claude_agent 0.7.8 → 0.7.9

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e9304128d18cec59aa0b0221f199b2674d60a5a2813141562b9df51b88057987
-  data.tar.gz: 8bd6696b854f8660e21744bc02d178e3e402a734119554c2869cc4fbb188a75d
+  metadata.gz: 90d9094d4e4428fc7dfc1e452759403a92f2e4badf5d1ac5f2f90e950c511255
+  data.tar.gz: fd2a8407c794aa672a4b03a774a6573dac3682bc09b8fe666170b90f5bd9f79a
 SHA512:
-  metadata.gz: 8ad2321250e0967d302d41d23724fe6f368f3e8fe6cb4461f483f6d13d3b9b35368189cce06dd28f5fa427013a9a272c6a4e1b3e94ee75b8a38dd7ed422c5068
-  data.tar.gz: d17257cfcabecbb033ce1427a0237456e5ecbeb0c1c3bedb9bc21dac1336be6a21b96a85ffc96919f7c4774a3f51498a4876f31d11249a05f327966af23f0ae8
+  metadata.gz: ce7d705ebe01a38ad1ac86a8caeaabd700831e200e73f134c9adef507b6f7b695effece63ae902849ee408714c3ae576930012dbbefe2b05da403d0de8549317
+  data.tar.gz: 7f06c5b0b5ed5ad798442672ad8ca969a0fdefc414a4c273e86911df0e171d24d5a378b9a88117bd3c60092a6affb5219d06cc58dd0107245643b571e2236440

data/CHANGELOG.md

@@ -7,6 +7,63 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.7.9] - 2026-02-25
+
+### Added
+- `Conversation` — high-level wrapper managing the full conversation lifecycle with auto-connect, multi-turn history, callbacks, and tool activity timeline
+- `Conversation#say(prompt)` — send a message and receive a `TurnResult`, auto-connecting on first call
+- `Conversation.open { |c| ... }` — block form with automatic cleanup
+- `Conversation.resume(session_id)` — resume a previous conversation
+- `Conversation` callbacks: `on_text`, `on_stream`, `on_tool_use`, `on_tool_result`, `on_thinking`, `on_result`, `on_message`, `on_permission`
+- `Conversation#total_cost`, `#session_id`, `#usage`, `#pending_permission`, `#pending_permissions?` — convenience accessors
+- `ClaudeAgent.conversation(**kwargs)` and `ClaudeAgent.resume_conversation(session_id)` — module-level convenience methods
+- `ToolActivity` — immutable `Data.define` pairing a `ToolUseBlock` with its `ToolResultBlock`, turn index, and timing; delegates `name`, `display_label`, `summary`, `file_path`, `id` to the tool use block
+- `Conversation#tool_activity` — unified timeline of all tool executions across turns with duration tracking
+- `PermissionRequest` — deferred permission promise that can be resolved from any thread via `allow!` / `deny!`, with `defer!` for hybrid callback/queue mode
+- `PermissionQueue` — thread-safe queue of pending `PermissionRequest` objects with `poll`, `pop(timeout:)`, and `drain!`
+- `Client#permission_queue` — returns the queue; always available on connected clients
+- `Client#pending_permission` — non-blocking poll for the next pending permission request
+- `Client#pending_permissions?` — check if any permission requests are waiting
+- `Options#permission_queue` — set to `true` to enable queue-based permissions (auto-sets `permission_prompt_tool_name: "stdio"`)
+- `ToolPermissionContext#request` — access the `PermissionRequest` from within a `can_use_tool` callback for hybrid defer mode
+- `ControlProtocol` now supports three permission modes: synchronous callback, queue-based, and hybrid (callback with selective `defer!`)
+- `TurnResult` — represents a complete agent turn, accumulating all messages between sending a prompt and receiving the `ResultMessage`
+- `TurnResult#text`, `#thinking` — concatenated text/thinking across all assistant messages
+- `TurnResult#tool_uses`, `#tool_results`, `#tool_executions` — tool use blocks, result blocks, and matched use/result pairs
+- `TurnResult#usage`, `#cost`, `#duration_ms`, `#session_id`, `#model`, `#success?`, `#error?` — convenient accessors from the result
+- `Client#send_and_receive(content)` — send a message and receive the complete `TurnResult` in one call
+- `Client#receive_turn` — like `receive_response` but returns a `TurnResult` instead of yielding raw messages
+- `ClaudeAgent.query_turn(prompt:)` — one-shot query returning a `TurnResult`
+- `EventHandler` — register typed event callbacks (`:text`, `:thinking`, `:tool_use`, `:tool_result`, `:result`, `:message`) instead of writing `case` statements over raw messages
+- `Client#on(event, &block)` and `#on_text`, `#on_tool_use`, `#on_tool_result`, `#on_result`, etc. — register persistent event handlers that fire during `receive_turn` and `send_and_receive`
+- `ClaudeAgent.query_turn` now accepts `events:` parameter for standalone `EventHandler` use
+- `TaskProgressMessage` for real-time background task (subagent) progress reporting (TypeScript SDK v0.2.51 parity)
+- `mcp_authenticate` control request on `ControlProtocol` and `Client` for MCP server OAuth authentication (TypeScript SDK v0.2.52 parity)
+- `mcp_clear_auth` control request on `ControlProtocol` and `Client` for clearing MCP server credentials (TypeScript SDK v0.2.52 parity)
+- RBS signatures for all new types and methods
+- `CumulativeUsage` — built-in cumulative usage tracking across conversation turns; automatically updated by `Client` as `ResultMessage`s are received
+- `Client#cumulative_usage` — returns the usage accumulator with `input_tokens`, `output_tokens`, `cache_read_input_tokens`, `cache_creation_input_tokens`, `total_cost_usd`, `num_turns`, `duration_ms`, `duration_api_ms`
+- `GenericMessage` type — wraps unknown top-level message types instead of raising `MessageParseError`, with dynamic field access via `[]` and `method_missing`
+- `GenericBlock` type — wraps unknown content block types instead of returning raw Hashes, with dynamic field access via `[]` and `method_missing`
+- `ToolUseBlock#file_path` — returns the file path for file-based tools (`Read`, `Write`, `Edit`, `NotebookEdit`), nil otherwise
+- `ToolUseBlock#display_label` — one-line human-readable label (e.g. `"Read lib/foo.rb"`, `"Bash: git status"`, `"WebFetch: example.com"`)
+- `ToolUseBlock#summary(max:)` — detailed summary with truncation (e.g. `"Write: /path.rb (3 lines)"`, `"Grep: pattern in /path (*.rb)"`)
+- `ServerToolUseBlock#file_path`, `#display_label`, `#summary(max:)` — same interface with server context (e.g. `"server_name/tool_name"`)
+
+### Changed
+- `BaseHookInput.define_input` — declarative macro for generating hook input subclasses from a single declaration; replaces 18 hand-written classes with one-line definitions
+- `MessageParser` now uses a registry pattern (`MessageParser.register`) instead of nested `case` statements for message routing; adding a new message type is one `register` call
+- `MessageParser#parse` no longer raises `MessageParseError` for unknown message types; returns `GenericMessage` instead
+- `MessageParser#parse_content_block` no longer returns raw Hashes for unknown content block types; returns `GenericBlock` instead
+- All parsed message hashes now use snake_case symbol keys throughout (e.g. `msg.usage[:input_tokens]`, `event.files.first[:filename]`, `event.event[:type]`)
+- `MessageParser#parse` applies a single `deep_transform_keys` pass that normalizes camelCase→snake_case and string→symbol for the entire hash tree
+- `can_use_tool` callbacks receive symbol-keyed `input` hashes
+- Hook callbacks receive symbol-keyed `input` hashes
+- MCP tool handlers receive symbol-keyed `arguments` hashes
+
+### Removed
+- `MessageParser#fetch_dual` — no longer needed now that keys are normalized at the entry point
+
 ## [0.7.8] - 2026-02-22
 
 ### Added