Brainstorm

$brainstorm "Build real-time collaboration platform" --count 3
$brainstorm -y "Design payment system" --count 5
$brainstorm "Build notification system" --style-skill material-design
$brainstorm system-architect --session WFS-xxx
$brainstorm ux-expert --include-questions

┌──────────────────────────────────────────────────────────────────┐
│                    BRAINSTORM PIPELINE                            │
├──────────────────────────────────────────────────────────────────┤
│                                                                    │
│  Phase 1: Mode Detection & Routing                                │
│     ├─ Parse flags and arguments                                  │
│     └─ Route to Auto Mode or Single Role Mode                    │
│                                                                    │
│  ═══ Auto Mode ═══                                                │
│                                                                    │
│  Phase 2: Interactive Framework Generation                        │
│     ├─ Context collection → Topic analysis → Role selection       │
│     ├─ Generate guidance-specification.md                         │
│     ├─ Generate roles.csv (1 row per selected role)              │
│     └─ User validates (skip if -y)                               │
│                                                                    │
│  Phase 3: Wave Role Analysis (spawn_agents_on_csv)               │
│     ├─ spawn_agents_on_csv(role instruction template)            │
│     ├─ Each role agent produces analysis.md + sub-documents      │
│     └─ discoveries.ndjson shared across role agents              │
│                                                                    │
│  Phase 4: Synthesis Integration                                   │
│     ├─ Read all role analyses (read-only)                        │
│     ├─ Cross-role analysis → conflict detection                  │
│     ├─ Feature spec generation                                    │
│     └─ Output: feature-specs/ + feature-index.json               │
│                                                                    │
│  ═══ Single Role Mode ═══                                         │
│                                                                    │
│  Phase 3S: Single Role Analysis (spawn_agent)                    │
│     ├─ spawn_agent(conceptual_planning_agent)                    │
│     └─ Output: {role}/analysis*.md                               │
│                                                                    │
└──────────────────────────────────────────────────────────────────┘

roles.csv                      feature-specs/
┌──────────────┐              ┌──────────────────┐
│ R1: sys-arch │──findings───→│ F-001-auth.md    │
│ analysis.md  │              │ (cross-role spec) │
├──────────────┤              ├──────────────────┤
│ R2: ui-design│──findings───→│ F-002-ui.md      │
│ analysis.md  │              │ (cross-role spec) │
├──────────────┤              ├──────────────────┤
│ R3: test-str │──findings───→│ F-003-test.md    │
│ analysis.md  │              │ (cross-role spec) │
└──────────────┘              └──────────────────┘

Two context channels:
1. Directed: role findings → synthesis → feature specs
2. Broadcast: discoveries.ndjson (append-only shared board)

id,role,title,focus,deps,wave,status,findings,output_files,error
"R1","system-architect","系统架构师","Technical architecture, scalability","","1","pending","","",""
"R2","ui-designer","UI设计师","Visual design, mockups","","1","pending","","",""
"R3","test-strategist","测试策略师","Test strategy, quality","","1","pending","","",""

.workflow/active/WFS-{topic}/
├── workflow-session.json              # Session metadata
├── .process/
│   └── context-package.json           # Phase 0 context
├── roles.csv                          # Role analysis state (Phase 2-3)
├── discoveries.ndjson                 # Shared discovery board
└── .brainstorming/
    ├── guidance-specification.md      # Framework (Phase 2)
    ├── feature-index.json             # Feature index (Phase 4)
    ├── synthesis-changelog.md         # Synthesis audit trail (Phase 4)
    ├── feature-specs/                 # Feature specs (Phase 4)
    │   ├── F-001-{slug}.md
    │   └── F-00N-{slug}.md
    └── {role}/                        # Role analyses (Phase 3, immutable)
        ├── {role}-context.md          # Interactive Q&A
        ├── analysis.md                # Main/index document
        ├── analysis-cross-cutting.md  # Cross-feature
        └── analysis-F-{id}-{slug}.md  # Per-feature

const getUtc8ISOString = () => new Date(Date.now() + 8 * 60 * 60 * 1000).toISOString()

// Parse flags
const AUTO_YES = $ARGUMENTS.includes('--yes') || $ARGUMENTS.includes('-y')
const countMatch = $ARGUMENTS.match(/--count\s+(\d+)/)
const roleCount = countMatch ? Math.min(parseInt(countMatch[1]), 9) : 3
const sessionMatch = $ARGUMENTS.match(/--session\s+(\S+)/)
const existingSessionId = sessionMatch ? sessionMatch[1] : null
const skipQuestions = $ARGUMENTS.includes('--skip-questions')
const includeQuestions = $ARGUMENTS.includes('--include-questions')
const styleSkillMatch = $ARGUMENTS.match(/--style-skill\s+(\S+)/)
const styleSkill = styleSkillMatch ? styleSkillMatch[1] : null
const updateMode = $ARGUMENTS.includes('--update')

// Role detection
const VALID_ROLES = [
  'data-architect', 'product-manager', 'product-owner', 'scrum-master',
  'subject-matter-expert', 'system-architect', 'test-strategist',
  'ui-designer', 'ux-expert'
]
const cleanArgs = $ARGUMENTS
  .replace(/--yes|-y|--count\s+\d+|--session\s+\S+|--skip-questions|--include-questions|--style-skill\s+\S+|--update/g, '')
  .trim()
const firstArg = cleanArgs.split(/\s+/)[0]
const isRole = VALID_ROLES.includes(firstArg)

// Mode detection
let executionMode
if (AUTO_YES) {
  executionMode = 'auto'
} else if (isRole) {
  executionMode = 'single-role'
} else if (cleanArgs) {
  executionMode = 'auto'
} else {
  executionMode = null  // Ask user
}

const topic = isRole
  ? cleanArgs.replace(firstArg, '').trim()
  : cleanArgs.replace(/^["']|["']$/g, '')

2. **User Validation** (skip if AUTO_YES)

```javascript
if (!AUTO_YES) {
  console.log(`\n## Brainstorm Framework\n`)
  console.log(`Topic: ${topic}`)
  console.log(`Domain: ${analysis.domain} | Complexity: ${analysis.complexity}`)
  console.log(`\nSelected Roles (${roles.length}):`)
  roles.forEach(r => console.log(`  - [${r.id}] ${r.title}: ${r.focus}`))
  if (features.length > 0) {
    console.log(`\nFeatures (${features.length}):`)
    features.forEach(f => console.log(`  - [${f.id}] ${f.title}`))
  }

  const answer = functions.request_user_input({
    questions: [{
      question: "Approve brainstorm framework?",
      header: "Validate",
      options: [
        { label: "Approve", description: "Proceed with role analysis" },
        { label: "Modify Roles", description: "Change role selection" },
        { label: "Cancel", description: "Abort" }
      ]
    }]
  })

  if (answer.Validate === "Cancel") return
  if (answer.Validate === "Modify Roles") {
    // Allow user to adjust via request_user_input
    const roleAnswer = functions.request_user_input({
      questions: [{
        question: "Select roles for analysis:",
        header: "Roles",
        options: VALID_ROLES.map(r => ({
          label: r,
          description: roles.find(sel => sel.role === r)?.focus || ''
        }))
      }]
    })
    // Rebuild roles[] from selection
  }
}

4. **Generate roles.csv**

```javascript
const header = 'id,role,title,focus,deps,wave,status,findings,output_files,error'
const rows = roles.map(r =>
  [r.id, r.role, r.title, r.focus, '', '1', 'pending', '', '', '']
    .map(v => `"${String(v).replace(/"/g, '""')}"`)
    .join(',')
)
Write(`${sessionFolder}/roles.csv`, [header, ...rows].join('\n'))