name: "Product Requirements Review"
description: "Multi-role PRD review — PM analyzes requirements, architect assesses tech, UX researcher assesses experience, then synthesis"

agents_dir: "agency-agents"

llm:
  provider: "deepseek"
  model: "deepseek-chat"
  max_tokens: 4096
  timeout: 120000
  retry: 3

concurrency: 2

inputs:
  - name: prd_content
    description: "PRD document content"
    required: true

steps:
  - id: analyze
    role: "product/product-manager"
    task: |
      Analyze the following PRD and output a structured review.

      PRD:
      {{prd_content}}

      Format (markdown):

      ### Core Requirements (ranked P0/P1/P2)
      - P0: ...
      - P1: ...

      ### Target User Persona
      User type | Core pain point | Use case

      ### Risks & Uncertainties
      - [high/medium/low] risk description → suggested validation

      ### Key Success Metrics
      - Metric | Target | How measured
    output: requirements

  - id: tech_review
    role: "engineering/engineering-software-architect"
    task: |
      Based on the requirements analysis, assess technical feasibility.

      Requirements:
      {{requirements}}

      Format (markdown):

      ### Feasibility Score: [high / medium / low]

      ### Recommended Architecture
      Core components and interactions (3-5 sentences)

      ### Build Complexity
      | Module | Complexity (1-5) | Est. person-days | Main challenges |

      ### Technical Risks
      - [high/medium/low] risk → mitigation
    output: tech_report
    depends_on: [analyze]

  - id: design_review
    role: "design/design-ux-researcher"
    task: |
      Based on the requirements analysis, evaluate UX.

      Requirements:
      {{requirements}}

      Format (markdown):

      ### UX Risks
      - [high/medium/low] risk → scope of impact

      ### Usability Issues
      - Issue → proposed solution

      ### Top 3 Design Recommendations
      1. Recommendation → expected impact

      ### Assumptions to Validate
      - Assumption → validation method (user interviews / A-B test / prototype)
    output: design_report
    depends_on: [analyze]

  - id: final_summary
    role: "product/product-manager"
    task: |
      Synthesize tech and design feedback into a final verdict.

      ## Tech Review
      {{tech_report}}

      ## Design Review
      {{design_report}}

      Format (markdown):

      ### Verdict: [✅ Approved / ⚠️ Needs changes / ❌ Rejected]

      ### Must Resolve (before merge)
      1. [source: tech/design] issue → proposed fix

      ### Suggested Improvements (later iteration)
      1. Issue → priority

      ### Next Actions
      - [ ] Action item | Owner | Suggested timeline
    output: final_report
    depends_on: [tech_review, design_review]