prompt_objects 0.2.0 → 0.3.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 0fe57111f23be7238465e5a386d48d8d694cd4599e2113e6378150951efcd303
-  data.tar.gz: '07192c4f47af2d7aa9530fdc61b65cb97fba77ae84133a30533d3f07314e0b79'
+  metadata.gz: 676489da9fefa02b4d187f99137469c6cc0011d99e7e43b72574c0d57b19c8a7
+  data.tar.gz: acd7c61e837ac00485cb8286cf95caa3d2103c82d07aa5fef7ce49185a3a0f6a
 SHA512:
-  metadata.gz: 66d2e26018f49685968bccfd075b2e7cf74fb4bd0cbdf0730f4f3c68c3101be232235beb86a8a1cde73bdf2707c1a679b916d53c8b4c1fb2f9d455d965e4d246
-  data.tar.gz: 5fb3e4b088aaf7a6a99e08e144f090c68e2a63b0b009fe43929d93daa8da7b9e6b0c330ae048c2d6bda9fbcdef385000c0f255695a539e922393fc3c84a59e84
+  metadata.gz: dda01aa39c36088e3433894b004994b4a24b735bd6b59bb6ae510943be0f6f83931d1e45fa34051b4f49a7243b606b7ac71b8790f8e5701f9652373dff29d756
+  data.tar.gz: e02fb7e206ced06445391d81012a61d790277f4fc0f46343e3e0391f46da58aafd3a047abd13147e0c2e8fe150284b5333759f27ab1037ccb6a7e8020d18316a

data/CHANGELOG.md

@@ -0,0 +1,68 @@
+# Changelog
+
+All notable changes to PromptObjects are documented in this file.
+
+## [0.3.0] - 2025-02-05
+
+### Added
+
+- **Token usage & cost tracking** — Track input/output tokens and estimated costs per session and across delegation trees. Includes per-model pricing table and a Usage Panel in the web UI (right-click a thread to view).
+- **Ollama & OpenRouter support** — Connect to local Ollama models or OpenRouter's model marketplace. Both reuse the OpenAI adapter with configurable base URLs. Ollama models are auto-discovered from the local API.
+- **Thread export** — Export any conversation thread as Markdown or JSON, including full delegation chains. Delegation sub-threads render inline next to the tool call that triggered them, preserving the actual flow of work. Available via right-click context menu or REST API.
+- **ARC-AGI-1 template** — A template for solving ARC-AGI challenges with a solver PO, data manager PO, and 7 custom grid primitives (load, render, diff, info, find objects, transform, test solution).
+- **Persistent event log** — Message bus events are now persisted to SQLite for replay and debugging.
+- **REST message endpoint & events API** — Send messages to POs and retrieve bus events via HTTP. Includes server discovery for CLI commands.
+- **CLI `message` and `events` commands** — Interact with a running environment from the command line without opening the web UI.
+
+### Fixed
+
+- Custom primitives (created by POs or from templates) now auto-load on environment startup. Previously they were saved to `env/primitives/` but never registered on restart.
+- Message serialization bugs that caused crashes when tool calls contained non-string values.
+- Frontend auto-rebuilds when running `prompt_objects serve` in development.
+- Delegation sub-threads in exports now appear inline after the triggering tool call, not at the bottom of the document.
+- Tool result truncation limit increased from 2,000 to 10,000 characters to preserve detail in exports.
+- Full message content stored in bus; truncation applied only at display time.
+
+## [0.2.0] - 2025-01-23
+
+### Added
+
+- **GitHub Actions CI** — Automated test suite running on Ruby 3.2, 3.3, and 3.4.
+- **Conversation threads with delegation isolation** — Each PO-to-PO delegation runs in its own thread, keeping conversations clean.
+- **Thread sidebar** — Navigate between threads with auto-naming and instant feedback on creation.
+- **Real-time capability updates** — Adding/removing capabilities broadcasts changes to the web UI immediately.
+- **PO prompt editing** — Edit a Prompt Object's system prompt directly in the web UI with auto-save back to the markdown file.
+- **`modify_prompt` universal capability** — POs can rewrite their own system prompts at runtime.
+- **Environment recovery tools** — `remove_capability` and `delete_primitive` for cleaning up broken state.
+- **Streaming tool calls** — Tool call chains display in real-time as they execute, not just after completion.
+- **Capabilities panel** — Visual display of each PO's available primitives and PO-to-PO capabilities.
+- **Core PromptObject tests** — Unit test suite for the core framework.
+
+### Fixed
+
+- Session store binding and Gemini model name resolution.
+- Tool results missing function name for Gemini API compatibility.
+- `tool_calls` Hash vs ToolCall object handling across all adapters.
+- Claude API response parsing for new PO chat updates.
+- Thread switching now immediately shows the new thread on creation.
+- Session message counts and cross-session response routing.
+
+## [0.1.0] - 2025-01-15
+
+### Added
+
+- **Core framework** — Markdown files with YAML frontmatter act as LLM-backed autonomous entities.
+- **Unified capability interface** — Primitives (Ruby) and Prompt Objects (Markdown) share the same `receive(message, context:)` interface.
+- **Built-in primitives** — `read_file`, `list_files`, `write_file`, `http_get`.
+- **Universal capabilities** — `ask_human`, `think`, `request_capability`, `create_capability`, `add_capability`.
+- **Multi-provider LLM support** — OpenAI, Anthropic, and Gemini adapters with model selection UI.
+- **PO-to-PO communication** — Prompt Objects can call each other as capabilities through the message bus.
+- **Self-modification** — POs can create new Prompt Objects and primitives at runtime (with human approval).
+- **Web UI** — React frontend with real-time WebSocket updates, split-view layout, markdown rendering.
+- **Notification system** — Non-blocking human-in-the-loop via `ask_human` with notification bell and dropdown.
+- **Live filesystem watching** — Changes to `.md` files in the objects directory are reflected immediately.
+- **SQLite session storage** — Persistent conversation history with WAL mode for concurrent access.
+- **Environment management** — Create, list, and manage isolated environments with `prompt_objects env` commands.
+- **Templates** — Bootstrap new environments from templates (`basic`, `pair`, `team`, and more).
+- **MCP server mode** — Expose POs as tools via the Model Context Protocol for external client integration.
+- **CLI** — `prompt_objects` command with subcommands for environment management, serving, and interaction.

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    prompt_objects (0.1.0)
+    prompt_objects (0.3.0)
       anthropic (~> 1.0)
       async-websocket (~> 0.28)
       falcon (~> 0.50)

data/README.md

@@ -45,8 +45,8 @@ gem install prompt_objects
 # Create an environment from a template
 prompt_objects env create my-project --template basic
 
-# Run the web interface
-prompt_objects --env my-project
+# Run and open the web interface
+prompt_objects serve my-project --open
 ```
 
 ### Environment Commands

data/exe/prompt_objects

@@ -264,7 +264,7 @@ class REPL
       time = entry[:timestamp].strftime("%H:%M:%S")
       from = entry[:from]
       to = entry[:to]
-      msg = entry[:message]
+      msg = entry[:summary]
 
       line = "#{time}  #{from} -> #{to}: #{msg}"
       line = line[0, 70] + "..." if line.length > 73
@@ -283,6 +283,8 @@ def print_main_help
     Commands:
       env                          Manage environments (create, list, export, etc.)
       serve <env>                  Run environment as a web server
+      message <env> <po> "text"    Send a message to a PO and print the response
+      events <env> [--session ID]  Show recent events from the message bus
       repl [name] [objects_dir]    Start interactive REPL with a prompt object
       help                         Show this help message
 
@@ -293,6 +295,8 @@ def print_main_help
       prompt_objects env create demo --template demo  # Create from template
       prompt_objects serve my-env                 # Start web UI
       prompt_objects serve my-env --open          # Start and open browser
+      prompt_objects message my-env solver "Hello"     # Send a message
+      prompt_objects message my-env solver "Hello" --json  # JSON output
   HELP
 end
 
@@ -476,6 +480,9 @@ def run_serve(args)
   elsif options[:web]
     require "prompt_objects/server"
 
+    # Rebuild frontend to ensure built assets match source
+    rebuild_frontend
+
     # Open browser if requested
     if options[:open]
       url = "http://#{options[:host]}:#{options[:port]}"
@@ -491,6 +498,31 @@ def run_serve(args)
   end
 end
 
+def rebuild_frontend
+  # Find the frontend directory relative to this executable
+  gem_root = File.expand_path("../..", __FILE__)
+  frontend_dir = File.join(gem_root, "frontend")
+
+  unless Dir.exist?(frontend_dir)
+    # No frontend source (e.g., installed as gem without source) — skip
+    return
+  end
+
+  node_modules = File.join(frontend_dir, "node_modules")
+  unless Dir.exist?(node_modules)
+    puts "Frontend dependencies not installed. Run: cd frontend && npm install"
+    return
+  end
+
+  print "Building frontend... "
+  result = system("npm run build --prefix #{frontend_dir}", out: File::NULL, err: File::NULL)
+  if result
+    puts "done"
+  else
+    puts "failed (serving stale assets)"
+  end
+end
+
 def open_browser(url)
   case RUBY_PLATFORM
   when /darwin/
@@ -538,6 +570,356 @@ def load_all_objects(runtime, env_path)
   end
 end
 
+def print_message_help
+  puts <<~HELP
+    Usage: prompt_objects message [options] <environment> <po_name> "message"
+
+    Send a message to a prompt object and print the response.
+
+    Arguments:
+      environment  Environment name or path
+      po_name      Name of the prompt object to message
+      message      The message to send (quoted string)
+
+    Options:
+      --json           Output response as JSON
+      --events         Also print the event log for this interaction
+      --new-thread     Start a new thread for this message
+      --session ID     Use a specific session ID
+      --help, -h       Show this help
+
+    If a server is running for the environment, the message is sent via
+    HTTP to the running server (and streams to the web UI in real time).
+    Otherwise, a standalone runtime is started for the request.
+
+    Examples:
+      prompt_objects message my-env solver "What patterns do you see?"
+      prompt_objects message my-env solver "Solve task 1" --json
+      prompt_objects message my-env solver "Try again" --new-thread
+  HELP
+end
+
+def run_message(args)
+  options = {
+    json: false,
+    events: false,
+    new_thread: false,
+    session_id: nil
+  }
+
+  positional = []
+  skip_next = false
+
+  args.each_with_index do |arg, i|
+    if skip_next
+      skip_next = false
+      next
+    end
+
+    case arg
+    when "--json"
+      options[:json] = true
+    when "--events"
+      options[:events] = true
+    when "--new-thread"
+      options[:new_thread] = true
+    when "--session"
+      options[:session_id] = args[i + 1]
+      skip_next = true
+    when "--help", "-h"
+      print_message_help
+      exit 0
+    else
+      positional << arg
+    end
+  end
+
+  env_name = positional[0]
+  po_name = positional[1]
+  message_text = positional[2]
+
+  unless env_name && po_name && message_text
+    puts "Error: environment, po_name, and message are required"
+    puts "Run 'prompt_objects message --help' for usage"
+    exit 1
+  end
+
+  # Resolve environment
+  env_path = resolve_environment(env_name)
+  unless env_path
+    $stderr.puts "Error: environment '#{env_name}' not found"
+    exit 1
+  end
+
+  # Check for running server
+  require "prompt_objects/server"
+  server_info = PromptObjects::Server.read_server_file(env_path)
+
+  if server_info
+    # Send via HTTP to running server
+    send_message_via_http(server_info, po_name, message_text, options)
+  else
+    # Standalone mode
+    send_message_standalone(env_path, po_name, message_text, options)
+  end
+end
+
+def send_message_via_http(server_info, po_name, message_text, options)
+  require "net/http"
+  require "uri"
+
+  uri = URI("http://#{server_info[:host]}:#{server_info[:port]}/api/prompt_objects/#{po_name}/message")
+
+  body = { message: message_text }
+  body[:new_thread] = true if options[:new_thread]
+  body[:session_id] = options[:session_id] if options[:session_id]
+
+  begin
+    response = Net::HTTP.post(uri, body.to_json, "Content-Type" => "application/json")
+    data = JSON.parse(response.body)
+
+    if data["error"]
+      $stderr.puts "Error: #{data['error']}"
+      exit 1
+    end
+
+    if options[:json]
+      puts JSON.pretty_generate(data)
+    else
+      puts data["response"]
+    end
+
+    if options[:events] && data["session_id"]
+      events_uri = URI("http://#{server_info[:host]}:#{server_info[:port]}/api/events/session/#{data['session_id']}")
+      events_response = Net::HTTP.get(events_uri)
+      events_data = JSON.parse(events_response)
+
+      if events_data["events"]&.any?
+        $stderr.puts
+        $stderr.puts "--- Events ---"
+        events_data["events"].each do |e|
+          $stderr.puts "#{e['timestamp']}  #{e['from']} -> #{e['to']}: #{e['summary']}"
+        end
+        $stderr.puts "--------------"
+      end
+    end
+  rescue Errno::ECONNREFUSED
+    $stderr.puts "Error: Server at #{server_info[:host]}:#{server_info[:port]} is not responding"
+    $stderr.puts "The .server file may be stale. Try running standalone."
+    exit 1
+  end
+end
+
+def send_message_standalone(env_path, po_name, message_text, options)
+  runtime = PromptObjects::Runtime.new(env_path: env_path)
+  load_all_objects(runtime, env_path)
+
+  po = runtime.registry.get(po_name)
+  unless po.is_a?(PromptObjects::PromptObject)
+    $stderr.puts "Error: prompt object '#{po_name}' not found"
+    $stderr.puts "Available: #{runtime.loaded_objects.join(', ')}"
+    exit 1
+  end
+
+  # Create new thread if requested
+  po.new_thread if options[:new_thread]
+
+  # Switch session if specified
+  if options[:session_id]
+    po.switch_session(options[:session_id])
+  end
+
+  session_id = po.session_id
+  log_start = runtime.bus.log.length
+
+  context = runtime.context
+  context.current_capability = "human"
+
+  # Log to bus
+  runtime.bus.publish(from: "human", to: po.name, message: message_text, session_id: session_id)
+
+  response = po.receive(message_text, context: context)
+
+  # Log response to bus
+  runtime.bus.publish(from: po.name, to: "human", message: response, session_id: session_id)
+
+  if options[:json]
+    data = {
+      response: response,
+      po_name: po.name,
+      session_id: session_id,
+      event_count: runtime.bus.log.length - log_start
+    }
+    puts JSON.pretty_generate(data)
+  else
+    puts response
+  end
+
+  if options[:events]
+    entries = runtime.bus.log[log_start..]
+    if entries&.any?
+      $stderr.puts
+      $stderr.puts "--- Events ---"
+      entries.each do |entry|
+        time = entry[:timestamp].strftime("%H:%M:%S")
+        $stderr.puts "#{time}  #{entry[:from]} -> #{entry[:to]}: #{entry[:summary]}"
+      end
+      $stderr.puts "--------------"
+    end
+  end
+end
+
+def print_events_help
+  puts <<~HELP
+    Usage: prompt_objects events [options] <environment>
+
+    Show recent events from the message bus.
+
+    Arguments:
+      environment  Environment name or path
+
+    Options:
+      --session ID     Show events for a specific session
+      --count N        Number of events to show (default: 50)
+      --json           Output as JSON
+      --help, -h       Show this help
+
+    Examples:
+      prompt_objects events my-env
+      prompt_objects events my-env --session abc-123
+      prompt_objects events my-env --count 100 --json
+  HELP
+end
+
+def run_events(args)
+  options = { count: 50, session_id: nil, json: false }
+  positional = []
+  skip_next = false
+
+  args.each_with_index do |arg, i|
+    if skip_next
+      skip_next = false
+      next
+    end
+
+    case arg
+    when "--session"
+      options[:session_id] = args[i + 1]
+      skip_next = true
+    when "--count"
+      options[:count] = args[i + 1].to_i
+      skip_next = true
+    when "--json"
+      options[:json] = true
+    when "--help", "-h"
+      print_events_help
+      exit 0
+    else
+      positional << arg
+    end
+  end
+
+  env_name = positional[0]
+  unless env_name
+    puts "Error: environment name required"
+    puts "Run 'prompt_objects events --help' for usage"
+    exit 1
+  end
+
+  env_path = resolve_environment(env_name)
+  unless env_path
+    $stderr.puts "Error: environment '#{env_name}' not found"
+    exit 1
+  end
+
+  # Check for running server
+  require "prompt_objects/server"
+  server_info = PromptObjects::Server.read_server_file(env_path)
+
+  if server_info
+    show_events_via_http(server_info, options)
+  else
+    show_events_standalone(env_path, options)
+  end
+end
+
+def show_events_via_http(server_info, options)
+  require "net/http"
+  require "uri"
+
+  if options[:session_id]
+    uri = URI("http://#{server_info[:host]}:#{server_info[:port]}/api/events/session/#{options[:session_id]}")
+  else
+    uri = URI("http://#{server_info[:host]}:#{server_info[:port]}/api/events?count=#{options[:count]}")
+  end
+
+  response = Net::HTTP.get(uri)
+  data = JSON.parse(response)
+
+  if options[:json]
+    puts JSON.pretty_generate(data)
+  else
+    print_events(data["events"] || [])
+  end
+rescue Errno::ECONNREFUSED
+  $stderr.puts "Error: Server not responding"
+  exit 1
+end
+
+def show_events_standalone(env_path, options)
+  db_path = File.join(env_path, "sessions.db")
+  unless File.exist?(db_path)
+    puts "No event history (sessions.db not found)"
+    exit 0
+  end
+
+  store = PromptObjects::Session::Store.new(db_path)
+
+  events = if options[:session_id]
+    store.get_events(session_id: options[:session_id])
+  else
+    store.get_recent_events(options[:count])
+  end
+
+  if options[:json]
+    formatted = events.map do |e|
+      {
+        id: e[:id],
+        from: e[:from],
+        to: e[:to],
+        summary: e[:summary],
+        message: e[:message],
+        timestamp: e[:timestamp]&.iso8601,
+        session_id: e[:session_id]
+      }
+    end
+    puts JSON.pretty_generate(formatted)
+  else
+    print_events(events)
+  end
+
+  store.close
+end
+
+def print_events(events)
+  if events.empty?
+    puts "No events found."
+    return
+  end
+
+  events.each do |e|
+    ts = e[:timestamp] || e["timestamp"]
+    ts = ts.is_a?(Time) ? ts.strftime("%H:%M:%S") : ts&.split("T")&.last&.split(".")&.first
+    from = e[:from] || e["from"]
+    to = e[:to] || e["to"]
+    summary = e[:summary] || e["summary"]
+
+    line = "#{ts}  #{from} -> #{to}: #{summary}"
+    line = line[0, 120] + "..." if line.length > 123
+    puts line
+  end
+end
+
 # === Main Entry Point ===
 
 def run_env(args)
@@ -561,6 +943,10 @@ def main
     run_repl(args)
   when "serve"
     run_serve(args)
+  when "message", "msg"
+    run_message(args)
+  when "events"
+    run_events(args)
   when "help", "--help", "-h"
     print_main_help
   else

data/frontend/src/App.tsx

@@ -7,11 +7,12 @@ import { PODetail } from './components/PODetail'
 import { MessageBus } from './components/MessageBus'
 import { NotificationPanel } from './components/NotificationPanel'
 import { ThreadsSidebar } from './components/ThreadsSidebar'
+import { UsagePanel } from './components/UsagePanel'
 
 export default function App() {
-  const { sendMessage, respondToNotification, createSession, switchSession, switchLLM, createThread, updatePrompt } =
+  const { sendMessage, respondToNotification, createSession, switchSession, switchLLM, createThread, updatePrompt, requestUsage, exportThread } =
     useWebSocket()
-  const { selectedPO, busOpen, notifications } = useStore()
+  const { selectedPO, busOpen, notifications, usageData, clearUsageData } = useStore()
   const selectedPOData = useSelectedPO()
   const [splitView, setSplitView] = useState(true) // Default to split view
 
@@ -47,6 +48,8 @@ export default function App() {
                   po={selectedPOData}
                   switchSession={switchSession}
                   createThread={createThread}
+                  requestUsage={requestUsage}
+                  exportThread={exportThread}
                 />
               </aside>
             )}
@@ -81,7 +84,7 @@ export default function App() {
 
         {/* Message Bus sidebar */}
         {busOpen && (
-          <aside className="w-80 border-l border-po-border bg-po-surface overflow-hidden">
+          <aside className="w-80 flex-shrink-0 border-l border-po-border bg-po-surface overflow-hidden">
             <MessageBus />
           </aside>
         )}
@@ -91,6 +94,11 @@ export default function App() {
       {notifications.length > 0 && (
         <NotificationPanel respondToNotification={respondToNotification} />
       )}
+
+      {/* Usage panel modal */}
+      {usageData && (
+        <UsagePanel usage={usageData as any} onClose={clearUsageData} />
+      )}
     </div>
   )
 }

data/frontend/src/components/ContextMenu.tsx

@@ -0,0 +1,67 @@
+import { useEffect, useRef } from 'react'
+
+interface ContextMenuItem {
+  label: string
+  onClick: () => void
+  icon?: string
+  danger?: boolean
+}
+
+interface ContextMenuProps {
+  x: number
+  y: number
+  items: ContextMenuItem[]
+  onClose: () => void
+}
+
+export function ContextMenu({ x, y, items, onClose }: ContextMenuProps) {
+  const menuRef = useRef<HTMLDivElement>(null)
+
+  useEffect(() => {
+    const handleClickOutside = (e: MouseEvent) => {
+      if (menuRef.current && !menuRef.current.contains(e.target as Node)) {
+        onClose()
+      }
+    }
+    const handleEscape = (e: KeyboardEvent) => {
+      if (e.key === 'Escape') onClose()
+    }
+
+    document.addEventListener('mousedown', handleClickOutside)
+    document.addEventListener('keydown', handleEscape)
+    return () => {
+      document.removeEventListener('mousedown', handleClickOutside)
+      document.removeEventListener('keydown', handleEscape)
+    }
+  }, [onClose])
+
+  // Adjust position to stay within viewport
+  const adjustedStyle = {
+    top: y,
+    left: x,
+  }
+
+  return (
+    <div
+      ref={menuRef}
+      className="fixed z-50 bg-po-surface border border-po-border rounded-lg shadow-xl py-1 min-w-[160px]"
+      style={adjustedStyle}
+    >
+      {items.map((item, idx) => (
+        <button
+          key={idx}
+          onClick={() => {
+            item.onClick()
+            onClose()
+          }}
+          className={`w-full text-left px-3 py-1.5 text-xs hover:bg-po-bg transition-colors flex items-center gap-2 ${
+            item.danger ? 'text-red-400 hover:text-red-300' : 'text-gray-300 hover:text-white'
+          }`}
+        >
+          {item.icon && <span>{item.icon}</span>}
+          {item.label}
+        </button>
+      ))}
+    </div>
+  )
+}

data/frontend/src/components/MessageBus.tsx

@@ -41,9 +41,10 @@ export function MessageBus() {
                 </span>
               </div>
               <div className="text-gray-300 break-words">
-                {msg.content.length > 200
-                  ? msg.content.slice(0, 200) + '...'
-                  : msg.content}
+                {(() => {
+                  const text = msg.summary || (typeof msg.content === 'string' ? msg.content : JSON.stringify(msg.content))
+                  return text.length > 200 ? text.slice(0, 200) + '...' : text
+                })()}
               </div>
             </div>
           ))

data/frontend/src/components/ModelSelector.tsx

@@ -33,6 +33,8 @@ export function ModelSelector({ switchLLM }: Props) {
     openai: 'OpenAI',
     anthropic: 'Anthropic',
     gemini: 'Gemini',
+    ollama: 'Ollama',
+    openrouter: 'OpenRouter',
   }
 
   return (
@@ -62,7 +64,9 @@ export function ModelSelector({ switchLLM }: Props) {
               <div className="px-3 py-2 bg-po-bg text-xs font-medium text-gray-400 uppercase tracking-wide flex items-center justify-between">
                 <span>{providerNames[provider.name] || provider.name}</span>
                 {!provider.available && (
-                  <span className="text-red-400 text-[10px] normal-case">No API Key</span>
+                  <span className="text-red-400 text-[10px] normal-case">
+                    {provider.name === 'ollama' ? 'Not Running' : 'No API Key'}
+                  </span>
                 )}
               </div>
               {provider.models.map((model) => {