Add LLM-friendly documentation export feature

Implements automatic generation of LLM-friendly text files during Jekyll build:
- /llms/guides.txt: All 96 developer guides
- /llms/api-docs.txt: C, Android, iOS, JavaScript, and Rocky.js API documentation
- /llms/community.txt: Community tools, apps, and libraries

Features:
- Clean markdown format with minimal metadata (title + URL)
- Automatic Liquid tag/variable removal
- Filters out redirect pages and stub content
- Organized with "---" separators between pages
- Only generates files with actual content

Architecture:
- lib/llms_txt_builder.rb: Reusable builder module (DRY principle)
- plugins/generator_llms_txt.rb: Jekyll generator plugin (runs at :lowest priority)

Signed-off-by: Ilia Breitburg <[email protected]>

Author: breitburg

lib/llms_txt_builder.rb

@@ -0,0 +1,130 @@
+require 'fileutils'
+
+module LlmsTxt
+  # Builder module for generating LLM-friendly text files from Jekyll content
+  class Builder
+    SEPARATOR = "---"
+
+    def initialize(site)
+      @site = site
+      @base_url = site.config['url'] || ''
+    end
+
+    # Process a Jekyll collection and return formatted content
+    def process_collection(collection_name)
+      collection = @site.collections[collection_name]
+      return '' unless collection
+
+      content = []
+      collection.docs.each do |doc|
+        next if should_exclude?(doc)
+        content << format_page(doc)
+      end
+      content.join("\n\n#{SEPARATOR}\n\n")
+    end
+
+    # Process multiple collections and return formatted content
+    def process_collections(collection_names)
+      content = []
+      collection_names.each do |name|
+        @site.collections[name]&.docs&.each do |doc|
+          next if should_exclude?(doc)
+          content << format_page(doc)
+        end
+      end
+      content.join("\n\n#{SEPARATOR}\n\n")
+    end
+
+    # Process generated pages matching a URL pattern
+    def process_pages(url_pattern)
+      content = []
+      @site.pages.each do |page|
+        next unless page.url.match?(url_pattern)
+        next if should_exclude?(page)
+        content << format_page(page)
+      end
+      content.join("\n\n#{SEPARATOR}\n\n")
+    end
+
+    # Format a single page/document
+    def format_page(page)
+      title = extract_title(page)
+      url = build_url(page.url)
+      markdown_content = extract_content(page)
+
+      output = []
+      output << "# #{title}"
+      output << "URL: #{url}"
+      output << ""
+      output << markdown_content
+      output.join("\n")
+    end
+
+    # Write content to a file in the llms directory
+    def write_file(filename, content, section_name = nil)
+      # Create temp directory for llms files
+      tmp_dir = File.join(@site.source, '../tmp/')
+      llms_dir = File.join(tmp_dir, 'llms')
+      FileUtils.mkdir_p(llms_dir)
+
+      filepath = File.join(llms_dir, filename)
+
+      output = []
+      if section_name
+        output << "# #{section_name}"
+        output << ""
+        output << "This file contains all #{section_name.downcase} from the Pebble Developer Documentation."
+        output << ""
+        output << SEPARATOR
+        output << ""
+      end
+      output << content
+
+      File.write(filepath, output.join("\n"))
+
+      # Register the file with Jekyll as a static file so it gets copied to output
+      @site.static_files << Jekyll::StaticFile.new(@site, tmp_dir, 'llms', filename)
+
+      Jekyll.logger.info('LLMS.txt:', "Generated #{filename}")
+    end
+
+    private
+
+    # Check if a page should be excluded from LLM files
+    def should_exclude?(page)
+      return true if page.data['llms_exclude'] == true
+
+      # Exclude redirect pages (they have no useful content)
+      return true if page.data['layout'] == 'redirect'
+
+      # Exclude pages with very little content (likely stubs or redirects)
+      content = page.content || ''
+      return true if content.strip.length < 100
+
+      false
+    end
+
+    # Extract the title from a page
+    def extract_title(page)
+      page.data['title'] || page.data['name'] || File.basename(page.url, '.*')
+    end
+
+    # Build the full URL for a page
+    def build_url(path)
+      "#{@base_url}#{path}"
+    end
+
+    # Extract and clean content from a page
+    def extract_content(page)
+      content = page.content || ''
+
+      # Remove liquid tags and variables (including multiline)
+      content = content.gsub(/\{%.*?%\}/m, '')
+      content = content.gsub(/\{\{.*?\}\}/m, '')
+
+      # Clean up extra whitespace
+      content = content.gsub(/\n{3,}/, "\n\n")
+      content.strip
+    end
+  end
+end

plugins/generator_llms_txt.rb

@@ -0,0 +1,79 @@
+require_relative '../lib/llms_txt_builder.rb'
+
+module Jekyll
+  # Generator for creating LLM-friendly text files from documentation
+  class GeneratorLlmsTxt < Generator
+    # Run last so all other generators have completed
+    priority :lowest
+
+    def generate(site)
+      @site = site
+      @builder = LlmsTxt::Builder.new(site)
+
+      Jekyll.logger.info('LLMS.txt:', 'Generating LLM documentation files...')
+
+      begin
+        generate_guides
+        generate_api_docs
+        generate_community
+
+        Jekyll.logger.info('LLMS.txt:', 'Done.')
+      rescue StandardError => e
+        Jekyll.logger.error('LLMS.txt Error:', e.message)
+        Jekyll.logger.error('LLMS.txt Error:', e.backtrace.join("\n"))
+      end
+    end
+
+    private
+
+    # Generate guides.txt from the guides collection
+    def generate_guides
+      content = @builder.process_collection('guides')
+      if content.empty?
+        Jekyll.logger.warn('LLMS.txt:', 'No guides found to process')
+        return
+      end
+      @builder.write_file('guides.txt', content, 'Guides')
+    end
+
+    # Generate api-docs.txt from generated API documentation pages
+    def generate_api_docs
+      # Match pages under /docs/ that are actual API documentation
+      # This includes C, Android, iOS, JavaScript, and Rocky.js docs
+      api_patterns = [
+        %r{^/docs/c/},
+        %r{^/docs/pebblekit-android/},
+        %r{^/docs/pebblekit-ios/},
+        %r{^/docs/pebblekit-js/},
+        %r{^/docs/rockyjs/}
+      ]
+
+      content = []
+      api_patterns.each do |pattern|
+        pattern_content = @builder.process_pages(pattern)
+        content << pattern_content unless pattern_content.empty?
+      end
+
+      if content.empty?
+        Jekyll.logger.warn('LLMS.txt:', 'No API docs found to process (DOCS_URL not set or no valid content)')
+        return
+      end
+
+      separator = "\n\n#{LlmsTxt::Builder::SEPARATOR}\n\n"
+      @builder.write_file('api-docs.txt', content.join(separator), 'API Documentation')
+    end
+
+    # Generate community.txt from community collections
+    def generate_community
+      community_collections = ['community_tools', 'community_apps', 'community_libraries']
+      content = @builder.process_collections(community_collections)
+
+      if content.empty?
+        Jekyll.logger.warn('LLMS.txt:', 'No community resources found to process')
+        return
+      end
+
+      @builder.write_file('community.txt', content, 'Community Resources')
+    end
+  end
+end