ai_synth/backend/src/services/prompts.rs

//! Prompt construction for the LLM generation pipeline.
//!
//! Builds system and user prompts for:
//! - **Search pass** (Pass 1): web search and initial article discovery
//! - **Per-article classify**: per-article classification and summarization
//!
//! Prompts are provider-agnostic and parameterized by user settings.

use crate::models::settings::UserSettings;
use crate::models::source::Source;

/// Build the system prompt and user prompt for the search pass (Pass 1).
///
/// The search pass instructs the LLM to find recent news articles
/// matching the user's theme and categories, using web search grounding.
///
/// # Arguments
/// * `settings` — User's configured settings (theme, categories, etc.)
/// * `sources` — User's custom sources to prioritize
/// * `current_date` — Formatted date string for the prompt
/// * `recent_domains` — Domains used in recent syntheses to avoid if possible
pub fn build_search_prompt(
    settings: &UserSettings,
    sources: &[Source],
    current_date: &str,
    recent_domains: &[String],
    category_gaps: Option<&[(String, i32)]>,
) -> (String, String) {
    let sources_text = if sources.is_empty() {
        String::new()
    } else {
        let list = sources
            .iter()
            .map(|s| format!("- {} ({})", s.title, s.url))
            .collect::<Vec<_>>()
            .join("\n");
        format!(
            "\nEn plus des sources par defaut, tu DOIS imperativement consulter \
             et integrer les informations provenant de ces sources personnalisees :\n{}\n",
            list
        )
    };

    let categories_text = settings
        .categories
        .iter()
        .enumerate()
        .map(|(i, cat)| format!("{}. {}", i + 1, cat))
        .collect::<Vec<_>>()
        .join("\n");

    let behavior = if settings.search_agent_behavior.is_empty() {
        "Tu peux egalement utiliser d'autres sources pertinentes trouvees via la recherche Google."
            .to_string()
    } else {
        settings.search_agent_behavior.clone()
    };

    let system_prompt = format!(
        "Tu es un assistant IA precis. Tu dois TOUJOURS fournir des URLs completes et exactes. \
         Ne tronque jamais les URLs. Tu dois te concentrer UNIQUEMENT sur les actualites des {} \
         derniers jours.",
        settings.max_age_days
    );

    let user_prompt = format!(
        "Aujourd'hui, nous sommes le {date}.\n\
         Tu es un expert en analyse de l'actualite sur le theme : \"{theme}\".\n\
         Ta tache est de rechercher les actualites STRICTEMENT des {days} derniers jours.\n\
         Ne retourne AUCUNE actualite datant de plus de {days} jours.\n\n\
         Tu DOIS imperativement t'appuyer sur le contenu des sites web pertinents pour ce theme.\
         {sources}\
         {behavior}\n\n\
         La synthese doit etre divisee en {count} grandes sections :\n\
         {categories}\n\n\
         Pour chaque categorie, fournis exactement {max_items} actualites.\n\
         Pour chaque actualite, fournis un titre provisoire, l'URL source exacte et complete, \
         et un resume provisoire.\n\
         Ne retourne JAMAIS des URLs de pages d'accueil (homepage). Fournis toujours des liens \
         directs vers des articles specifiques avec un chemin complet (pas juste le nom de domaine).\n\
         Ne change jamais les URLs retournees, et ne les tronque jamais. \
         Retourne le resultat au format JSON en utilisant les cles category_0, category_1, etc. \
         correspondant a l'ordre des sections ci-dessus.",
        date = current_date,
        theme = settings.theme,
        days = settings.max_age_days,
        sources = sources_text,
        behavior = behavior,
        count = settings.categories.len(),
        categories = categories_text,
        max_items = settings.max_items_per_category,
    );

    let user_prompt = if recent_domains.is_empty() {
        user_prompt
    } else {
        let domains_list = recent_domains.join(", ");
        format!(
            "{}\n\nEvite si possible les sources deja utilisees dans les syntheses precedentes : {}.",
            user_prompt, domains_list
        )
    };

    // If we have specific category gaps (Phase 2), replace the generic "N per category" line
    let user_prompt = if let Some(gaps) = category_gaps {
        let gaps_text = gaps
            .iter()
            .map(|(cat, needed)| format!("- {} : {} articles", cat, needed))
            .collect::<Vec<_>>()
            .join("\n");
        user_prompt.replace(
            &format!("Pour chaque categorie, fournis exactement {} actualites.", settings.max_items_per_category),
            &format!("Fournis le nombre d'articles suivant par categorie :\n{}", gaps_text),
        )
    } else {
        user_prompt
    };

    (system_prompt, user_prompt)
}

/// Build a prompt for LLM-assisted link extraction from a source page.
pub fn build_link_extraction_prompt(body_html: &str) -> (String, String) {
    let system_prompt =
        "Tu es un assistant qui analyse des pages web. \
         Tu dois identifier les liens vers des articles d'actualite. \
         Reponds uniquement au format JSON demande."
            .to_string();

    let body_truncated: String = body_html.chars().take(12000).collect();

    let user_prompt = format!(
        "Voici le contenu HTML du body d'une page de blog ou de site d'actualites.\n\n\
         {body}\n\n\
         Extrais UNIQUEMENT les URLs qui pointent vers des articles \
         (pas les liens de navigation, tags, categories, login, pages statiques, topics, \
         archive, companies, events, company, event, collections, etc.).\n\
         Retourne les URLs completes, sans les modifier, dans le format JSON demande. \
         Ne change jamais les URLs retournees, et ne les tronque jamais.",
        body = body_truncated,
    );

    (system_prompt, user_prompt)
}

/// Build a prompt for per-article classification and summarization.
///
/// The LLM classifies the article into a category and generates a title + summary.
pub fn build_article_classify_prompt(
    title: &str,
    body_snippet: &str,
    categories: &[String],
) -> (String, String) {
    let system_prompt =
        "Tu es un assistant qui analyse des articles d'actualite. \
         Tu dois classer l'article dans une categorie et generer un titre et un resume. \
         Reponds uniquement au format JSON demande."
            .to_string();

    let categories_list = categories
        .iter()
        .map(|c| format!("- \"{}\"", c))
        .collect::<Vec<_>>()
        .join("\n");

    let user_prompt = format!(
        "Voici un article d'actualite.\n\n\
         Titre : {title}\n\n\
         Contenu (extrait) :\n{body}\n\n\
         Categories disponibles :\n{categories}\n\n\
         Classe cet article dans la categorie la plus appropriee.\n\
         Si aucune categorie ne correspond, utilise \"Autre\".\n\
         Genere un titre clair et un resume de 4 a 5 lignes.\n\
         Si le titre fourni est vide, genere un titre a partir du contenu.",
        title = if title.is_empty() { "(pas de titre)" } else { title },
        body = body_snippet,
        categories = categories_list,
    );

    (system_prompt, user_prompt)
}

#[cfg(test)]
mod tests {
    use super::*;
    use chrono::Utc;
    use uuid::Uuid;

    fn test_settings() -> UserSettings {
        UserSettings {
            user_id: Uuid::nil(),
            theme: "Intelligence Artificielle".to_string(),
            max_age_days: 7,
            categories: vec![
                "Annonces majeures".to_string(),
                "Recherche et innovation".to_string(),
            ],
            max_items_per_category: 4,
            max_articles_per_source: 3,
            use_llm_for_source_links: false,
            article_history_days: 90,
            batch_size: 5,
            search_agent_behavior: String::new(),
            ai_provider: String::new(),
            ai_model: String::new(),
            ai_model_websearch: String::new(),
            rate_limit_max_requests: None,
            rate_limit_time_window_seconds: None,
            updated_at: Utc::now(),
        }
    }

    #[test]
    fn search_prompt_includes_theme() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("Intelligence Artificielle"));
    }

    #[test]
    fn search_prompt_includes_date() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("lundi 21 mars 2026"));
    }

    #[test]
    fn search_prompt_includes_max_age() {
        let settings = test_settings();
        let (system, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("7 derniers jours"));
        assert!(system.contains("7"));
    }

    #[test]
    fn search_prompt_includes_categories() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("1. Annonces majeures"));
        assert!(user_prompt.contains("2. Recherche et innovation"));
        assert!(user_prompt.contains("2 grandes sections"));
    }

    #[test]
    fn search_prompt_includes_max_items() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("4 actualites"));
    }

    #[test]
    fn search_prompt_includes_custom_sources() {
        let settings = test_settings();
        let sources = vec![
            Source {
                id: Uuid::nil(),
                user_id: Uuid::nil(),
                title: "TechCrunch".into(),
                url: "https://techcrunch.com".into(),
                created_at: Utc::now(),
            },
            Source {
                id: Uuid::nil(),
                user_id: Uuid::nil(),
                title: "The Verge".into(),
                url: "https://theverge.com".into(),
                created_at: Utc::now(),
            },
        ];

        let (_, user_prompt) = build_search_prompt(&settings, &sources, "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("TechCrunch (https://techcrunch.com)"));
        assert!(user_prompt.contains("The Verge (https://theverge.com)"));
        assert!(user_prompt.contains("sources personnalisees"));
    }

    #[test]
    fn search_prompt_no_sources_no_section() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(!user_prompt.contains("sources personnalisees"));
    }

    #[test]
    fn search_prompt_custom_behavior() {
        let mut settings = test_settings();
        settings.search_agent_behavior =
            "Concentre-toi sur les sources europeennes.".to_string();

        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("Concentre-toi sur les sources europeennes."));
        assert!(!user_prompt.contains("recherche Google"));
    }

    #[test]
    fn search_prompt_default_behavior_when_empty() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("recherche Google"));
    }

    #[test]
    fn search_prompt_warns_against_homepage_urls() {
        let settings = test_settings();
        let (_, user_prompt) = build_search_prompt(&settings, &[], "lundi 21 mars 2026", &[], None);
        assert!(user_prompt.contains("pages d'accueil"));
        assert!(user_prompt.contains("articles specifiques"));
    }

    #[test]
    fn search_prompt_includes_recent_domains_avoidance() {
        let settings = test_settings();
        let sources = vec![];
        let date = "lundi 17 mars 2026";
        let domains = vec!["techcrunch.com".to_string(), "theverge.com".to_string()];
        let (_, user_prompt) = build_search_prompt(&settings, &sources, date, &domains, None);
        assert!(user_prompt.contains("Evite si possible"));
        assert!(user_prompt.contains("techcrunch.com"));
        assert!(user_prompt.contains("theverge.com"));
    }

    #[test]
    fn search_prompt_no_avoidance_when_domains_empty() {
        let settings = test_settings();
        let sources = vec![];
        let date = "lundi 17 mars 2026";
        let (_, user_prompt) = build_search_prompt(&settings, &sources, date, &[], None);
        assert!(!user_prompt.contains("Evite si possible"));
    }

    #[test]
    fn search_prompt_with_category_gaps() {
        let settings = test_settings();
        let sources = vec![];
        let date = "lundi 17 mars 2026";
        let gaps = vec![
            ("AI News".to_string(), 2),
            ("Cybersecurity".to_string(), 4),
        ];
        let (_, user_prompt) = build_search_prompt(&settings, &sources, date, &[], Some(&gaps));
        assert!(user_prompt.contains("AI News : 2 articles"));
        assert!(user_prompt.contains("Cybersecurity : 4 articles"));
        assert!(!user_prompt.contains("exactement"));
    }

    #[test]
    fn search_prompt_without_gaps_uses_default() {
        let settings = test_settings();
        let sources = vec![];
        let date = "lundi 17 mars 2026";
        let (_, user_prompt) = build_search_prompt(&settings, &sources, date, &[], None);
        assert!(user_prompt.contains("exactement"));
    }

    #[test]
    fn link_extraction_prompt_includes_body() {
        let (sys, user) = build_link_extraction_prompt("<a href='/post'>P</a>");
        assert!(user.contains("<a href='/post'>"));
        assert!(user.contains("articles"));
        assert!(sys.contains("liens"));
        assert!(!user.contains("<head>"));
    }

    #[test]
    fn link_extraction_prompt_truncates_body() {
        let long_body = "x".repeat(20000);
        let (_, user) = build_link_extraction_prompt(&long_body);
        assert!(user.len() < 18000); // 12000 chars of body + prompt text
    }

    #[test]
    fn article_classify_prompt_includes_content() {
        let (sys, user) = build_article_classify_prompt(
            "GPT-5 Released",
            "OpenAI released GPT-5 today",
            &["AI News".into(), "Autre".into()],
        );
        assert!(user.contains("GPT-5 Released"));
        assert!(user.contains("AI News"));
        assert!(user.contains("Autre"));
        assert!(sys.contains("classer"));
    }

    #[test]
    fn article_classify_prompt_handles_empty_title() {
        let (_, user) = build_article_classify_prompt("", "Some content", &["Tech".into(), "Autre".into()]);
        assert!(user.contains("(pas de titre)"));
    }

}