ims/app/api/incidents/ai/quality-check/route.ts

export const dynamic = 'force-dynamic'

import { NextRequest, NextResponse } from 'next/server'
import { createClient } from '@/lib/supabase/server'
import { createAnthropicClient } from '@/lib/claude/client'
import { getApiKey } from '@/lib/settings'

export async function POST(request: NextRequest) {
  const supabase = await createClient()
  const { data: { user }, error: authError } = await supabase.auth.getUser()
  if (authError || !user) return NextResponse.json({ error: 'Unauthorized' }, { status: 401 })

  const since = new Date(Date.now() - 60_000).toISOString()
  const { count: recentCount } = await supabase
    .from('audit_log')
    .select('id', { count: 'exact', head: true })
    .eq('changed_by', user.id)
    .eq('action', 'ai_quality_check')
    .gte('changed_at', since)
  if ((recentCount ?? 0) > 0)
    return NextResponse.json({ error: 'Rate limited — please wait 60 seconds' }, { status: 429 })

  const anthropicKey = await getApiKey(supabase, 'ANTHROPIC_API_KEY')
  const anthropic = createAnthropicClient(anthropicKey)

  let body: { description?: string; incident_type?: string }
  try {
    body = await request.json()
  } catch {
    return NextResponse.json({ error: 'Invalid JSON body' }, { status: 400 })
  }
  if (!body.description || !body.incident_type) {
    return NextResponse.json({ error: 'description and incident_type required' }, { status: 422 })
  }

  let message: Awaited<ReturnType<typeof anthropic.messages.create>>
  try {
    message = await anthropic.messages.create({
      model: 'claude-opus-4-8',
      thinking: { type: 'adaptive' },
      max_tokens: 1024,
      tools: [{
        name: 'assess_quality',
        description: 'Assess HSE incident report description quality',
        input_schema: {
          type: 'object' as const,
          properties: {
            score: { type: 'number', description: '1-10 quality score' },
            passes: { type: 'boolean', description: 'True when score is 6 or above' },
            feedback: { type: 'string', description: 'One-sentence quality summary' },
            suggestions: {
              type: 'array',
              items: { type: 'string' },
              description: 'Up to 3 concrete suggestions to improve the description',
            },
          },
          required: ['score', 'passes', 'feedback', 'suggestions'],
        },
      }],
      tool_choice: { type: 'tool', name: 'assess_quality' },
      messages: [{
        role: 'user',
        content: `You are an HSE reporting assistant for a Malaysian 3PL warehouse. Assess this incident report description.

Incident type: ${body.incident_type}
Description: ${body.description}

Score 1–10 based on: specificity (location, time, persons involved), completeness (what happened + immediate actions), and clarity. Score 6 or above passes. If score is below 6, give up to 3 actionable suggestions.`,
      }],
    })
  } catch {
    return NextResponse.json({ error: 'AI service unavailable' }, { status: 503 })
  }

  const toolBlock = message.content.find(b => b.type === 'tool_use')
  if (!toolBlock || toolBlock.type !== 'tool_use') {
    return NextResponse.json({ error: 'AI assessment failed' }, { status: 500 })
  }

  const input = toolBlock.input as {
    score?: unknown
    passes?: unknown
    feedback?: unknown
    suggestions?: unknown
  }
  if (
    typeof input.score !== 'number' ||
    typeof input.passes !== 'boolean' ||
    typeof input.feedback !== 'string' ||
    !Array.isArray(input.suggestions)
  ) {
    return NextResponse.json({ error: 'AI returned unexpected structure' }, { status: 500 })
  }

  await supabase.rpc('write_audit_log', {
    p_table_name: 'incidents',
    p_record_id: user.id,
    p_action: 'ai_quality_check',
    p_new_value: { score: input.score, passes: input.passes, model: 'claude-opus-4-8' } as never,
  })

  return NextResponse.json(input)
}