import Anthropic from '@anthropic-ai/sdk'
import type { AIPayload } from '../renderer/types/editor'

let _client: Anthropic | null = null

function getClient(): Anthropic {
  if (!_client) {
    const apiKey = process.env.ANTHROPIC_API_KEY
if (!apiKey || apiKey === 'your-api-key-here') {
      throw new Error(
        'ANTHROPIC_API_KEY is not set. Add it to app/.env.local'
      )
    }
    _client = new Anthropic({ apiKey })
  }
  return _client
}

function buildSystemPrompt(payload: AIPayload): string {
  const chapterName =
    payload.documentPath.split('/').pop()?.replace(/\.md$/, '') ?? 'Unknown chapter'

  const chapterContext = `You are a literary editor assistant helping with a gothic/historical fiction novel set in the Basque Country. The current chapter is: "${chapterName}".

Key characters: Esti, Marko, Garbi, Irati, Cardinal Nikolai, Amaya, Izotz, Sua, Señor Jiménez, the Genboa family.

The chapter text is provided below. When identifying specific passages, quote the EXACT text from the document so it can be located and highlighted in the editor.

--- CHAPTER ---
${payload.documentContent}
--- END CHAPTER ---`

  const modeInstructions: Record<AIPayload['mode'], string> = {
    chat: 'Answer questions about the chapter, characters, plot, or craft. Be specific and cite passages where relevant.',

    passive_voice: `Identify ALL instances of passive voice in this chapter.

For each instance, respond in this exact format:
PASSIVE: "[exact quoted sentence]"
WHY: [brief explanation]
SUGGESTION: "[rewritten in active voice]"

List every instance you find, then give a brief overall summary.`,

    consistency: `Check this chapter carefully for consistency issues:
- Character names spelled or used inconsistently
- Timeline contradictions or impossibilities
- Repeated words or phrases appearing too close together (within a page)
- Setting details that seem contradictory
- Character behaviour inconsistent with their established personality

For each issue found:
ISSUE: [type of issue]
PASSAGE: "[exact quoted text]"
PROBLEM: [explanation]
SUGGESTION: [how to fix it]`,

    style: `Analyze the writing style and provide specific improvement suggestions:
- Pacing: identify slow passages or rushed moments
- Sentence variety: flag runs of similar length or structure
- Show don't tell: identify passages that tell emotion/state rather than showing it
- Gothic atmosphere: passages where the atmospheric tone is inconsistent
- Dialogue: any dialogue that feels stilted or unnatural

For each suggestion:
ISSUE: [type: Pacing / Sentence Variety / Show-Don't-Tell / Atmosphere / Dialogue]
PASSAGE: "[exact quoted text]"
PROBLEM: [specific explanation]
SUGGESTION: [concrete rewrite or approach]`,

    critique: `Give an honest, detailed critique of this chapter as a whole. Structure your response as follows:

**Overall impression** (2–3 sentences on what the chapter achieves and its most significant weakness)

**What works well**
Identify 2–4 specific strengths — scenes, lines, or moments that land effectively. Quote the passage and explain why it works.

STRENGTH: "[exact quoted passage]"
WHY: [explanation]

**What needs work**
Identify 2–4 areas where the chapter falls short. Be direct. Quote the passage and give a concrete direction for improvement.

ISSUE: "[exact quoted passage]"
PROBLEM: [explanation]
SUGGESTION: [concrete direction]

**One priority**
Name the single most important thing to fix in a revision of this chapter.`
  }

  return `${chapterContext}\n\n${modeInstructions[payload.mode]}`
}

export async function streamMessage(
  payload: AIPayload,
  onChunk: (chunk: string) => void
): Promise<void> {
  const client = getClient()

  const messages = [
    ...payload.conversationHistory.slice(-10),
    { role: 'user' as const, content: payload.userMessage }
  ]

  const stream = client.messages.stream({
    model: 'claude-sonnet-4-5',
    max_tokens: 4096,
    system: buildSystemPrompt(payload),
    messages
  })

  for await (const chunk of stream) {
    if (
      chunk.type === 'content_block_delta' &&
      chunk.delta.type === 'text_delta'
    ) {
      onChunk(chunk.delta.text)
    }
  }
}