<?php
/**
 * Standalone mode — value without an account.
 *
 * Why this exists: until now every module read its content from the server bundle
 * (AEO_Connect_Cache::get_bundle()). With no pairing there is no bundle, so a fresh
 * install did strictly nothing — the user activated the plugin, saw no effect, and
 * had no reason to go through the 6-step pairing flow. Measured on 2026-08-11:
 * 463 downloads, ~8/day sustained, and zero third-party site ever connected.
 *
 * This class produces the two artefacts that need no server at all:
 *   - AI-bot directives appended to robots.txt (purely deterministic);
 *   - a llms.txt built from what WordPress already knows about the site.
 *
 * Both are deliberately conservative: they only ever ADD, they never shadow a
 * physical file, and the server bundle always wins once the site is paired.
 *
 * @package AEO_Tool_Connect
 * @since   1.4.0
 */

defined( 'ABSPATH' ) || exit;

/**
 * Local generation of robots.txt directives and llms.txt, with no account.
 */
class AEO_Connect_Standalone {

	/**
	 * Option storing whether standalone mode is active. Default on: a plugin that
	 * does nothing on activation is the problem this class solves.
	 */
	const OPTION = 'aeo_connect_standalone';

	/**
	 * Is standalone mode usable right now?
	 *
	 * It applies only while the site is NOT paired. As soon as a bundle exists the
	 * server is the single source of truth, and this class steps aside — otherwise
	 * a paired site could show directives its dashboard never approved.
	 *
	 * @return bool
	 */
	public static function is_active() {
		if ( ! get_option( 'aeo_connect_enabled', true ) ) {
			return false;
		}

		if ( ! get_option( self::OPTION, true ) ) {
			return false;
		}

		return null === AEO_Connect_Cache::get_bundle();
	}

	/**
	 * AI crawler directives, appended to robots.txt.
	 *
	 * Allow-listed by default: the point is to be *visible* to answer engines, not
	 * to block them. A site owner who wants to opt out flips the module off.
	 *
	 * @return string
	 */
	public static function robots_directives() {
		$bots = array(
			'GPTBot',            // OpenAI — ChatGPT training/browsing.
			'OAI-SearchBot',     // OpenAI — search index.
			'ChatGPT-User',      // OpenAI — user-initiated fetches.
			'ClaudeBot',         // Anthropic.
			'anthropic-ai',      // Anthropic (legacy token, still honoured).
			'PerplexityBot',     // Perplexity.
			'Google-Extended',   // Google — Gemini / AI Overviews grounding.
			'Applebot-Extended', // Apple Intelligence.
			'CCBot',             // Common Crawl — feeds many models.
			'Bingbot',           // Microsoft — Copilot freshness.
		);

		$lines = array();

		foreach ( $bots as $bot ) {
			$lines[] = 'User-agent: ' . $bot;
			$lines[] = 'Allow: /';
			$lines[] = '';
		}

		$lines[] = 'Sitemap: ' . esc_url_raw( home_url( '/wp-sitemap.xml' ) );

		return implode( "\n", $lines );
	}

	/**
	 * Build a llms.txt from the site's own data.
	 *
	 * Title, tagline, the site's pages and latest posts. Descriptions are pulled
	 * from the SEO plugin's meta description when one is set (Yoast, Rank Math,
	 * SEOPress), then from the excerpt, then from the content itself — so entries
	 * carry a real description instead of a bare link. Content marked noindex by
	 * the SEO plugin is left out: if it is hidden from search engines it should
	 * not be advertised to answer engines either. No claim is made that this
	 * replaces a full audit — the admin screen says as much and points to
	 * everlange.com for the rest.
	 *
	 * @return string Empty string when the site has nothing worth listing.
	 */
	public static function llms_txt() {
		$name    = wp_strip_all_tags( (string) get_bloginfo( 'name' ) );
		$tagline = wp_strip_all_tags( (string) get_bloginfo( 'description' ) );

		if ( '' === trim( $name ) ) {
			return '';
		}

		$out   = array();
		$out[] = '# ' . $name;
		$out[] = '';

		if ( '' !== trim( $tagline ) ) {
			$out[] = '> ' . $tagline;
			$out[] = '';
		}

		$out[] = 'Site: ' . esc_url_raw( home_url( '/' ) );
		$out[] = '';

		$pages = get_pages(
			array(
				'sort_column' => 'menu_order,post_title',
				'number'      => 25,
				'post_status' => 'publish',
			)
		);

		$out = array_merge( $out, self::section( __( 'Pages', 'aeo-tool-connect' ), is_array( $pages ) ? $pages : array() ) );

		$posts = get_posts(
			array(
				'numberposts' => 25,
				'post_status' => 'publish',
			)
		);

		$out = array_merge( $out, self::section( __( 'Posts', 'aeo-tool-connect' ), is_array( $posts ) ? $posts : array() ) );

		$out[] = '---';
		$out[] = 'Generated locally by Everlange Connect (standalone mode).';
		$out[] = 'A full AEO audit across 10 answer engines is available at https://everlange.com/';

		return implode( "\n", $out ) . "\n";
	}

	/**
	 * Render one llms.txt section ("## Pages", "## Posts") as Markdown lines.
	 *
	 * @param string    $heading Section heading, already translated.
	 * @param WP_Post[] $items   Published posts or pages.
	 * @return string[] Lines including the trailing blank line, or an empty
	 *                  array when nothing in $items is listable.
	 */
	private static function section( $heading, $items ) {
		$lines = array();

		foreach ( $items as $item ) {
			if ( ! $item instanceof WP_Post ) {
				continue;
			}

			$title = wp_strip_all_tags( (string) $item->post_title );

			if ( '' === trim( $title ) ) {
				continue;
			}

			if ( self::is_noindexed( $item->ID ) ) {
				continue;
			}

			$line        = '- [' . $title . '](' . esc_url_raw( get_permalink( $item ) ) . ')';
			$description = self::description( $item );

			if ( '' !== $description ) {
				$line .= ' : ' . $description;
			}

			$lines[] = $line;
		}

		if ( empty( $lines ) ) {
			return array();
		}

		array_unshift( $lines, '## ' . $heading );
		$lines[] = '';

		return $lines;
	}

	/**
	 * Best available description for a post: the SEO plugin's meta description
	 * first (that is the one the owner actually wrote for machines), then the
	 * excerpt, then the content itself, trimmed.
	 *
	 * AIOSEO is not read here: it stores its descriptions in its own database
	 * table, not in post meta, and querying another plugin's table directly is
	 * not worth the coupling. Sites running AIOSEO fall back to the excerpt.
	 *
	 * @param WP_Post $item The post or page.
	 * @return string Possibly empty.
	 */
	private static function description( $item ) {
		$meta_keys = array(
			'_yoast_wpseo_metadesc', // Yoast SEO.
			'rank_math_description', // Rank Math.
			'_seopress_titles_desc', // SEOPress.
		);

		foreach ( $meta_keys as $meta_key ) {
			$value = get_post_meta( $item->ID, $meta_key, true );

			if ( is_string( $value ) && '' !== trim( $value ) ) {
				return wp_trim_words( wp_strip_all_tags( $value ), 30, '…' );
			}
		}

		$excerpt = wp_strip_all_tags( (string) $item->post_excerpt );

		if ( '' !== trim( $excerpt ) ) {
			return wp_trim_words( $excerpt, 30, '…' );
		}

		$content = wp_strip_all_tags( strip_shortcodes( (string) $item->post_content ) );

		if ( '' !== trim( $content ) ) {
			return wp_trim_words( $content, 30, '…' );
		}

		return '';
	}

	/**
	 * Has the owner's SEO plugin marked this content noindex?
	 *
	 * Checked per plugin because each stores the flag differently: Yoast uses a
	 * '1' string, Rank Math an array of robots tokens, SEOPress a 'yes' string.
	 * Absence of any flag means indexable, which is WordPress's own default.
	 *
	 * @param int $post_id Post ID.
	 * @return bool
	 */
	private static function is_noindexed( $post_id ) {
		if ( '1' === (string) get_post_meta( $post_id, '_yoast_wpseo_meta-robots-noindex', true ) ) {
			return true;
		}

		$rank_math = get_post_meta( $post_id, 'rank_math_robots', true );

		if ( is_array( $rank_math ) && in_array( 'noindex', $rank_math, true ) ) {
			return true;
		}

		if ( 'yes' === (string) get_post_meta( $post_id, '_seopress_robots_index', true ) ) {
			return true;
		}

		return false;
	}
}