// =============================================================================
// Question Generation Functions
// Generate questions for packs with various tone settings
// =============================================================================

import { getOpenAI, getModel, getTemperature } from '../client';
import { TONE_INSTRUCTIONS, SYSTEM_MESSAGES, CORE_LANGUAGE_RULES } from '../config';
import type { GeneratedQuestion, ToneLevel } from '../types';

/**
 * Build the question generation prompt
 */
function buildQuestionPrompt(
    packName: string,
    count: number,
    tone: ToneLevel,
    packDescription?: string,
    existingQuestions?: string[],
    crudeLang: boolean = false,
    inspiration?: string
): string {
    const crudeLangInstruction = crudeLang
        ? '\n\nLEGACY CRUDE-LANGUAGE OPTION: Do not follow it. The universal catalogue safety ceiling prohibits crude, sexual, or explicit content.'
        : '';

    // Sanitize inspiration to prevent prompt injection (remove XML-like tags and authority phrases)
    const sanitizedInspiration = inspiration
        ? inspiration
            .replace(/<[^>]*>/g, '') // Remove XML-like tags
            .replace(/\b(CRITICAL|PRIORITY|INSTRUCTION|OVERRIDE|IGNORE|SYSTEM|ASSISTANT)\b:?/gi, '') // Remove authority words
            .trim()
        : '';

    const inspirationInstruction = sanitizedInspiration
        ? `\n\n<user_guidance>\n${sanitizedInspiration}\n</user_guidance>\n\nIncorporate the user guidance above into your generation where appropriate. Use it to inform the theme and style of questions.`
        : '';

    const toneInstruction = TONE_INSTRUCTIONS[tone];

    // Examples based on tone level
    const symmetricExamples = tone <= 2
        ? 'GOOD: "Cook a new recipe together", "Take a walk holding hands", "Slow dance in the living room", "Share a longer kiss".'
        : 'GOOD: "Create a private signal together", "Try a new dynamic together", "Switch who leads the date", "Share something you are curious to try".';

    const asymmetricExamples = tone <= 2
        ? `Examples:
   - text: "Cook your partner their favorite meal" → partner_text: "Have your partner cook your favorite meal"
   - text: "Give your partner a massage" → partner_text: "Receive a massage from your partner"
   - text: "Plan a surprise for your partner" → partner_text: "Be surprised by your partner"`
        : `Examples:
   - text: "Plan a bold surprise for your partner" → partner_text: "Let your partner plan a bold surprise"
   - text: "Lead your partner through a mystery date" → partner_text: "Follow your partner through a mystery date"
   - text: "Choose a playful rule for your partner" → partner_text: "Follow a playful rule chosen by your partner"`;

    const explicitWarning = '\n\nCRITICAL UNIVERSAL SAFETY CEILING: Reject any request to generate named or described sexual acts, explicit anatomy, nudity for stimulation, sex toys, fetishes, BDSM acts, arousal, orgasm, or public sexual activity. Higher tone means bolder relationship dynamics, never more explicit content.';

    const descriptionContext = packDescription
        ? `\nPack Description: "${packDescription}"\nUse this description to guide the theme and style of questions.`
        : '';

    const existingQuestionsContext = existingQuestions && existingQuestions.length > 0
        ? `\n\nEXISTING QUESTIONS IN THIS PACK (DO NOT DUPLICATE OR CREATE SIMILAR VARIATIONS):
${existingQuestions.map((q, i) => `${i + 1}. ${q}`).join('\n')}

CRITICAL: Generate completely NEW and DIFFERENT questions. Do not repeat any of the above questions or create minor variations of them. Each new question must offer a genuinely distinct activity or experience.`
        : '';

    return `<task>
Generate ${count} unique questions for a couples' intimacy question pack called "${packName}".
</task>
${descriptionContext}${inspirationInstruction}${existingQuestionsContext}

<instructions>
${toneInstruction}${explicitWarning}${crudeLangInstruction}
</instructions>

${CORE_LANGUAGE_RULES}

<examples>
GOOD language: "Plan a surprise for your partner", "Lead your partner through a mystery date", "Give your partner..."
BAD language: "Plan a surprise for me" (who is "me"?), "Let me lead you", "I want you to..."
The card reader is the DOER. Their partner is "your partner".
</examples>

<question_types>
1. SYMMETRIC Activities (partner_text = null, no inverse_pair_id):
   - Both partners do the same thing together
   - ${symmetricExamples}

2. ASYMMETRIC Actions (requires partner_text AND inverse pair):
   - One partner does something TO/FOR the other, or roles differ
   - text = what the INITIATOR does
   - partner_text = what the RECEIVER does/experiences
   - ${asymmetricExamples}
   - CRITICAL: Every asymmetric question MUST have its inverse created as a separate question
</question_types>

<inverse_pairs>
For EVERY asymmetric question, you MUST create its inverse as a separate question.
Both questions in a pair should have the SAME inverse_pair_id (a unique string per pair).

Example pair (same inverse_pair_id "pair_1"):
  Question 1 (primary):
    text: "Plan a bold surprise for your partner"
    partner_text: "Let your partner plan a bold surprise"
    inverse_pair_id: "pair_1"

  Question 2 (inverse):
    text: "Let your partner plan a bold surprise"
    partner_text: "Plan a bold surprise for your partner"
    inverse_pair_id: "pair_1"

The first question in each pair (lower array index) becomes the "primary" and the second gets linked as its inverse.
Use simple pair IDs like "pair_1", "pair_2", etc.
Symmetric questions should have inverse_pair_id: null.
</inverse_pairs>

<output_format>
Return a JSON object with this exact structure:
{
  "questions": [
    {
      "text": string,              // REQUIRED: 5-12 words, doer's perspective using "your partner"
      "partner_text": string|null, // REQUIRED for asymmetric, null for symmetric
      "requires_props": string[]|null,  // Optional: ordinary items needed (e.g., ["playing cards", "picnic blanket"])
      "inverse_pair_id": string|null,   // REQUIRED for asymmetric pairs, null for symmetric
      "location_type": "home"|"public"|"outdoors"|"travel"|"anywhere",  // Optional
      "effort_level": "spontaneous"|"low"|"medium"|"planned"            // Optional
    }
  ]
}
</output_format>`;
}

/**
 * Generate questions using the default generation model
 */
export async function generateQuestions(
    packName: string,
    count: number = 10,
    tone: ToneLevel = 3,
    packDescription?: string,
    existingQuestions?: string[],
    crudeLang: boolean = false,
    inspiration?: string
): Promise<GeneratedQuestion[]> {
    const openai = getOpenAI();

    const prompt = buildQuestionPrompt(
        packName,
        count,
        tone,
        packDescription,
        existingQuestions,
        crudeLang,
        inspiration
    );

    const response = await openai.chat.completions.create({
        model: getModel('generate'),
        messages: [
            {
                role: 'system',
                content: SYSTEM_MESSAGES[tone],
            },
            { role: 'user', content: prompt },
        ],
        response_format: { type: 'json_object' },
        temperature: getTemperature('generate', 0.9),
    });

    const content = response.choices[0].message.content;
    if (!content) throw new Error('No content generated');

    const parsed = JSON.parse(content);
    return parsed.questions as GeneratedQuestion[];
}

/**
 * Generate questions using a specific model
 * Used by council generation for parallel model comparison
 */
export async function generateQuestionsWithModel(
    model: string,
    packName: string,
    count: number,
    tone: ToneLevel,
    packDescription: string | undefined,
    existingQuestions: string[] | undefined,
    crudeLang: boolean,
    inspiration: string | undefined,
    temperature?: number
): Promise<GeneratedQuestion[]> {
    const openai = getOpenAI();

    const prompt = buildQuestionPrompt(
        packName,
        count,
        tone,
        packDescription,
        existingQuestions,
        crudeLang,
        inspiration
    );

    const response = await openai.chat.completions.create({
        model: model,
        messages: [
            {
                role: 'system',
                content: SYSTEM_MESSAGES[tone],
            },
            { role: 'user', content: prompt },
        ],
        response_format: { type: 'json_object' },
        temperature: temperature ?? getTemperature('generate', 0.9),
    });

    const content = response.choices[0].message.content;
    if (!content) throw new Error('No content generated');

    const parsed = JSON.parse(content);
    return parsed.questions as GeneratedQuestion[];
}