prompt_objects 0.2.0 → 0.3.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 0fe57111f23be7238465e5a386d48d8d694cd4599e2113e6378150951efcd303
-  data.tar.gz: '07192c4f47af2d7aa9530fdc61b65cb97fba77ae84133a30533d3f07314e0b79'
+  metadata.gz: 0a1ec32b66d34c2fe11c964759513487335794f9a8c336bd32f7ae9aeed5380c
+  data.tar.gz: 9a6650251228ab4dae25f16120e8ae8c1992e4458faf0c858894b6e4848ae80b
 SHA512:
-  metadata.gz: 66d2e26018f49685968bccfd075b2e7cf74fb4bd0cbdf0730f4f3c68c3101be232235beb86a8a1cde73bdf2707c1a679b916d53c8b4c1fb2f9d455d965e4d246
-  data.tar.gz: 5fb3e4b088aaf7a6a99e08e144f090c68e2a63b0b009fe43929d93daa8da7b9e6b0c330ae048c2d6bda9fbcdef385000c0f255695a539e922393fc3c84a59e84
+  metadata.gz: 867fc4cb49114ace4af4682feff5a062eb414f0ce84e4be4e48632609d68cec2ac201be2cb43fb995ac5e9cb4ec0480399aa6eaad7580da81fd5451905b0ddf5
+  data.tar.gz: 4b3d3c3d96d45a2573829345fd1f822fa3f8918c9d4088cccc46d82ba4fa797bb7c9838e74d443017069f7cd9b9d343d3bd92219930b12098f036d038e05ef83

data/CHANGELOG.md

@@ -0,0 +1,80 @@
+# Changelog
+
+All notable changes to PromptObjects are documented in this file.
+
+## [0.3.1] - 2025-02-08
+
+### Added
+
+- **Thread Explorer** — Standalone HTML visualizer for exploring conversation thread exports. Three views: sequence diagram (swim lanes showing PO communication), timeline (flat chronological event list), and detail panel (full conversation with tool calls). Includes token cost bar, search, per-PO filtering, breadcrumb navigation, and structural event highlighting.
+- **`explore` CLI command** — Open Thread Explorer from the command line. `prompt_objects explore <env>` lists root threads and opens the most recent; `--session ID` targets a specific thread. Data is embedded directly so it opens ready to go.
+
+### Fixed
+
+- Delegation messages now correctly show the calling PO's name instead of "human". Previously `context.current_capability` resolved to the target PO (matching its own name), causing a fallback to "human". Now uses `context.calling_po`.
+- CLI integration tests skip gracefully in CI when no LLM API key is available, instead of failing.
+
+## [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 (0.3.1)
       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