@cloudkinetix/bmad-enhanced/expansion-packs/ck-llm-agent-dev/workflows/llm-agent-design.yaml

Cloud-Kinetix enhanced fork of BMAD-METHOD - Breakthrough Method of Agile AI-driven Development with robust versioning and unified validation.

name: llm-agent-design
title: LLM Agent Design Workflow
description: Research-driven workflow for designing and architecting AI agents with safety governance
type: design
category: llm-development
estimated_time: 2-3 days

agents:
  - llm-architect
  - llm-engineer
  - llm-safety-governance
  - architect
  - pm

prerequisites:
  - Clear understanding of agent purpose
  - Target user personas defined
  - Integration requirements identified
  - Safety and compliance requirements known

startup_sequence:
  - agent: llm-architect
    task: initial-research
    message: "Beginning AI agent design research and capability analysis"

research_phase:
  - id: 1.1
    agent: llm-architect
    task: domain-research
    outputs:
      - Domain knowledge report
      - Existing solutions analysis
      - Technology landscape
      - Best practices summary
    decision_points:
      - id: D1
        name: Architecture Pattern
        description: Choose agent architecture approach

  - id: 1.2
    agent: llm-architect
    task: capability-mapping
    inputs:
      - Domain research
      - User requirements
    outputs:
      - Required capabilities list
      - Technical feasibility assessment
      - Integration points identification

  - id: 1.3
    agent: llm-safety-governance
    task: safety-requirements
    outputs:
      - Safety guidelines document
      - Risk assessment matrix
      - Compliance requirements
      - Ethical considerations

design_phase:
  - id: 2.1
    agent: llm-architect
    task: create-agent-spec
    inputs:
      - Capability requirements
      - Safety guidelines
    outputs:
      - AI agent specification
      - Architecture design document
      - Interface definitions
    decision_points:
      - id: D2
        name: Model Selection
        description: Choose AI model(s) and approach

  - id: 2.2
    agent: llm-engineer
    task: prompt-engineering-design
    inputs:
      - Agent specification
      - Use case scenarios
    outputs:
      - Prompt templates
      - Chain-of-thought designs
      - Few-shot examples
      - Validation criteria

  - id: 2.3
    agent: architect
    task: system-integration-design
    inputs:
      - Agent architecture
      - Existing system landscape
    outputs:
      - Integration architecture
      - API specifications
      - Data flow diagrams
      - Security design

validation_phase:
  - id: 3.1
    agent: llm-engineer
    task: design-evaluation-suite
    outputs:
      - Test scenario definitions
      - Evaluation metrics
      - Benchmark criteria
      - Edge case catalog

  - id: 3.2
    agent: llm-safety-governance
    task: safety-review
    inputs:
      - Agent design
      - Evaluation suite
    outputs:
      - Safety validation report
      - Risk mitigation strategies
      - Governance checklist
      - Approval recommendations

  - id: 3.3
    agent: pm
    task: stakeholder-review
    outputs:
      - Design review presentation
      - Feedback incorporation
      - Approval documentation
      - Implementation plan

decision_points:
  - id: D1
    step: 1.1
    description: Choose agent architecture pattern
    options:
      - Single-agent with tools
      - Multi-agent orchestration
      - Hybrid human-AI workflow
      - Autonomous agent system
    impacts:
      - System complexity
      - Integration requirements
      - Monitoring needs
      - Safety considerations

  - id: D2
    step: 2.1
    description: Select AI model approach
    options:
      - Single large model
      - Ensemble of specialized models
      - Fine-tuned custom model
      - Hybrid approach
    impacts:
      - Performance characteristics
      - Cost implications
      - Latency requirements
      - Maintenance complexity

  - id: D3
    step: 3.2
    description: Safety governance level
    options:
      - Basic safety checks
      - Comprehensive governance
      - Regulatory compliance
      - Mission-critical standards
    impacts:
      - Development timeline
      - Testing requirements
      - Documentation needs
      - Operational procedures

outputs:
  - Comprehensive AI agent specification
  - Architecture design documents
  - Prompt engineering templates
  - Safety and governance framework
  - Evaluation suite design
  - Integration specifications
  - Risk assessment and mitigations
  - Implementation roadmap

success_criteria:
  - All stakeholders aligned on design
  - Safety requirements fully addressed
  - Technical feasibility validated
  - Integration points clearly defined
  - Evaluation criteria established
  - Governance framework approved

next_steps:
  - Proceed to implementation workflow
  - Set up development environment
  - Begin prompt testing iterations
  - Establish monitoring infrastructure