prompt_objects 0.3.1 → 0.5.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 0a1ec32b66d34c2fe11c964759513487335794f9a8c336bd32f7ae9aeed5380c
-  data.tar.gz: 9a6650251228ab4dae25f16120e8ae8c1992e4458faf0c858894b6e4848ae80b
+  metadata.gz: 6dd3c14942ef80d1979ded542802ad9d1a24b32cd479aa4ba0a08bb0bf2e349d
+  data.tar.gz: 3642d5424aa827754dfefde82a9e9e5c770935ebf08e36cfc429d8f86e5e8502
 SHA512:
-  metadata.gz: 867fc4cb49114ace4af4682feff5a062eb414f0ce84e4be4e48632609d68cec2ac201be2cb43fb995ac5e9cb4ec0480399aa6eaad7580da81fd5451905b0ddf5
-  data.tar.gz: 4b3d3c3d96d45a2573829345fd1f822fa3f8918c9d4088cccc46d82ba4fa797bb7c9838e74d443017069f7cd9b9d343d3bd92219930b12098f036d038e05ef83
+  metadata.gz: f4e4449d2a8764a762c13ea14b6ccac17b07d1ae06a053a94e5b66ce7ebf38b8ca2903c140f0bb06810e09fa830bdac1dcfb8485d2f6f7fe32f7b0dffdf374bb
+  data.tar.gz: 1faa31a8c32f067cdfb133c06e6c3d45e55fec3381d86f649e8845b9d2c58bc367e2bed3821cbec4a92a18776b723ccf69c276f9e913347198816005c4608353

data/CHANGELOG.md

@@ -2,6 +2,38 @@
 
 All notable changes to PromptObjects are documented in this file.
 
+## [0.5.0] - 2026-02-13
+
+### Added
+
+- **Smalltalk System Browser redesign** — Complete frontend overhaul replacing the chat-app UI with a multi-pane object browser. POs are treated as live objects in a running image: permanent ObjectList (left pane), multi-pane Inspector with MethodList + SourcePane, REPL-style Workspace, and bottom Transcript. Warm charcoal + amber palette with Geist fonts. All panels resizable via drag handles.
+- **Collapsible inspector top pane** — The Methods + Source pane has a thin header bar with a collapse/expand toggle. When collapsed, the Workspace fills the full inspector height. Collapse state persists across PO switches.
+- **Source entry in method list** — A "Source" entry at the top of the method list provides a clear way to navigate back to the PO's prompt after inspecting a capability.
+- **Dynamic Ollama model discovery** — LLM config now queries the Ollama API for installed models instead of using a static list.
+
+### Fixed
+
+- **Capabilities disappearing on file save** — `po_modified` events were sending capabilities as plain string names instead of rich objects, overwriting the store. All serialization paths now emit consistent `{name, description, parameters}` objects.
+- **Centralized PO serialization** — Moved duplicated state/message/session serialization from WebSocketHandler and API routes into `PromptObject` (`to_state_hash`, `to_summary_hash`, `to_inspect_hash`). Eliminates inconsistent serialization as a class of bug.
+- **Missing items field in array tool schemas** — LLM APIs reject array parameters without an `items` field. Added a defensive sanitizer in `Capability#descriptor` as a fallback.
+- **OpenAI adapter error details** — 4xx errors from Ollama now surface the actual rejection reason instead of just "status 400".
+- **All WebSocket message types handled** — Added frontend handlers for `prompt_updated`, `llm_error`, `session_created`, and `session_switched`. Removed defensive normalization workarounds.
+
+## [0.4.0] - 2026-02-11
+
+### Added
+
+- **Spatial Canvas** — Three.js 2D visualization at `/canvas` showing POs as glowing hexagonal nodes with force-directed layout, tool call diamonds, animated message arcs with traveling particles, and click-to-inspect side panels. Real-time updates from the same WebSocket feed as the dashboard. Zoom, pan, and keyboard shortcuts (F to fit, Escape to deselect).
+- **PO-to-PO delegation broadcasting** — Server now broadcasts `po_delegation_started` and `po_delegation_completed` WebSocket events when one PO calls another. Delegated POs show as active in both the canvas (cyan glow, "called by X" status) and dashboard views. Replaces client-side inference from message history scanning.
+- **Ruby 4 support** — CI now tests against Ruby 4. Fixed empty required parameter handling for compatibility. Thanks to [@radanskoric](https://github.com/radanskoric) for the contribution! ([#2](https://github.com/works-on-your-machine/prompt_objects/pull/2))
+
+### Fixed
+
+- **WebSocket reconnection lifecycle** — Fixed duplicate connections on page refresh caused by zombie `onclose` handlers. Added socket identity guards, close-before-reconnect, and `handleMessageRef` pattern to prevent stale closures.
+- **Stale state on disconnect** — PO statuses now reset to idle when WebSocket disconnects, preventing stuck "thinking" indicators and stale streaming content.
+- **Chat input locked on disconnect** — Chat input is now only disabled when the PO is busy AND connected. Shows "Reconnecting..." indicator when disconnected instead of permanently locking.
+- **Tool calls not appearing on canvas** — Tool call visualization now extracts from PO message history instead of looking for a format bus messages don't use.
+
 ## [0.3.1] - 2025-02-08
 
 ### Added

data/CLAUDE.md

@@ -6,17 +6,21 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 **PromptObjects** is a Ruby framework where markdown files with LLM-backed behavior act as first-class autonomous entities. The core insight: **everything is a capability**—primitives (Ruby code) and Prompt-Objects (markdown files) share the same interface, differing only in interpretation complexity.
 
-**Current Status**: Design phase—`design-doc-v2.md` contains the full specification, `IMPLEMENTATION_PLAN.md` has detailed phased build plans. No implementation code exists yet.
+**Current Status**: v0.5.0 — The core framework is fully implemented and functional. The original 6-phase implementation plan is complete. Active development is focused on visualization, developer experience, and exploring new primitives. See `CHANGELOG.md` for release history and `design-doc-v2.md` / `IMPLEMENTATION_PLAN.md` for original design context.
 
 ## Architecture
 
 ```
-ENVIRONMENT
+RUNTIME (Environment)
 ├── CAPABILITY REGISTRY
 │   ├── PRIMITIVES (Ruby) - deterministic interpretation
-│   └── PROMPT-OBJECTS (Markdown) - semantic interpretation via LLM
-├── MESSAGE BUS - routes messages, logs for visualization
-└── TERMINAL UI (Charm) - capability bar, message log, conversation, input
+│   ├── PROMPT-OBJECTS (Markdown) - semantic interpretation via LLM
+│   └── UNIVERSAL CAPABILITIES - available to all POs automatically
+├── MESSAGE BUS - routes messages, logs to SQLite for replay
+├── SESSION STORE (SQLite) - persistent conversation threads, delegation tracking
+├── HUMAN QUEUE - non-blocking ask_human requests
+├── WEB SERVER (Sinatra + WebSocket) - serves React frontend
+└── MCP SERVER - exposes POs as tools via Model Context Protocol
 ```
 
 ### Unified Capability Interface
@@ -53,56 +57,120 @@ You are a careful, thoughtful file reader...
 Available to all Prompt-Objects automatically (no frontmatter declaration needed):
 - `ask_human` - pause for human input/confirmation
 - `think` - internal reasoning (not shown to human)
-- `request_capability` - ask environment for new capability
+- `create_capability` / `add_capability` / `remove_capability` - self-modification
+- `list_capabilities` / `list_primitives` - introspection
+- `create_primitive` / `add_primitive` / `delete_primitive` / `verify_primitive` / `modify_primitive` / `request_primitive` - primitive management
+- `modify_prompt` - rewrite own system prompt at runtime
 
-## Technology Stack
-
-- **Ruby** - core implementation
-- **LLM APIs** - OpenAI, Anthropic, Gemini (adapter pattern)
-- **Charm** - Terminal UI (Bubble Tea for interaction, Lipgloss for styling, Glamour for markdown)
-- **MCP** - Model Context Protocol integration
+### PO-to-PO Delegation
+When a PO calls another PO, the system creates an isolated delegation thread in the target PO. The caller's context is tracked so messages show correct provenance. Delegation start/complete events are broadcast via WebSocket for real-time UI updates.
 
-### TUI Architecture Note
+## Technology Stack
 
-The TUI uses a **single Bubble Tea program** with internal screen states (picker, wizard, main). Do NOT run multiple sequential `Bubbletea.run` calls—this causes terminal state corruption. See `docs/phase-8-environments.md` for details.
+- **Ruby** (>= 3.2, tested through Ruby 4) - core implementation
+- **LLM APIs** - OpenAI, Anthropic, Gemini, Ollama, OpenRouter (adapter pattern via `LLM::Factory`)
+- **Sinatra** - web server for REST API and static file serving
+- **faye-websocket** - real-time WebSocket communication
+- **React + TypeScript** - web frontend (dashboard, chat, capabilities panel)
+- **Three.js** - spatial canvas visualization (force-directed PO graph)
+- **SQLite** - session persistence and event log storage
+- **MCP** - Model Context Protocol server mode
 
-## Planned File Structure
+## File Structure
 
 ```
 prompt_objects/
-├── exe/prompt_objects          # CLI entrypoint
+├── exe/prompt_objects              # CLI entrypoint
 ├── lib/
-│   ├── prompt_objects.rb       # Main entry
+│   ├── prompt_objects.rb           # Main entry, requires all modules
 │   └── prompt_objects/
-│       ├── environment.rb      # Runtime container
-│       ├── capability.rb       # Base interface
-│       ├── prompt_object.rb    # PO implementation
-│       ├── primitive.rb        # Primitive tool wrapper
-│       ├── loader.rb           # Parses frontmatter + body
-│       ├── registry.rb         # Capability registration
-│       ├── message_bus.rb      # Message routing and logging
-│       ├── llm/                # LLM adapters
-│       ├── primitives/         # Built-in primitives (read_file, etc.)
-│       ├── universal/          # Universal capabilities
-│       ├── mcp/                # MCP integration
-│       └── ui/                 # Charm-based terminal UI
-├── objects/                    # Where Prompt-Objects live (.md files)
-└── primitives/                 # Optional user-defined primitives
+│       ├── environment.rb          # Runtime container (registry, bus, LLM, sessions)
+│       ├── capability.rb           # Base capability interface
+│       ├── prompt_object.rb        # PO implementation (LLM conversation loop)
+│       ├── primitive.rb            # Primitive tool wrapper
+│       ├── loader.rb               # Parses frontmatter + body from .md files
+│       ├── registry.rb             # Capability registration and lookup
+│       ├── message_bus.rb          # Message routing, logging, SQLite persistence
+│       ├── human_queue.rb          # Non-blocking human interaction queue
+│       ├── cli.rb                  # CLI command definitions
+│       ├── server.rb               # Web server setup
+│       ├── server/
+│       │   ├── app.rb              # Sinatra application
+│       │   ├── api/routes.rb       # REST API endpoints
+│       │   ├── websocket_handler.rb # WebSocket event handling
+│       │   └── file_watcher.rb     # Live .md file change detection
+│       ├── llm/
+│       │   ├── factory.rb          # Provider/model selection
+│       │   ├── response.rb         # Unified response object
+│       │   ├── pricing.rb          # Token cost calculation
+│       │   ├── openai_adapter.rb   # OpenAI + Ollama + OpenRouter
+│       │   ├── anthropic_adapter.rb
+│       │   └── gemini_adapter.rb
+│       ├── primitives/             # Built-in: read_file, list_files, write_file, http_get
+│       ├── universal/              # 14 universal capabilities (see list above)
+│       ├── connectors/             # Interface adapters (base, mcp)
+│       ├── mcp/                    # MCP server and tool definitions
+│       ├── session/
+│       │   └── store.rb            # SQLite session/thread persistence
+│       └── environment/
+│           ├── manager.rb          # Create/list/clone environments
+│           ├── manifest.rb         # Environment metadata (manifest.yml)
+│           ├── git.rb              # Auto-commit integration
+│           ├── exporter.rb         # Environment export
+│           └── importer.rb         # Environment import
+├── frontend/                       # React + TypeScript web UI
+│   └── src/
+│       ├── App.tsx
+│       ├── components/             # Dashboard, chat, capabilities panel
+│       ├── canvas/                 # Three.js spatial visualization
+│       ├── hooks/                  # WebSocket, state management hooks
+│       ├── store/                  # Frontend state
+│       └── types/                  # TypeScript type definitions
+├── objects/                        # Default POs: greeter, reader, coordinator
+├── templates/                      # Environment templates (basic, developer, writer, arc-agi-1, etc.)
+├── tools/                          # Development tooling
+└── test/                           # Unit and integration tests
 ```
 
-## Implementation Phases
-
-1. **Core Loop**: Capability base, PromptObject, Loader, single LLM adapter, simple REPL
-2. **Primitives & Binding**: Primitive base, built-in primitives, Registry
-3. **Multi-Capability**: Message bus with logging, PO↔PO communication
-4. **Self-Modification**: Universal capabilities including create_capability
-5. **Polish & UI**: Full Charm integration, all UI components
-6. **Demo Ready**: Error handling, testing, practice runs
-
 ## Key Concepts
 
 - **Semantic Binding**: Natural language → capability call (visible in message log)
-- **PO↔PO Communication**: Prompt-Objects can call each other as capabilities
-- **Self-Modification**: System can create new Prompt-Objects at runtime (with human approval)
-- **Human-in-the-Loop**: POs ask for confirmation on dangerous actions, clarification when ambiguous
-- **Non-blocking Human Queue**: When POs need human input (`ask_human`), they suspend (via Fibers) and queue a notification. Multiple POs can wait simultaneously. Human responds via notification panel, PO resumes automatically.
+- **PO↔PO Communication**: Prompt-Objects call each other as capabilities through isolated delegation threads
+- **Self-Modification**: POs can create new POs and primitives at runtime (with human approval)
+- **Human-in-the-Loop**: POs use `ask_human` to pause and queue notifications; human responds asynchronously via the web UI
+- **Environments**: Isolated collections of POs with their own sessions, git history, and configuration. Created from templates via CLI.
+- **Thread Export**: Conversation threads (including delegation chains) exportable as Markdown or JSON
+
+## Development
+
+```bash
+# Install dependencies
+bundle install
+cd frontend && npm install && cd ..
+
+# Run tests
+bundle exec rake test
+
+# Serve an environment with web UI
+prompt_objects serve <env-name> --open
+
+# Environment management
+prompt_objects env create <name> --template basic
+prompt_objects env list
+prompt_objects env info <name>
+```
+
+## Releases
+
+Always tag releases. Pushing a version tag triggers the Discord notification workflow (`.github/workflows/discord-release.yml`), which extracts the matching section from `CHANGELOG.md` and posts it.
+
+```bash
+# 1. Update version in prompt_objects.gemspec
+# 2. Add changelog entry under ## [X.Y.Z] - YYYY-MM-DD
+# 3. Commit: "Release vX.Y.Z — short description"
+# 4. Tag and push:
+git tag vX.Y.Z
+git push && git push origin vX.Y.Z
+# 5. Publish gem:
+gem build prompt_objects.gemspec && gem push prompt_objects-X.Y.Z.gem
+```

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    prompt_objects (0.3.1)
+    prompt_objects (0.4.0)
       anthropic (~> 1.0)
       async-websocket (~> 0.28)
       falcon (~> 0.50)
@@ -17,24 +17,25 @@ GEM
   specs:
     addressable (2.8.8)
       public_suffix (>= 2.0.2, < 8.0)
-    anthropic (1.16.3)
+    anthropic (1.19.0)
+      cgi
       connection_pool
     ast (2.4.3)
-    async (2.35.2)
+    async (2.36.0)
       console (~> 1.29)
       fiber-annotation
       io-event (~> 1.11)
       metrics (~> 0.12)
       traces (~> 0.18)
-    async-container (0.27.7)
+    async-container (0.30.0)
       async (~> 2.22)
-    async-container-supervisor (0.9.2)
+    async-container-supervisor (0.9.3)
       async-service
       io-endpoint
       memory (~> 0.7)
       memory-leak (~> 0.5)
       process-metrics
-    async-http (0.94.0)
+    async-http (0.94.2)
       async (>= 2.10.2)
       async-pool (~> 0.11)
       io-endpoint (~> 0.14)
@@ -49,9 +50,9 @@ GEM
       async-http (~> 0.56)
     async-pool (0.11.1)
       async (>= 2.0)
-    async-service (0.16.0)
+    async-service (0.19.0)
       async
-      async-container (~> 0.16)
+      async-container (~> 0.29)
       string-format (~> 0.2)
     async-websocket (0.30.0)
       async-http (~> 0.76)
@@ -63,6 +64,7 @@ GEM
       samovar (~> 2.1)
     base64 (0.3.0)
     bigdecimal (4.0.1)
+    cgi (0.5.1)
     connection_pool (3.0.2)
     console (1.34.2)
       fiber-annotation
@@ -74,20 +76,20 @@ GEM
       reline (>= 0.3.8)
     erb (6.0.1)
     event_stream_parser (1.0.0)
-    falcon (0.53.1)
+    falcon (0.54.2)
       async
       async-container (~> 0.20)
       async-container-supervisor (~> 0.6)
       async-http (~> 0.75)
       async-http-cache (~> 0.4)
-      async-service (~> 0.16)
+      async-service (~> 0.19)
       bundler
       localhost (~> 1.1)
       openssl (>= 3.0)
       protocol-http (~> 0.31)
       protocol-rack (~> 0.7)
       samovar (~> 2.3)
-    faraday (2.14.0)
+    faraday (2.14.1)
       faraday-net_http (>= 2.0, < 3.5)
       json
       logger
@@ -103,29 +105,29 @@ GEM
     fiber-storage (1.0.1)
     front_matter_parser (1.0.1)
     io-console (0.8.2)
-    io-endpoint (0.16.0)
+    io-endpoint (0.17.2)
     io-event (1.14.2)
     io-stream (0.11.1)
-    irb (1.16.0)
+    irb (1.17.0)
       pp (>= 0.6.0)
+      prism (>= 1.3.0)
       rdoc (>= 4.0.0)
       reline (>= 0.4.2)
-    json (2.18.0)
+    json (2.18.1)
     json-schema (6.1.0)
       addressable (~> 2.8)
       bigdecimal (>= 3.1, < 5)
-    json_rpc_handler (0.1.1)
     language_server-protocol (3.17.0.5)
     lint_roller (1.1.0)
-    listen (3.9.0)
+    listen (3.10.0)
+      logger
       rb-fsevent (~> 0.10, >= 0.10.3)
       rb-inotify (~> 0.9, >= 0.9.10)
-    localhost (1.6.0)
+    localhost (1.7.0)
     logger (1.7.0)
     mapping (1.1.3)
-    mcp (0.4.0)
+    mcp (0.6.0)
       json-schema (>= 4.1)
-      json_rpc_handler (~> 0.1)
     memory (0.12.0)
       bake (~> 0.15)
       console
@@ -139,25 +141,25 @@ GEM
       uri (>= 0.11.1)
     openssl (4.0.0)
     parallel (1.27.0)
-    parser (3.3.10.0)
+    parser (3.3.10.1)
       ast (~> 2.4.1)
       racc
     pp (0.6.3)
       prettyprint
     prettyprint (0.2.0)
-    prism (1.7.0)
+    prism (1.9.0)
     process-metrics (0.8.0)
       console (~> 1.8)
       json (~> 2)
       samovar (~> 2.1)
     protocol-hpack (1.5.1)
-    protocol-http (0.58.0)
-    protocol-http1 (0.36.0)
+    protocol-http (0.59.0)
+    protocol-http1 (0.37.0)
       protocol-http (~> 0.58)
-    protocol-http2 (0.23.0)
+    protocol-http2 (0.24.0)
       protocol-hpack (~> 1.4)
       protocol-http (~> 0.47)
-    protocol-rack (0.21.0)
+    protocol-rack (0.21.1)
       io-stream (>= 0.10)
       protocol-http (~> 0.58)
       rack (>= 1.0)
@@ -167,7 +169,7 @@ GEM
     psych (5.3.1)
       date
       stringio
-    public_suffix (7.0.1)
+    public_suffix (7.0.2)
     racc (1.8.1)
     rack (3.2.4)
     rainbow (3.1.1)
@@ -175,14 +177,14 @@ GEM
     rb-fsevent (0.11.2)
     rb-inotify (0.11.1)
       ffi (~> 1.0)
-    rdoc (7.0.3)
+    rdoc (7.2.0)
       erb
       psych (>= 4.0.0)
       tsort
     regexp_parser (2.11.3)
     reline (0.6.3)
       io-console (~> 0.5)
-    rubocop (1.82.1)
+    rubocop (1.84.1)
       json (~> 2.3)
       language_server-protocol (~> 3.17.0.2)
       lint_roller (~> 1.1.0)
@@ -190,7 +192,7 @@ GEM
       parser (>= 3.3.0.2)
       rainbow (>= 2.2.2, < 4.0)
       regexp_parser (>= 2.9.3, < 3.0)
-      rubocop-ast (>= 1.48.0, < 2.0)
+      rubocop-ast (>= 1.49.0, < 2.0)
       ruby-progressbar (~> 1.7)
       unicode-display_width (>= 2.4.0, < 4.0)
     rubocop-ast (1.49.0)

data/README.md

@@ -12,6 +12,11 @@ Prompt Objects applies this to AI. Instead of treating LLMs as external APIs you
 
 This is a new computing primitive: semantic late binding at runtime, where natural language becomes the interface between intelligent components.
 
+Blog Posts
+==========
+
+- [What if we took message passing seriously?](https://worksonmymachine.ai/p/what-if-we-took-message-passing-seriously) — The origin story and motivation behind Prompt Objects.
+
 Who
 ===
 

data/frontend/index.html

@@ -4,9 +4,13 @@
     <meta charset="UTF-8" />
     <link rel="icon" type="image/svg+xml" href="/vite.svg" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <link rel="preconnect" href="https://cdn.jsdelivr.net" crossorigin />
+    <link href="https://cdn.jsdelivr.net/fontsource/fonts/geist@latest/latin-400-normal.woff2" as="font" type="font/woff2" crossorigin />
+    <link href="https://cdn.jsdelivr.net/npm/geist@1/dist/fonts/geist-sans/style.css" rel="stylesheet" />
+    <link href="https://cdn.jsdelivr.net/npm/geist@1/dist/fonts/geist-mono/style.css" rel="stylesheet" />
     <title>PromptObjects</title>
   </head>
-  <body class="bg-po-bg text-gray-100">
+  <body class="bg-po-bg text-po-text-primary font-ui">
     <div id="root"></div>
     <script type="module" src="/src/main.tsx"></script>
   </body>