openclacky 0.5.1 → 0.5.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e8253fadf5425c27f9b8ebfecfe195c4a21c0f0d944e0c22030dc77411dda41e
-  data.tar.gz: 345301c2cc266afab03564ea08c3801bc1adae239e129d7fc9ac99280e47ec07
+  metadata.gz: 52ab39c40c69d821b45d01ed1d14086a50446309421891e9cd6724a9b32864cf
+  data.tar.gz: 0d61ef02a23f8a04157f8b7b2308a4f0c68fd37bba1407894b9841cd707c6873
 SHA512:
-  metadata.gz: 2fe927a2d4c86be0ecd45e4defe9ae8646ded14153ac05178bf751a29f108b4ed9744f2e7fd64714437e70e4be1dcc092640154400e9c639deb29d558c42fa0e
-  data.tar.gz: 25f7e33a14d03a1a03e5a0947dab559a5e27244cd7333342bd1014033cce6fac141ae3794eb0fc6dd435260045a695c161bbe07f39f0a253974939cda048edb9
+  metadata.gz: d654bd29a093bf23f1e0c162867251e08a28ec722deaae7a9d234f2aed913329d991df29391778612c458c51b6775881ee0867cb2d959c0f7072b14e662d0c38
+  data.tar.gz: 8056dff60d1c845751021e2c852e58071360c8d6a6b81cec72be238e647521ae36a211d72af88602239f8abcab5a15b7a345b8efd98e68b62a0a6dfde8a63b23

data/.clackyrules

@@ -22,7 +22,9 @@ It provides chat functionality and autonomous AI agent capabilities with tool us
 - Use frozen string literals: `# frozen_string_literal: true`
 - Keep classes focused and single-responsibility
 - Use meaningful variable and method names
+- **IMPORTANT**: All code comments must be written in English
 - Add descriptive comments for complex logic
+- Use clear, self-documenting code with English naming
 
 ### Architecture Patterns
 - Tools inherit from `Clacky::Tools::Base`

data/README.md

@@ -12,6 +12,7 @@ A command-line interface for interacting with AI models. OpenClacky supports Ope
 - 🎨 Colorful terminal output
 - 🌐 OpenAI-compatible API support (OpenAI, Gitee AI, DeepSeek, etc.)
 - 🛠️ Rich built-in tools: file operations, web search, code execution, and more
+- ⚡ Prompt caching support for Claude models (reduces costs up to 90%)
 
 ## Installation
 

data/lib/clacky/agent.rb

@@ -2,13 +2,14 @@
 
 require "securerandom"
 require "json"
-require "readline"
+require "tty-prompt"
 require "set"
 require_relative "utils/arguments_parser"
 
 module Clacky
   class Agent
-    attr_reader :session_id, :messages, :iterations, :total_cost, :working_dir, :created_at, :total_tasks, :todos
+    attr_reader :session_id, :messages, :iterations, :total_cost, :working_dir, :created_at, :total_tasks, :todos,
+                :cache_stats
 
     # Pricing per 1M tokens (approximate - adjust based on actual model)
     PRICING = {
@@ -65,6 +66,12 @@ module Clacky
       @todos = []  # Store todos in memory
       @iterations = 0
       @total_cost = 0.0
+      @cache_stats = {
+        cache_creation_input_tokens: 0,
+        cache_read_input_tokens: 0,
+        total_requests: 0,
+        cache_hit_requests: 0
+      }
       @start_time = nil
       @working_dir = working_dir || Dir.pwd
       @created_at = Time.now.iso8601
@@ -91,6 +98,14 @@ module Clacky
       @created_at = session_data[:created_at]
       @total_tasks = session_data.dig(:stats, :total_tasks) || 0
 
+      # Restore cache statistics if available
+      @cache_stats = session_data.dig(:stats, :cache_stats) || {
+        cache_creation_input_tokens: 0,
+        cache_read_input_tokens: 0,
+        total_requests: 0,
+        cache_hit_requests: 0
+      }
+
       # Check if the session ended with an error
       last_status = session_data.dig(:stats, :last_status)
       last_error = session_data.dig(:stats, :last_error)
@@ -103,7 +118,7 @@ module Clacky
           @messages = @messages[0...last_user_index]
 
           # Trigger a hook to notify about the rollback
-          trigger_hook(:session_rollback, {
+          @hooks.trigger(:session_rollback, {
             reason: "Previous session ended with error",
             error_message: last_error,
             rolled_back_message_index: last_user_index
@@ -122,7 +137,14 @@ module Clacky
       # Add system prompt as the first message if this is the first run
       if @messages.empty?
         system_prompt = build_system_prompt
-        @messages << { role: "system", content: system_prompt }
+        system_message = { role: "system", content: system_prompt }
+
+        # Enable caching for system prompt if configured and model supports it
+        if @config.enable_prompt_caching
+          system_message[:cache_control] = { type: "ephemeral" }
+        end
+
+        @messages << system_message
       end
 
       @messages << { role: "user", content: user_input }
@@ -222,7 +244,8 @@ module Clacky
         total_iterations: @iterations,
         total_cost_usd: @total_cost.round(4),
         duration_seconds: @start_time ? (Time.now - @start_time).round(2) : 0,
-        last_status: status.to_s
+        last_status: status.to_s,
+        cache_stats: @cache_stats
       }
 
       # Add error message if status is error
@@ -240,6 +263,7 @@ module Clacky
           max_iterations: @config.max_iterations,
           max_cost_usd: @config.max_cost_usd,
           enable_compression: @config.enable_compression,
+          enable_prompt_caching: @config.enable_prompt_caching,
           keep_recent_messages: @config.keep_recent_messages,
           max_tokens: @config.max_tokens,
           verbose: @config.verbose
@@ -366,7 +390,8 @@ module Clacky
             model: @config.model,
             tools: tools_to_send,
             max_tokens: @config.max_tokens,
-            verbose: @config.verbose
+            verbose: @config.verbose,
+            enable_caching: @config.enable_prompt_caching
           )
         rescue Faraday::ConnectionFailed, Faraday::TimeoutError, Errno::ECONNREFUSED, Errno::ETIMEDOUT => e
           retries += 1
@@ -386,6 +411,54 @@ module Clacky
 
         track_cost(response[:usage])
 
+        # Handle truncated responses (when max_tokens limit is reached)
+        if response[:finish_reason] == "length"
+          # Count recent truncations to prevent infinite loops
+          recent_truncations = @messages.last(5).count { |m|
+            m[:role] == "user" && m[:content]&.include?("[SYSTEM] Your response was truncated")
+          }
+
+          if recent_truncations >= 2
+            # Too many truncations - task is too complex
+            progress.finish
+            puts "\n⚠️  Response truncated multiple times. Task is too complex for a single response." if @config.verbose
+
+            # Create a response that tells the user to break down the task
+            error_response = {
+              content: "I apologize, but this task is too complex to complete in a single response. " \
+                       "Please break it down into smaller steps, or reduce the amount of content to generate at once.\n\n" \
+                       "For example, when creating a long document:\n" \
+                       "1. First create the file with a basic structure\n" \
+                       "2. Then use edit() to add content section by section",
+              finish_reason: "stop",
+              tool_calls: nil
+            }
+
+            # Add this as an assistant message so it appears in conversation
+            @messages << {
+              role: "assistant",
+              content: error_response[:content]
+            }
+
+            return error_response
+          end
+
+          # Insert system message to guide LLM to retry with smaller steps
+          @messages << {
+            role: "user",
+            content: "[SYSTEM] Your response was truncated due to length limit. Please retry with a different approach:\n" \
+                     "- For long file content: create the file with structure first, then use edit() to add content section by section\n" \
+                     "- Break down large tasks into multiple smaller steps\n" \
+                     "- Avoid putting more than 2000 characters in a single tool call argument\n" \
+                     "- Use multiple tool calls instead of one large call"
+          }
+
+          puts "⚠️  Response truncated due to length limit. Retrying with smaller steps..." if @config.verbose
+
+          # Recursively retry
+          return think(&block)
+        end
+
         # Add assistant response to messages
         msg = { role: "assistant" }
         # Always include content field (some APIs require it even with tool_calls)
@@ -543,6 +616,18 @@ module Clacky
       input_cost = (usage[:prompt_tokens] / 1_000_000.0) * PRICING[:input]
       output_cost = (usage[:completion_tokens] / 1_000_000.0) * PRICING[:output]
       @total_cost += input_cost + output_cost
+
+      # Track cache usage statistics
+      @cache_stats[:total_requests] += 1
+
+      if usage[:cache_creation_input_tokens]
+        @cache_stats[:cache_creation_input_tokens] += usage[:cache_creation_input_tokens]
+      end
+
+      if usage[:cache_read_input_tokens]
+        @cache_stats[:cache_read_input_tokens] += usage[:cache_read_input_tokens]
+        @cache_stats[:cache_hit_requests] += 1
+      end
     end
 
     def compress_messages_if_needed
@@ -582,7 +667,15 @@ module Clacky
         summary = summarize_messages(messages_to_compress)
 
         # Rebuild messages array: [system, summary, recent_messages]
-        @messages = [system_msg, summary, *recent_messages].compact
+        # Preserve cache_control on system message if it exists
+        rebuilt_messages = [system_msg, summary, *recent_messages].compact
+
+        # Re-apply cache control to system message if caching is enabled
+        if @config.enable_prompt_caching && rebuilt_messages.first&.dig(:role) == "system"
+          rebuilt_messages.first[:cache_control] = { type: "ephemeral" }
+        end
+
+        @messages = rebuilt_messages
 
         final_size = @messages.size
 
@@ -610,7 +703,7 @@ module Clacky
 
       # Track which messages to include
       messages_to_include = Set.new
-      
+
       # Start from the end and work backwards
       i = messages.size - 1
       messages_collected = 0
@@ -631,13 +724,13 @@ module Clacky
         # If this is an assistant message with tool_calls, we MUST include ALL corresponding tool results
         if msg[:role] == "assistant" && msg[:tool_calls]
           tool_call_ids = msg[:tool_calls].map { |tc| tc[:id] }
-          
+
           # Find all tool results that belong to this assistant message
           # They should be in the messages immediately following this assistant message
           j = i + 1
           while j < messages.size
             next_msg = messages[j]
-            
+
             # If we find a tool result for one of our tool_calls, include it
             if next_msg[:role] == "tool" && tool_call_ids.include?(next_msg[:tool_call_id])
               messages_to_include.add(j)
@@ -645,7 +738,7 @@ module Clacky
               # Stop when we hit a non-tool message (start of next turn)
               break
             end
-            
+
             j += 1
           end
         end
@@ -766,18 +859,28 @@ module Clacky
       prompt_text = format_tool_prompt(call)
       puts "\n❓ #{prompt_text}"
 
-      # Use Readline for better input handling (backspace, arrow keys, etc.)
-      response = Readline.readline("   (Enter/y to approve, n to deny, or provide feedback): ", true)
+      # Use TTY::Prompt for better input handling
+      tty_prompt = TTY::Prompt.new(interrupt: :exit)
 
-      if response.nil?  # Handle EOF/pipe input
+      begin
+        response = tty_prompt.ask("   (Enter/y to approve, n to deny, or provide feedback):", required: false) do |q|
+          q.modify :strip
+        end
+      rescue TTY::Reader::InputInterrupt
+        # Handle Ctrl+C
+        puts
         return { approved: false, feedback: nil }
       end
 
-      response = response.chomp
+      # Handle nil response (EOF/pipe input)
+      if response.nil? || response.empty?
+        return { approved: true, feedback: nil }  # Empty means approved
+      end
+
       response_lower = response.downcase
 
-      # Empty response (just Enter) or "y"/"yes" = approved
-      if response.empty? || response_lower == "y" || response_lower == "yes"
+      # "y"/"yes" = approved
+      if response_lower == "y" || response_lower == "yes"
         return { approved: true, feedback: nil }
       end
 
@@ -978,6 +1081,7 @@ module Clacky
         iterations: @iterations,
         duration_seconds: Time.now - @start_time,
         total_cost_usd: @total_cost.round(4),
+        cache_stats: @cache_stats,
         messages: @messages,
         error: error
       }

data/lib/clacky/agent_config.rb

@@ -7,7 +7,8 @@ module Clacky
 
     attr_accessor :model, :max_iterations, :max_cost_usd, :timeout_seconds,
                   :permission_mode, :allowed_tools, :disallowed_tools,
-                  :max_tokens, :verbose, :enable_compression, :keep_recent_messages
+                  :max_tokens, :verbose, :enable_compression, :keep_recent_messages,
+                  :enable_prompt_caching
 
     def initialize(options = {})
       @model = options[:model] || "gpt-3.5-turbo"
@@ -21,6 +22,8 @@ module Clacky
       @verbose = options[:verbose] || false
       @enable_compression = options[:enable_compression].nil? ? true : options[:enable_compression]
       @keep_recent_messages = options[:keep_recent_messages] || 20
+      # Enable prompt caching by default for cost savings
+      @enable_prompt_caching = options[:enable_prompt_caching].nil? ? true : options[:enable_prompt_caching]
     end