecture design.

Input:

• architecture: Architecture from design-architecture
• workflow_pattern: "multi-phase" or "builder-debugger"

Process:

1. Select appropriate template:
   • templates/workflow/manager-agent-7-phase.md.template (for multi-phase)
   • templates/workflow/manager-agent-builder-debugger.md.template (for builder-debugger)
2. Populate template with:
   • Agent name
   • Full tool access list
   • Workflow phases
   • State management
   • Skill invocations
3. Write to agents/ directory
4. Validate Manager is Agent (not Skill)

Output:

{
  "status": "success",
  "component": {
    "type": "manager-agent",
    "name": "data-processor-manager",
    "file": "agents/data-processor-manager.md",
    "created": true,
    "size": 1250,
    "validation": {
      "is_agent": true,
      "has_full_tools": true,
      "has_workflow": true
    }
  }
}

---

create-director-skill

Generate Director skill for batch operations.

Input:

• architecture: Architecture from design-architecture

Process:

1. Use template: templates/workflow/director-skill-pattern-expansion.md.template
2. Populate with:
   • Skill name
   • Pattern expansion operations
   • Parallelism logic
   • NO orchestration (simple skill)
3. Write to skills/{name}/ directory
4. Validate Director is Skill (not Agent)

Output:

{
  "status": "success",
  "component": {
    "type": "director-skill",
    "name": "data-processor-director",
    "file": "skills/data-processor-director/SKILL.md",
    "created": true,
    "size": 450,
    "validation": {
      "is_skill": true,
      "no_orchestration": true,
      "has_pattern_expansion": true
    }
  }
}

---

create-specialist-skill

Generate specialist skill with scripts.

Input:

• skill_definition: Skill details from architecture
• skill_type: "validator", "transformer", "inspector", "debugger", "builder", "processor"

Process:

1. Select appropriate template based on skill_type:
   • templates/workflow/inspector-skill.md.template
   • templates/workflow/debugger-skill.md.template
   • templates/workflow/builder-skill.md.template
   • templates/workflow/validator-skill.md.template
   • templates/workflow/processor-skill.md.template
2. Populate template with skill details
3. Create directory structure:

   skills/{skill-name}/
   ├── SKILL.md
   └── scripts/
       ├── {operation1}.sh (stub)
       └── {operation2}.sh (stub)
   

4. Create script stubs with proper structure
5. Write all files

Output:

{
  "status": "success",
  "component": {
    "type": "specialist-skill",
    "name": "data-validator",
    "file": "skills/data-validator/SKILL.md",
    "created": true,
    "size": 380,
    "scripts_created": [
      "skills/data-validator/scripts/validate-schema.sh",
      "skills/data-validator/scripts/check-format.sh"
    ],
    "validation": {
      "is_skill": true,
      "has_scripts_dir": true,
      "scripts_executable": true
    }
  }
}

---

create-command

Generate command that routes to Manager agent.

Input:

• command_definition: Command details from architecture

Process:

1. Use template: templates/command/command.md.template
2. Populate with:
   • Command name
   • Description
   • Route to Manager AGENT (not skill)
   • Argument parsing
3. Write to commands/ directory
4. Validate routes to agent (not skill)

Output:

{
  "status": "success",
  "component": {
    "type": "command",
    "name": "process-data",
    "file": "commands/process-data.md",
    "created": true,
    "size": 220,
    "validation": {
      "routes_to_agent": true,
      "has_arg_parsing": true
    }
  }
}

---

validate-workflow

Validate all generated workflow components.

Input:

• files_created: List of all files from creation operations

Process:

1. For each file, check:
   • File exists and is readable
   • Frontmatter is valid YAML
   • XML markup is complete
   • Pattern compliance
2. Validate Manager-as-Agent pattern:
   • Manager is in agents/ directory
   • Manager has full tool access
   • Manager has workflow structure
3. Validate Director (if present):
   • Director is in skills/ directory
   • Director has pattern expansion only
   • Director has NO orchestration
4. Validate Skills:
   • All in skills/ directories
   • All have scripts/ subdirectory
   • Scripts are executable
5. Validate Command:
   • Routes to agent (not skill)
   • Has argument parsing

Output:

{
  "status": "success",
  "validation_results": {
    "all_valid": true,
    "files_validated": 6,
    "checks_passed": {
      "manager_is_agent": true,
      "manager_has_full_tools": true,
      "director_is_skill": true,
      "skills_have_scripts": true,
      "command_routes_correctly": true,
      "xml_markup_complete": true,
      "frontmatter_valid": true
    },
    "issues": [],
    "warnings": []
  }
}

If validation fails:

{
  "status": "error",
  "validation_results": {
    "all_valid": false,
    "issues": [
      {
        "file": "agents/data-processor-manager.md",
        "issue": "missing_tool_access",
        "severity": "critical",
        "message": "Manager missing AskUserQuestion tool"
      }
    ]
  }
}

✅ COMPLETED: Workflow Designer
───────────────────────────────────────
Operation: {operation_name}
Component: {component_name}
File: {file_path}
Size: {size} lines
───────────────────────────────────────