swarm_memory 2.1.2 → 2.1.3

data/lib/swarm_sdk/swarm_loader.rb

@@ -0,0 +1,145 @@
+# frozen_string_literal: true
+
+module SwarmSDK
+  # Loader for creating swarm instances from multiple sources
+  #
+  # SwarmLoader loads swarm configurations from:
+  # - Files: .rb (DSL) or .yml (YAML)
+  # - YAML strings: Direct YAML content
+  # - DSL blocks: Inline Ruby blocks
+  #
+  # All loaded swarms get hierarchical swarm_id and parent_swarm_id.
+  #
+  # ## Features
+  # - Supports Ruby DSL (.rb files or blocks)
+  # - Supports YAML (.yml/.yaml files or strings)
+  # - Sets hierarchical swarm_id based on parent + registration name
+  # - Isolates loading in separate context
+  # - Proper error handling for missing/invalid sources
+  #
+  # ## Examples
+  #
+  #   # From file
+  #   swarm = SwarmLoader.load_from_file(
+  #     "./swarms/code_review.rb",
+  #     swarm_id: "main/code_review",
+  #     parent_swarm_id: "main"
+  #   )
+  #
+  #   # From YAML string
+  #   swarm = SwarmLoader.load_from_yaml_string(
+  #     "version: 2\nswarm:\n  name: Test\n...",
+  #     swarm_id: "main/testing",
+  #     parent_swarm_id: "main"
+  #   )
+  #
+  #   # From block
+  #   swarm = SwarmLoader.load_from_block(
+  #     proc { id "team"; name "Team"; agent :dev { ... } },
+  #     swarm_id: "main/team",
+  #     parent_swarm_id: "main"
+  #   )
+  #
+  class SwarmLoader
+    class << self
+      # Load a swarm from a file (.rb or .yml)
+      #
+      # @param file_path [String] Path to swarm file
+      # @param swarm_id [String] Hierarchical swarm ID to assign
+      # @param parent_swarm_id [String] Parent swarm ID
+      # @return [Swarm] Loaded swarm instance with overridden IDs
+      # @raise [ConfigurationError] If file not found or unsupported type
+      def load_from_file(file_path, swarm_id:, parent_swarm_id:)
+        path = Pathname.new(file_path).expand_path
+
+        raise ConfigurationError, "Swarm file not found: #{path}" unless path.exist?
+
+        # Determine file type and load
+        case path.extname
+        when ".rb"
+          load_from_ruby_file(path, swarm_id, parent_swarm_id)
+        when ".yml", ".yaml"
+          load_from_yaml_file(path, swarm_id, parent_swarm_id)
+        else
+          raise ConfigurationError, "Unsupported swarm file type: #{path.extname}. Use .rb, .yml, or .yaml"
+        end
+      end
+
+      # Load a swarm from YAML string
+      #
+      # @param yaml_content [String] YAML configuration content
+      # @param swarm_id [String] Hierarchical swarm ID to assign
+      # @param parent_swarm_id [String] Parent swarm ID
+      # @return [Swarm] Loaded swarm instance with overridden IDs
+      # @raise [ConfigurationError] If YAML is invalid
+      def load_from_yaml_string(yaml_content, swarm_id:, parent_swarm_id:)
+        # Use Configuration to parse YAML string
+        config = Configuration.new(yaml_content, base_dir: Dir.pwd)
+        config.load_and_validate
+        swarm = config.to_swarm
+
+        # Override swarm_id and parent_swarm_id
+        swarm.override_swarm_ids(swarm_id: swarm_id, parent_swarm_id: parent_swarm_id)
+
+        swarm
+      end
+
+      # Load a swarm from DSL block
+      #
+      # @param block [Proc] Block containing SwarmSDK DSL
+      # @param swarm_id [String] Hierarchical swarm ID to assign
+      # @param parent_swarm_id [String] Parent swarm ID
+      # @return [Swarm] Loaded swarm instance with overridden IDs
+      def load_from_block(block, swarm_id:, parent_swarm_id:)
+        # Execute block in Builder context
+        builder = Swarm::Builder.new
+        builder.instance_eval(&block)
+        swarm = builder.build_swarm
+
+        # Override swarm_id and parent_swarm_id
+        swarm.override_swarm_ids(swarm_id: swarm_id, parent_swarm_id: parent_swarm_id)
+
+        swarm
+      end
+
+      private
+
+      # Load swarm from Ruby DSL file
+      #
+      # @param path [Pathname] Path to .rb file
+      # @param swarm_id [String] Swarm ID to assign
+      # @param parent_swarm_id [String] Parent swarm ID
+      # @return [Swarm] Loaded swarm with overridden IDs
+      def load_from_ruby_file(path, swarm_id, parent_swarm_id)
+        content = File.read(path)
+
+        # Execute DSL in isolated context
+        # The DSL should return a swarm via SwarmSDK.build { ... }
+        swarm = eval(content, binding, path.to_s) # rubocop:disable Security/Eval
+
+        # Override swarm_id and parent_swarm_id
+        # These must be set after build to ensure hierarchical structure
+        swarm.override_swarm_ids(swarm_id: swarm_id, parent_swarm_id: parent_swarm_id)
+
+        swarm
+      end
+
+      # Load swarm from YAML file
+      #
+      # @param path [Pathname] Path to .yml file
+      # @param swarm_id [String] Swarm ID to assign
+      # @param parent_swarm_id [String] Parent swarm ID
+      # @return [Swarm] Loaded swarm with overridden IDs
+      def load_from_yaml_file(path, swarm_id, parent_swarm_id)
+        # Use Configuration to load and convert YAML to swarm
+        config = Configuration.load_file(path.to_s)
+        swarm = config.to_swarm
+
+        # Override swarm_id and parent_swarm_id
+        swarm.override_swarm_ids(swarm_id: swarm_id, parent_swarm_id: parent_swarm_id)
+
+        swarm
+      end
+    end
+  end
+end

data/lib/swarm_sdk/swarm_registry.rb

@@ -0,0 +1,136 @@
+# frozen_string_literal: true
+
+module SwarmSDK
+  # Registry for managing sub-swarms in composable swarms
+  #
+  # SwarmRegistry handles lazy loading, caching, and lifecycle management
+  # of child swarms registered via the `swarms` DSL block.
+  #
+  # ## Features
+  # - Lazy loading: Sub-swarms are only loaded when first accessed
+  # - Caching: Loaded swarms are cached for reuse
+  # - Hierarchical IDs: Sub-swarms get IDs based on parent + registration name
+  # - Context control: keep_context determines if swarm state persists
+  # - Lifecycle management: Cleanup cascades through all sub-swarms
+  #
+  # ## Example
+  #
+  #   registry = SwarmRegistry.new(parent_swarm_id: "main_app")
+  #   registry.register("code_review", file: "./swarms/code_review.rb", keep_context: true)
+  #
+  #   # Lazy load on first access
+  #   swarm = registry.load_swarm("code_review")
+  #   # => Swarm with swarm_id = "main_app/code_review"
+  #
+  #   # Reset if keep_context: false
+  #   registry.reset_if_needed("code_review")
+  #
+  class SwarmRegistry
+    # Initialize a new swarm registry
+    #
+    # @param parent_swarm_id [String] ID of the parent swarm
+    def initialize(parent_swarm_id:)
+      @parent_swarm_id = parent_swarm_id
+      @registered_swarms = {}
+      # Format: { "code_review" => { file: "...", keep_context: true, instance: nil } }
+    end
+
+    # Register a sub-swarm for lazy loading
+    #
+    # @param name [String] Registration name for the swarm
+    # @param source [Hash] Source specification with :type and :value
+    #   - { type: :file, value: "./path/to/swarm.rb" }
+    #   - { type: :yaml, value: "version: 2\n..." }
+    #   - { type: :block, value: Proc }
+    # @param keep_context [Boolean] Whether to preserve conversation state between calls (default: true)
+    # @return [void]
+    # @raise [ArgumentError] If swarm with same name already registered
+    def register(name, source:, keep_context: true)
+      raise ArgumentError, "Swarm '#{name}' already registered" if @registered_swarms.key?(name)
+
+      @registered_swarms[name] = {
+        source: source,
+        keep_context: keep_context,
+        instance: nil, # Lazy load
+      }
+    end
+
+    # Check if a swarm is registered
+    #
+    # @param name [String] Swarm registration name
+    # @return [Boolean] True if swarm is registered
+    def registered?(name)
+      @registered_swarms.key?(name)
+    end
+
+    # Load a registered swarm (lazy load + cache)
+    #
+    # Loads the swarm from its source (file, yaml, or block) on first access, then caches it.
+    # Sets hierarchical swarm_id based on parent_swarm_id + registration name.
+    #
+    # @param name [String] Swarm registration name
+    # @return [Swarm] Loaded swarm instance
+    # @raise [ConfigurationError] If swarm not registered
+    def load_swarm(name)
+      entry = @registered_swarms[name]
+      raise ConfigurationError, "Swarm '#{name}' not registered" unless entry
+
+      # Return cached instance if exists
+      return entry[:instance] if entry[:instance]
+
+      # Load from appropriate source
+      swarm_id = "#{@parent_swarm_id}/#{name}" # Hierarchical
+      source = entry[:source]
+
+      swarm = case source[:type]
+      when :file
+        SwarmLoader.load_from_file(
+          source[:value],
+          swarm_id: swarm_id,
+          parent_swarm_id: @parent_swarm_id,
+        )
+      when :yaml
+        SwarmLoader.load_from_yaml_string(
+          source[:value],
+          swarm_id: swarm_id,
+          parent_swarm_id: @parent_swarm_id,
+        )
+      when :block
+        SwarmLoader.load_from_block(
+          source[:value],
+          swarm_id: swarm_id,
+          parent_swarm_id: @parent_swarm_id,
+        )
+      else
+        raise ConfigurationError, "Unknown source type: #{source[:type]}"
+      end
+
+      entry[:instance] = swarm
+      swarm
+    end
+
+    # Reset swarm context if keep_context: false
+    #
+    # @param name [String] Swarm registration name
+    # @return [void]
+    def reset_if_needed(name)
+      entry = @registered_swarms[name]
+      return if entry[:keep_context]
+
+      entry[:instance]&.reset_context!
+    end
+
+    # Cleanup all registered swarms
+    #
+    # Stops all loaded swarm instances and clears the registry.
+    # Should be called when parent swarm is done.
+    #
+    # @return [void]
+    def shutdown_all
+      @registered_swarms.each_value do |entry|
+        entry[:instance]&.cleanup
+      end
+      @registered_swarms.clear
+    end
+  end
+end

data/lib/swarm_sdk/tools/delegate.rb

@@ -14,10 +14,12 @@ module SwarmSDK
       #
       # @param delegate_name [String] Name of the delegate agent (e.g., "backend")
       # @param delegate_description [String] Description of the delegate agent
-      # @param delegate_chat [AgentChat] The chat instance for the delegate agent
+      # @param delegate_chat [AgentChat, nil] The chat instance for the delegate agent (nil if delegating to swarm)
       # @param agent_name [Symbol, String] Name of the agent using this tool
       # @param swarm [Swarm] The swarm instance
       # @param hook_registry [Hooks::Registry] Registry for callbacks
+      # @param call_stack [Array] Delegation call stack for circular dependency detection
+      # @param swarm_registry [SwarmRegistry, nil] Registry for sub-swarms (nil if not using composable swarms)
       # @param delegating_chat [Agent::Chat, nil] The chat instance of the agent doing the delegating (for accessing hooks)
       def initialize(
         delegate_name:,
@@ -26,6 +28,8 @@ module SwarmSDK
         agent_name:,
         swarm:,
         hook_registry:,
+        call_stack:,
+        swarm_registry: nil,
         delegating_chat: nil
       )
         super()
@@ -36,6 +40,8 @@ module SwarmSDK
         @agent_name = agent_name
         @swarm = swarm
         @hook_registry = hook_registry
+        @call_stack = call_stack
+        @swarm_registry = swarm_registry
         @delegating_chat = delegating_chat
 
         # Generate tool name in the expected format: DelegateTaskTo[AgentName]
@@ -63,6 +69,13 @@ module SwarmSDK
       # @param task [String] Task to delegate
       # @return [String] Result from delegate agent or error message
       def execute(task:)
+        # Check for circular dependency
+        if @call_stack.include?(@delegate_target)
+          emit_circular_warning
+          return "Error: Circular delegation detected: #{@call_stack.join(" -> ")} -> #{@delegate_target}. " \
+            "Please restructure your delegation to avoid infinite loops."
+        end
+
         # Get agent-specific hooks from the delegating chat instance
         agent_hooks = if @delegating_chat&.respond_to?(:hook_agent_hooks)
           @delegating_chat.hook_agent_hooks || {}
@@ -94,9 +107,16 @@ module SwarmSDK
           return result.value
         end
 
-        # Proceed with delegation
-        response = @delegate_chat.ask(task)
-        delegation_result = response.content
+        # Determine delegation type and proceed
+        delegation_result = if @delegate_chat
+          # Delegate to agent
+          delegate_to_agent(task)
+        elsif @swarm_registry&.registered?(@delegate_target)
+          # Delegate to registered swarm
+          delegate_to_swarm(task)
+        else
+          raise ConfigurationError, "Unknown delegation target: #{@delegate_target}"
+        end
 
         # Trigger post_delegation callback
         post_context = Hooks::Context.new(
@@ -127,10 +147,12 @@ module SwarmSDK
         LogStream.emit(
           type: "delegation_error",
           agent: @agent_name,
+          swarm_id: @swarm.swarm_id,
+          parent_swarm_id: @swarm.parent_swarm_id,
           delegate_to: @tool_name,
           error_class: e.class.name,
           error_message: "Request timed out",
-          backtrace: e.backtrace&.first(5) || [],
+          error_backtrace: e.backtrace&.first(5) || [],
         )
         "Error: Request to #{@tool_name} timed out. The agent may be overloaded or the LLM service is not responding. Please try again or simplify the task."
       rescue Faraday::Error => e
@@ -138,10 +160,12 @@ module SwarmSDK
         LogStream.emit(
           type: "delegation_error",
           agent: @agent_name,
+          swarm_id: @swarm.swarm_id,
+          parent_swarm_id: @swarm.parent_swarm_id,
           delegate_to: @tool_name,
           error_class: e.class.name,
           error_message: e.message,
-          backtrace: e.backtrace&.first(5) || [],
+          error_backtrace: e.backtrace&.first(5) || [],
         )
         "Error: Network error communicating with #{@tool_name}: #{e.class.name}. Please check connectivity and try again."
       rescue StandardError => e
@@ -150,15 +174,76 @@ module SwarmSDK
         LogStream.emit(
           type: "delegation_error",
           agent: @agent_name,
+          swarm_id: @swarm.swarm_id,
+          parent_swarm_id: @swarm.parent_swarm_id,
           delegate_to: @tool_name,
           error_class: e.class.name,
           error_message: e.message,
-          backtrace: backtrace_array,
+          error_backtrace: backtrace_array,
         )
         # Return error string for LLM
         backtrace_str = backtrace_array.join("\n  ")
         "Error: #{@tool_name} encountered an error: #{e.class.name}: #{e.message}\nBacktrace:\n  #{backtrace_str}"
       end
+
+      private
+
+      # Delegate to an agent
+      #
+      # @param task [String] Task to delegate
+      # @return [String] Result from agent
+      def delegate_to_agent(task)
+        # Push delegate target onto call stack to track delegation chain
+        @call_stack.push(@delegate_target)
+        begin
+          response = @delegate_chat.ask(task)
+          response.content
+        ensure
+          # Always pop from stack, even if delegation fails
+          @call_stack.pop
+        end
+      end
+
+      # Delegate to a registered swarm
+      #
+      # @param task [String] Task to delegate
+      # @return [String] Result from swarm's lead agent
+      def delegate_to_swarm(task)
+        # Load sub-swarm (lazy load + cache)
+        subswarm = @swarm_registry.load_swarm(@delegate_target)
+
+        # Push delegate target onto call stack to track delegation chain
+        @call_stack.push(@delegate_target)
+        begin
+          # Execute sub-swarm's lead agent
+          lead_agent = subswarm.agents[subswarm.lead_agent]
+          response = lead_agent.ask(task)
+          result = response.content
+
+          # Reset if keep_context: false
+          @swarm_registry.reset_if_needed(@delegate_target)
+
+          result
+        ensure
+          # Always pop from stack, even if delegation fails
+          @call_stack.pop
+        end
+      end
+
+      # Emit circular dependency warning event
+      #
+      # @return [void]
+      def emit_circular_warning
+        LogStream.emit(
+          type: "delegation_circular_dependency",
+          agent: @agent_name,
+          swarm_id: @swarm.swarm_id,
+          parent_swarm_id: @swarm.parent_swarm_id,
+          target: @delegate_target,
+          call_stack: @call_stack,
+          timestamp: Time.now.utc.iso8601,
+        )
+      end
     end
   end
 end

data/lib/swarm_sdk/tools/read.rb

@@ -92,25 +92,37 @@ module SwarmSDK
           return validation_error("Path is a directory, not a file. Use Bash with ls to read directories.")
         end
 
-        # Register this read in the tracker (use resolved path)
-        Stores::ReadTracker.register_read(@agent_name, resolved_path)
-
         # Check if it's a document and try to convert it
         converter = find_converter_for_file(resolved_path)
         if converter
           result = converter.new.convert(resolved_path)
+          # For document files, register the converted text content
+          # Extract text from result (may be wrapped in system-reminder tags)
+          if result.is_a?(String)
+            # Remove system-reminder wrapper if present to get clean text for digest
+            text_content = result.gsub(%r{<system-reminder>.*?</system-reminder>}m, "").strip
+            Stores::ReadTracker.register_read(@agent_name, resolved_path, text_content)
+          end
           return result
         end
 
         # Try to read as text, handle binary files separately
         content = read_file_content(resolved_path)
 
-        # If content is a Content object (binary file), return it directly
-        return content if content.is_a?(RubyLLM::Content)
+        # If content is a Content object (binary file), track with binary digest and return
+        if content.is_a?(RubyLLM::Content)
+          # For binary files, read raw bytes for digest
+          binary_content = File.binread(resolved_path)
+          Stores::ReadTracker.register_read(@agent_name, resolved_path, binary_content)
+          return content
+        end
 
         # Return early if we got an error message or system reminder
         return content if content.is_a?(String) && (content.start_with?("Error:") || content.start_with?("<system-reminder>"))
 
+        # At this point, we have valid text content - register the read with digest
+        Stores::ReadTracker.register_read(@agent_name, resolved_path, content)
+
         # Check if file is empty
         if content.empty?
           return format_with_reminder(

data/lib/swarm_sdk/tools/scratchpad/scratchpad_list.rb

@@ -12,8 +12,29 @@ module SwarmSDK
 
         description <<~DESC
           List all entries in scratchpad with their metadata.
-          Shows path, title, size, and last updated time for each entry.
-          Use this to discover what's stored in the scratchpad.
+
+          ## When to Use ScratchpadList
+
+          Use ScratchpadList to:
+          - Discover what content is available in the scratchpad
+          - Check what other agents have stored
+          - Find relevant entries before reading them
+          - Review all stored outputs and analysis
+          - Check entry sizes and last update times
+
+          ## Best Practices
+
+          - Use this before ScratchpadRead if you don't know what's stored
+          - Filter by prefix to narrow down results (e.g., 'notes/' lists all notes)
+          - Shows path, title, size, and last updated time for each entry
+          - Any agent can see all scratchpad entries
+          - Helps coordinate multi-agent workflows
+
+          ## Examples
+
+          - List all entries: (no prefix parameter)
+          - List notes only: prefix='notes/'
+          - List analysis results: prefix='analysis/'
         DESC
 
         param :prefix,

data/lib/swarm_sdk/tools/scratchpad/scratchpad_read.rb

@@ -12,8 +12,29 @@ module SwarmSDK
 
         description <<~DESC
           Read content from scratchpad.
-          Use this to retrieve temporary notes, results, or messages stored by any agent.
-          Any agent can read any scratchpad content.
+
+          ## When to Use ScratchpadRead
+
+          Use ScratchpadRead to:
+          - Retrieve previously stored content and outputs
+          - Access detailed analysis or results from earlier steps
+          - Read messages or notes left by other agents
+          - Access cached computed data
+          - Retrieve content that was too long for direct responses
+
+          ## Best Practices
+
+          - Any agent can read any scratchpad content
+          - Content is returned with line numbers for easy reference
+          - Use ScratchpadList first if you don't know what's stored
+          - Scratchpad data is temporary and lost when swarm ends
+          - For persistent data, use MemoryRead instead
+
+          ## Examples
+
+          - Read status: file_path='status'
+          - Read analysis: file_path='api_analysis'
+          - Read agent notes: file_path='notes/backend'
         DESC
 
         param :file_path,

data/lib/swarm_sdk/tools/scratchpad/scratchpad_write.rb

@@ -13,12 +13,29 @@ module SwarmSDK
 
         description <<~DESC
           Store content in scratchpad for temporary cross-agent communication.
-          Use this for quick notes, intermediate results, or coordination messages.
-          Any agent can read this content. Data is lost when the swarm ends.
 
-          For persistent storage that survives across sessions, use MemoryWrite instead.
+          ## When to Use Scratchpad
 
-          Choose a simple, descriptive path. Examples: 'status', 'result', 'notes/agent_x'
+          Use ScratchpadWrite to:
+          - Store detailed outputs, analysis, or results that are too long for direct responses
+          - Share information that would otherwise clutter your responses
+          - Store intermediate results during multi-step tasks
+          - Leave coordination messages for other agents
+          - Cache computed data for quick retrieval
+
+          ## Best Practices
+
+          - Choose simple, descriptive paths: 'status', 'result', 'notes/agent_x'
+          - Use hierarchical paths for organization: 'analysis/step1', 'analysis/step2'
+          - Keep entries focused - one piece of information per entry
+          - Any agent can read scratchpad content
+          - Data is lost when the swarm ends (use MemoryWrite for persistent storage)
+          - Maximum 1MB per entry
+
+          ## Examples
+
+          Good paths: 'status', 'api_analysis', 'test_results', 'notes/backend'
+          Bad paths: 'scratch/temp/file123.txt', 'output.log'
         DESC
 
         param :file_path,

data/lib/swarm_sdk/tools/stores/read_tracker.rb

@@ -3,39 +3,74 @@
 module SwarmSDK
   module Tools
     module Stores
-      # ReadTracker manages read-file tracking for all agents
+      # ReadTracker manages read-file tracking for all agents with content digest verification
       #
       # This module maintains a global registry of which files each agent has read
-      # during their conversation. This enables enforcement of the "read-before-write"
-      # and "read-before-edit" rules that ensure agents have context before modifying files.
+      # during their conversation along with SHA256 digests of the content. This enables
+      # enforcement of the "read-before-write" and "read-before-edit" rules that ensure
+      # agents have context before modifying files, AND prevents editing files that have
+      # changed externally since being read.
       #
-      # Each agent maintains an independent set of read files, keyed by agent identifier.
+      # Each agent maintains an independent map of read files to content digests.
       module ReadTracker
-        @read_files = {}
+        @read_files = {} # { agent_id => { file_path => sha256_digest } }
         @mutex = Mutex.new
 
         class << self
-          # Register that an agent has read a file
+          # Register that an agent has read a file with content digest
           #
           # @param agent_id [Symbol] The agent identifier
           # @param file_path [String] The absolute path to the file
-          def register_read(agent_id, file_path)
+          # @param content [String] File content (for digest calculation)
+          # @return [String] The calculated SHA256 digest
+          def register_read(agent_id, file_path, content)
             @mutex.synchronize do
-              @read_files[agent_id] ||= Set.new
-              @read_files[agent_id] << File.expand_path(file_path)
+              @read_files[agent_id] ||= {}
+              digest = Digest::SHA256.hexdigest(content)
+              @read_files[agent_id][File.expand_path(file_path)] = digest
+              digest
             end
           end
 
-          # Check if an agent has read a file
+          # Check if an agent has read a file AND content hasn't changed
           #
           # @param agent_id [Symbol] The agent identifier
           # @param file_path [String] The absolute path to the file
-          # @return [Boolean] true if the agent has read this file
+          # @return [Boolean] true if agent read file and content matches
           def file_read?(agent_id, file_path)
             @mutex.synchronize do
               return false unless @read_files[agent_id]
 
-              @read_files[agent_id].include?(File.expand_path(file_path))
+              expanded_path = File.expand_path(file_path)
+              stored_digest = @read_files[agent_id][expanded_path]
+              return false unless stored_digest
+
+              # Check if file still exists and matches stored digest
+              return false unless File.exist?(expanded_path)
+
+              current_digest = Digest::SHA256.hexdigest(File.read(expanded_path))
+              current_digest == stored_digest
+            end
+          end
+
+          # Get all read files with digests for snapshot
+          #
+          # @param agent_id [Symbol] The agent identifier
+          # @return [Hash] { file_path => digest }
+          def get_read_files(agent_id)
+            @mutex.synchronize do
+              @read_files[agent_id]&.dup || {}
+            end
+          end
+
+          # Restore read files with digests from snapshot
+          #
+          # @param agent_id [Symbol] The agent identifier
+          # @param files_with_digests [Hash] { file_path => digest }
+          # @return [void]
+          def restore_read_files(agent_id, files_with_digests)
+            @mutex.synchronize do
+              @read_files[agent_id] = files_with_digests.dup
             end
           end