claude_agent 0.6.0 → 0.7.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4d11f6fa8fa6ecc4b00e0bb6f7913582d0ac8675f649f2d5b423306660a44247
-  data.tar.gz: 1c7259588ea2b823ad06c7f263830418a590b5488c7db3d9d825b3e46b31495a
+  metadata.gz: 72db3aeea7a724d0271a3fc8bb0bd7900425a34e7114e526e9bfe8bbbe26e19f
+  data.tar.gz: a31e9321dc56507d0a05d8319e50fded02aaa420a9709a0970a298826ea2f4ad
 SHA512:
-  metadata.gz: e57643b8293c0cba79fb96322daa1c8fd08a068664f63bad5a6787f9731e60232ff0ee806698ff79707c4a000dd600bfc1e459ea9a0933e7685e5b5fefd87911
-  data.tar.gz: 6ac435262b2489ef696db91db6440e8df9f8000e09a46b97cc80decdda1b04cedfcd561a6d56e8b09aa8d373fa223f1ecdad2ff08f82ba9ea350375dd5772eed
+  metadata.gz: 54d586a6a4e368adf4b64ec8afdbf4b04ad165473e81bcf8e0c244c7c04ac5c524b6c1186e34dd7bf6c31cc3b521da6d5526b187a30d147c3839ba5b708056ae
+  data.tar.gz: 4af34d4756c9559570ea9c8d0ed9819ae98827773eed4332d9cc7bc50c8b5edc6584c0ee25fbfa197c3bc79513acace10ee88ec3bf1674141afcfe213ead2f02

data/.claude/skills/spec/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: spec
+description: Manage the SDK feature parity specification. Use 'update' to regenerate SPEC.md from current implementations, or 'complete' to implement missing features.
+argument-hint: [update|complete]
+disable-model-invocation: true
+allowed-tools: Bash, Read, Glob, Grep, Write, Edit, TodoWrite, Task, AskUserQuestion, EnterPlanMode
+---
+
+# SDK Feature Parity Management
+
+Manage the feature parity specification for the TypeScript, Python, and Ruby Claude Agent SDKs.
+
+## Reference SDKs
+
+Ensure reference SDKs are current before starting any workflow:
+
+```bash
+bin/update-reference-sdks
+```
+
+This clones/updates:
+- Python SDK from GitHub -> `vendor/claude-agent-sdk-python`
+- TypeScript SDK from npm -> `vendor/claude-agent-sdk-npm`
+- TypeScript SDK repo from GitHub -> `vendor/claude-agent-sdk-typescript` (changelog/releases)
+
+## Key Source Files
+
+**TypeScript SDK (Primary Reference)**
+- `vendor/claude-agent-sdk-npm/sdk.d.ts` - Complete API surface with all types
+- `vendor/claude-agent-sdk-typescript/CHANGELOG.md` - Release changelog
+
+**Python SDK**
+- `vendor/claude-agent-sdk-python/src/claude_agent_sdk/types.py` - Type definitions
+- `vendor/claude-agent-sdk-python/src/claude_agent_sdk/_internal/` - Internal implementation
+
+**Ruby SDK (This Repository)**
+- `lib/claude_agent/` - All implementation files
+
+## Mode Selection
+
+Based on the first argument:
+
+- **update**: Follow the workflow in [commands/update.md](commands/update.md)
+- **complete**: Follow the workflow in [commands/complete.md](commands/complete.md)
+- If no argument or unrecognized: ask the user which mode they want

data/.claude/{commands/spec → skills/spec/commands}/complete.md

@@ -1,21 +1,12 @@
----
-description: Implement missing Ruby SDK features identified in SPEC.md to achieve full parity with TypeScript/Python SDKs
-allowed-tools: Bash, Read, Glob, Grep, Write, Edit, TodoWrite, Task, AskUserQuestion, EnterPlanMode
----
-
 # Complete SDK Feature Parity
 
-You are implementing missing features in the Ruby SDK to achieve full parity with the official TypeScript and Python Claude Agent SDKs.
+Implement missing features in the Ruby SDK to achieve full parity with the official TypeScript and Python Claude Agent SDKs.
 
 ## Step 1: Identify Missing Features
 
-Read SPEC.md and identify all features where the Ruby SDK shows ❌ (not implemented):
-
-```
-Read SPEC.md
-```
+Read SPEC.md and identify all features where the Ruby SDK is not implemented.
 
-Create a prioritized list of missing features:
+Create a prioritized list:
 1. **High Priority** - Features implemented in BOTH TypeScript and Python SDKs
 2. **Medium Priority** - Features only in TypeScript SDK (Python also missing)
 3. **Low Priority** - TypeScript-only features that may be JS-specific
@@ -24,31 +15,23 @@ Use TodoWrite to track the features to implement.
 
 ## Step 2: Research Each Feature
 
-For each missing feature, gather complete context before planning:
+Gather complete context before planning:
 
 ### 2a. Official Documentation
 
-Use the `claude-code-guide` agent to understand the feature's intended behavior:
+Use the `claude-code-guide` agent to understand each feature's intended behavior:
 
 ```
 Task(subagent_type: "claude-code-guide", prompt: "How does [feature] work in the Claude Agent SDK? What are all the options, behaviors, and edge cases?")
 ```
 
-### 2b. TypeScript Implementation
-
-Read the TypeScript type definitions to understand the API surface:
-- `vendor/claude-agent-sdk-npm/sdk.d.ts`
-
-### 2c. Python Implementation (if available)
+### 2b. Reference Implementations
 
-If Python has the feature, read their implementation for patterns:
-- `vendor/claude-agent-sdk-python/src/claude_agent_sdk/types.py`
-- `vendor/claude-agent-sdk-python/src/claude_agent_sdk/_internal/`
+Read the TypeScript type definitions and changelog listed in SKILL.md to understand the API surface and when features were added. If Python has the feature, read their implementation for patterns.
 
 ## Step 3: Clarify Requirements
 
-Use AskUserQuestion to resolve any ambiguities:
-
+Use AskUserQuestion to resolve ambiguities:
 - Implementation approach choices
 - Ruby-specific design decisions
 - Whether certain features should be skipped (e.g., JS-specific)
@@ -56,15 +39,9 @@ Use AskUserQuestion to resolve any ambiguities:
 
 ## Step 4: Enter Plan Mode
 
-Enter plan mode to design the implementation:
-
-```
-EnterPlanMode
-```
-
-The plan should include:
+Enter plan mode to design the implementation. The plan should include:
 - Which files to create or modify
-- Data structures (use `Data.define` for immutable types)
+- Data structures
 - Public API design
 - Test coverage requirements
 - Any breaking changes or deprecations
@@ -80,26 +57,22 @@ After plan approval, implement each feature:
 5. Update RBS type signatures in `sig/` directory
 6. Write unit tests in `test/claude_agent/`
 7. Write integration tests in `test/integration/` for features that interact with CLI/API
-8. Update SPEC.md to mark feature as ✅
+8. Update SPEC.md to mark feature as complete
 
 ### Testing Requirements
 
 - **Unit tests** - For all new types, data structures, and internal logic
-- **Integration tests** - Required for anything that:
-  - Spawns the CLI subprocess
-  - Sends/receives messages via the control protocol
-  - Interacts with MCP servers
-  - Uses file checkpointing or session management
+- **Integration tests** - Required for anything that spawns the CLI subprocess, sends/receives messages via the control protocol, interacts with MCP servers, or uses file checkpointing/session management
 
 ## Step 6: Update Specification
 
-After implementing features, run `/spec:update` to refresh SPEC.md with the new implementation status.
+After implementing features, run `/spec update` to refresh SPEC.md with the new implementation status.
 
 ## Output
 
 Provide a summary of:
 - Features implemented
-- Any features skipped (with reasons)
+- Features skipped (with reasons)
 - Breaking changes introduced
 - Test coverage added
 - Remaining gaps (if any)

data/.claude/skills/spec/commands/update.md

@@ -0,0 +1,48 @@
+# Update SDK Specification
+
+Update SPEC.md to reflect the current feature parity between the TypeScript, Python, and Ruby Claude Agent SDKs.
+
+## Step 1: Update Reference SDKs
+
+Run `bin/update-reference-sdks` to ensure all reference SDKs are current.
+
+## Step 2: Research Official Documentation
+
+Use the `claude-code-guide` agent to check for new or updated SDK features:
+
+```
+Task(subagent_type: "claude-code-guide", prompt: "What are all the features and options in the Claude Agent SDK? Include configuration options, hooks, permissions, MCP support, and control protocol methods.")
+```
+
+This helps catch features documented but not yet in SDK source files.
+
+## Step 3: Analyze All Three SDKs
+
+Read and compare the key source files listed in SKILL.md. Pay special attention to:
+- TypeScript `sdk.d.ts` as the authoritative API surface
+- TypeScript `CHANGELOG.md` for recently added features
+- Python `types.py` for implementation patterns
+- All Ruby SDK files for current implementation status
+
+## Step 4: Update SPEC.md
+
+1. **Reference Versions** - Update TypeScript SDK version and Python SDK commit
+2. **Feature Tables** - Update all markers based on current implementations
+3. **New Features** - Add any new features found in the TypeScript SDK
+4. **Removed Features** - Remove any deprecated features
+
+## Guidelines
+
+- Check every field and option in each SDK
+- TypeScript sdk.d.ts is the authoritative reference
+- Preserve the existing table structure and markdown formatting
+- Always update the reference version info at the top
+- Note significant changes after updating
+
+## Output
+
+Provide a brief summary of:
+- SDK versions checked
+- New features added to the spec
+- Features removed from the spec
+- Notable gaps in the Ruby SDK

data/CHANGELOG.md

@@ -7,6 +7,17 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.7.0] - 2026-01-31
+
+### Added
+- MCP tool annotations support (`readOnlyHint`, `destructiveHint`, `idempotentHint`, `openWorldHint`, `title`) on `MCP::Tool` and `MCP.tool` convenience method (TypeScript SDK v0.2.27 parity)
+- README documentation for `UserMessageReplay`, `HookStartedMessage`, `HookProgressMessage`, `ToolUseSummaryMessage`, `FilesPersistedEvent` message types
+- README documentation for `mcp_reconnect` and `mcp_toggle` client methods
+- README documentation for MCP tool annotations
+
+### Changed
+- Updated SPEC.md to reference TypeScript SDK v0.2.27 and Python SDK v0.1.26
+
 ## [0.6.0] - 2026-01-30
 
 ### Added

data/CLAUDE.md

@@ -1,6 +1,6 @@
 # ClaudeAgent Ruby SDK
 
-Ruby SDK for building autonomous AI agents that interact with Claude Code CLI.
+Ruby SDK for building autonomous AI agents that interact with Claude Code CLI. See [README.md](README.md) for API usage, message types, configuration, and examples.
 
 ## Stack
 
@@ -32,16 +32,6 @@ bin/rbs-validate                   # Validate RBS signatures
 bin/release VERSION                # Release gem (e.g., bin/release 1.2.0)
 ```
 
-## Architecture
-
-| Component                   | Purpose                                            |
-|-----------------------------|----------------------------------------------------|
-| `Query`                     | One-shot stateless prompts                         |
-| `Client`                    | Multi-turn bidirectional conversations             |
-| `ControlProtocol`           | Handles handshake, hooks, permissions, MCP routing |
-| `Transport::Subprocess`     | Spawns CLI, manages stdin/stdout                   |
-| `MCP::Tool` / `MCP::Server` | Custom tool definitions                            |
-
 ## Conventions
 
 - **Immutable data types**: All messages and options use `Data.define`
@@ -50,25 +40,6 @@ bin/release VERSION                # Release gem (e.g., bin/release 1.2.0)
 - **Error hierarchy**: All errors inherit from `ClaudeAgent::Error` with context (exit code, stderr, etc.)
 - **Protocol flow**: Transport → ControlProtocol → MessageParser → typed message objects
 
-## Key Patterns
-
-```ruby
-# One-shot query
-result = ClaudeAgent.query("prompt", model: "sonnet")
-
-# Interactive client
-client = ClaudeAgent::Client.new(options)
-client.send_message("prompt") { |msg| handle(msg) }
-
-# Content blocks are polymorphic
-message.content.each do |block|
-  case block
-  when ClaudeAgent::TextBlock then ...
-  when ClaudeAgent::ToolUseBlock then ...
-  end
-end
-```
-
 ## Testing Notes
 
 - Unit tests in `test/claude_agent/` - run without Claude CLI

data/README.md

@@ -5,7 +5,7 @@ Ruby gem for building AI-powered applications with the [Claude Agent SDK](https:
 ## Requirements
 
 - Ruby 3.2+ (uses `Data.define`)
-- [Claude Code CLI](https://docs.anthropic.com/en/docs/claude-code/getting-started) v2.0.0+
+- [Claude Code CLI](https://code.claude.com/docs/en/getting-started) v2.0.0+
 
 ## Installation
 
@@ -226,6 +226,19 @@ result.errors          # Array of error messages (if any)
 result.permission_denials  # Array of SDKPermissionDenial (if any)
 ```
 
+### UserMessageReplay
+
+Replayed user messages when resuming a session with existing history:
+
+```ruby
+replay.content             # Message content
+replay.uuid                # Message UUID
+replay.session_id          # Session identifier
+replay.parent_tool_use_id  # Parent tool use ID (if tool result)
+replay.replay?             # true if this is a replayed message
+replay.synthetic?          # true if this is a synthetic message
+```
+
 ### SystemMessage
 
 Internal system events:
@@ -289,6 +302,48 @@ hook_response.stderr     # Hook stderr
 hook_response.exit_code  # Exit code
 ```
 
+### HookStartedMessage
+
+Hook execution start notification:
+
+```ruby
+hook_started.hook_id    # Hook identifier
+hook_started.hook_name  # Hook name
+hook_started.hook_event # Hook event type
+```
+
+### HookProgressMessage
+
+Progress during hook execution:
+
+```ruby
+hook_progress.hook_id    # Hook identifier
+hook_progress.hook_name  # Hook name
+hook_progress.hook_event # Hook event type
+hook_progress.stdout     # Hook stdout so far
+hook_progress.stderr     # Hook stderr so far
+hook_progress.output     # Combined output
+```
+
+### ToolUseSummaryMessage
+
+Summary of tool use for collapsed display:
+
+```ruby
+summary.summary                  # Human-readable summary text
+summary.preceding_tool_use_ids   # Tool use IDs this summarizes
+```
+
+### FilesPersistedEvent
+
+Files persisted to storage during a session:
+
+```ruby
+persisted.files        # Array of successfully persisted file paths
+persisted.failed       # Array of files that failed to persist
+persisted.processed_at # Timestamp of persistence
+```
+
 ### AuthStatusMessage
 
 Authentication status during login:
@@ -413,6 +468,33 @@ schema: {
 }
 ```
 
+### Tool Annotations
+
+Annotate tools with hints about their behavior:
+
+```ruby
+tool = ClaudeAgent::MCP::Tool.new(
+  name: "search",
+  description: "Search the web",
+  schema: { query: String },
+  annotations: {
+    readOnlyHint: true,       # Tool only reads data, no side effects
+    destructiveHint: false,   # Tool does not destroy/delete data
+    idempotentHint: true,     # Repeated calls with same args have same effect
+    openWorldHint: true,      # Tool interacts with external systems
+    title: "Web Search"       # Human-readable display name
+  }
+) { |args| "Results for #{args['query']}" }
+
+# Or with the convenience method
+tool = ClaudeAgent::MCP.tool(
+  "search", "Search the web", { query: String },
+  annotations: { readOnlyHint: true, openWorldHint: true }
+) { |args| "Results" }
+```
+
+All annotation fields are optional hints — omit any that don't apply.
+
 ### Tool Return Values
 
 Tools can return various formats:
@@ -622,6 +704,13 @@ result = client.set_mcp_servers({
 })
 puts "Added: #{result.added}, Removed: #{result.removed}"
 
+# Reconnect a disconnected MCP server
+client.mcp_reconnect("my-server")
+
+# Enable or disable an MCP server
+client.mcp_toggle("my-server", enabled: false)  # Disable
+client.mcp_toggle("my-server", enabled: true)   # Re-enable
+
 # Query capabilities
 client.supported_commands.each { |cmd| puts "#{cmd.name}: #{cmd.description}" }
 client.supported_models.each { |model| puts "#{model.value}: #{model.display_name}" }

data/SPEC.md

@@ -3,11 +3,11 @@
 This document provides a comprehensive specification of the Claude Agent SDK, comparing feature parity across the official TypeScript and Python SDKs with this Ruby implementation.
 
 **Reference Versions:**
-- TypeScript SDK: v0.2.25 (npm package)
-- Python SDK: v0.1.25 from GitHub (commit 1bf665c)
+- TypeScript SDK: v0.2.27 (npm package)
+- Python SDK: v0.1.26 from GitHub (commit 438ddf7)
 - Ruby SDK: This repository
 
-**Last Updated:** 2026-01-30
+**Last Updated:** 2026-01-31
 
 ---
 
@@ -424,7 +424,7 @@ Permission handling and updates.
 |------------------|:----------:|:------:|:----:|---------------------------|
 | `signal`         |     ✅      |   ✅    |  ✅   | Abort signal              |
 | `suggestions`    |     ✅      |   ✅    |  ✅   | Permission suggestions    |
-| `blockedPath`    |     ✅      |   ✅    |  ✅   | Blocked file path         |
+| `blockedPath`    |     ✅      |   ❌    |  ✅   | Blocked file path         |
 | `decisionReason` |     ✅      |   ❌    |  ✅   | Why permission triggered  |
 | `toolUseID`      |     ✅      |   ❌    |  ✅   | Tool call ID              |
 | `agentID`        |     ✅      |   ❌    |  ✅   | Subagent ID if applicable |
@@ -474,11 +474,12 @@ Model Context Protocol server support.
 
 ### SDK MCP Server
 
-| Feature              | TypeScript | Python |         Ruby         | Notes                  |
-|----------------------|:----------:|:------:|:--------------------:|------------------------|
-| `createSdkMcpServer` |     ✅      |   ❌    |          ✅           | Create SDK server      |
-| `tool()` helper      |     ✅      |   ❌    |          ✅           | Create tool definition |
-| Tool input schema    |  ✅ (Zod)   |   ❌    | ✅ (Hash/JSON Schema) | Schema definition      |
+| Feature              | TypeScript | Python |         Ruby         | Notes                    |
+|----------------------|:----------:|:------:|:--------------------:|--------------------------|
+| `createSdkMcpServer` |     ✅      |   ❌    |          ✅           | Create SDK server        |
+| `tool()` helper      |     ✅      |   ❌    |          ✅           | Create tool definition   |
+| Tool input schema    |  ✅ (Zod)   |   ❌    | ✅ (Hash/JSON Schema) | Schema definition        |
+| Tool annotations     |     ✅      |   ❌    |          ✅           | MCP tool hints (v0.2.27) |
 
 ---
 
@@ -611,6 +612,8 @@ Public API surface for SDK clients.
 | `accountInfo()`          |     ✅      |   ❌    |  ✅   | Get account info       |
 | `rewindFiles()`          |     ✅      |   ✅    |  ✅   | Rewind file changes    |
 | `setMcpServers()`        |     ✅      |   ❌    |  ✅   | Dynamic MCP servers    |
+| `reconnectMcpServer()`   |     ✅      |   ❌    |  ✅   | Reconnect MCP server   |
+| `toggleMcpServer()`      |     ✅      |   ❌    |  ✅   | Enable/disable MCP     |
 | `streamInput()`          |     ✅      |   ✅    |  ✅   | Stream user input      |
 | `close()`                |     ✅      |   ✅    |  ✅   | Close query/session    |
 
@@ -663,9 +666,12 @@ Public API surface for SDK clients.
 - v0.2.19 adds `HookStartedMessage`, `HookProgressMessage`, and `ToolUseSummaryMessage`
 - v0.2.25 adds `SDKFilesPersistedEvent` message type for file persistence confirmation
 - v0.2.25 adds `McpClaudeAIProxyServerConfig` (`claudeai-proxy` type) for managed proxy servers
+- v0.2.21 adds `reconnectMcpServer()` and `toggleMcpServer()` query methods
+- v0.2.21 adds `config`, `scope`, `tools` fields and `disabled` status to `McpServerStatus`
+- v0.2.27 adds optional `annotations` support to `tool()` helper for MCP tool hints
 
 ### Python SDK
-- Full source available (v0.1.25)
+- Full source available (v0.1.26)
 - Now has `Transport` abstract class and several query control methods
 - Supports `interrupt()`, `set_permission_mode()`, `set_model()`, `rewind_files()`, `stream_input()`, `close()`, `get_mcp_status()` in query
 - Client also supports `interrupt()`, `set_permission_mode()`, `set_model()`, `rewind_files()`
@@ -676,6 +682,7 @@ Public API surface for SDK clients.
 - `excludedCommands` in sandbox now supported
 - v0.1.25 adds `PostToolUseFailure` hook event support
 - v0.1.25 adds `permissionDecisionReason` to `PreToolUseHookSpecificOutput`
+- v0.1.26 bumps bundled CLI to v2.1.27 (no new SDK features beyond PostToolUseFailure)
 - `additionalContext` supported in `UserPromptSubmitHookSpecificOutput`
 - `PreToolUseHookInput` does not include `tool_use_id` (unlike TypeScript)
 - `ToolPermissionContext` missing `blockedPath`, `decisionReason`, `toolUseID`, `agentID`
@@ -690,3 +697,4 @@ Public API surface for SDK clients.
 - `executable`/`executableArgs` marked N/A (JS runtime options not applicable to Ruby)
 - Added: `FilesPersistedEvent` message type (new in TS v0.2.25)
 - `claudeai-proxy` MCP server type handled transparently via Hash-based config passthrough (new in TS v0.2.25)
+- Added: MCP tool `annotations` support (new in TS v0.2.27)

data/lib/claude_agent/mcp/server.rb

@@ -150,8 +150,8 @@ module ClaudeAgent
     #     "Hello, #{args['name']}!"
     #   end
     #
-    def self.tool(name, description, schema = {}, &handler)
-      Tool.new(name: name, description: description, schema: schema, &handler)
+    def self.tool(name, description, schema = {}, annotations: nil, &handler)
+      Tool.new(name: name, description: description, schema: schema, annotations: annotations, &handler)
     end
 
     # Convenience method to create a server

data/lib/claude_agent/mcp/tool.rb

@@ -34,16 +34,18 @@ module ClaudeAgent
     #   end
     #
     class Tool
-      attr_reader :name, :description, :schema, :handler
+      attr_reader :name, :description, :schema, :annotations, :handler
 
       # @param name [String] Tool name (must be unique within server)
       # @param description [String] Description of what the tool does
       # @param schema [Hash] Input schema (simple Ruby types or JSON Schema)
+      # @param annotations [Hash, nil] MCP tool annotations (readOnlyHint, destructiveHint, etc.)
       # @param handler [Proc] Block to execute when tool is called
-      def initialize(name:, description:, schema: {}, &handler)
+      def initialize(name:, description:, schema: {}, annotations: nil, &handler)
         @name = name.to_s
         @description = description.to_s
         @schema = normalize_schema(schema)
+        @annotations = annotations
         @handler = handler || ->(args) { raise "No handler defined for tool #{name}" }
       end
 
@@ -60,11 +62,13 @@ module ClaudeAgent
       # Convert to MCP tool definition format
       # @return [Hash]
       def to_mcp_definition
-        {
+        definition = {
           name: @name,
           description: @description,
           inputSchema: @schema
         }
+        definition[:annotations] = @annotations if @annotations.present?
+        definition
       end
 
       private

data/lib/claude_agent/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module ClaudeAgent
-  VERSION = "0.6.0"
+  VERSION = "0.7.0"
 end

data/sig/claude_agent.rbs

@@ -967,15 +967,16 @@ module ClaudeAgent
 
   # MCP module
   module MCP
-    def self.tool: (String name, String description, ?Hash[Symbol, untyped] schema) { (Hash[String, untyped]) -> untyped } -> Tool
+    def self.tool: (String name, String description, ?Hash[Symbol, untyped] schema, ?annotations: Hash[Symbol, untyped]?) { (Hash[String, untyped]) -> untyped } -> Tool
     def self.create_server: (name: String, ?tools: Array[Tool]) -> Server
 
     class Tool
       attr_reader name: String
       attr_reader description: String
       attr_reader schema: Hash[Symbol, untyped]
+      attr_reader annotations: Hash[Symbol, untyped]?
 
-      def initialize: (name: String, description: String, ?schema: Hash[Symbol, untyped]) { (Hash[String, untyped]) -> untyped } -> void
+      def initialize: (name: String, description: String, ?schema: Hash[Symbol, untyped], ?annotations: Hash[Symbol, untyped]?) { (Hash[String, untyped]) -> untyped } -> void
       def call: (Hash[String, untyped] arguments) -> Hash[Symbol, untyped]
       def to_mcp_definition: () -> Hash[Symbol, untyped]
     end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: claude_agent
 version: !ruby/object:Gem::Version
-  version: 0.6.0
+  version: 0.7.0
 platform: ruby
 authors:
 - Thomas Carr
@@ -34,14 +34,15 @@ executables: []
 extensions: []
 extra_rdoc_files: []
 files:
-- ".claude/commands/spec/complete.md"
-- ".claude/commands/spec/update.md"
 - ".claude/rules/conventions.md"
 - ".claude/rules/git.md"
 - ".claude/rules/pull-requests.md"
 - ".claude/rules/releases.md"
 - ".claude/rules/testing.md"
 - ".claude/settings.json"
+- ".claude/skills/spec/SKILL.md"
+- ".claude/skills/spec/commands/complete.md"
+- ".claude/skills/spec/commands/update.md"
 - CHANGELOG.md
 - CLAUDE.md
 - LICENSE.txt

data/.claude/commands/spec/update.md

@@ -1,95 +0,0 @@
----
-description: Update SPEC.md with latest feature parity comparison across TypeScript, Python, and Ruby SDKs
-allowed-tools: Bash, Read, Glob, Grep, Write, Edit, TodoWrite, Task
----
-
-# Update SDK Specification Document
-
-You are updating the SPEC.md file to reflect the current feature parity between the TypeScript, Python, and Ruby Claude Agent SDKs.
-
-## Step 1: Update Reference SDKs
-
-First, ensure the reference SDKs are up to date:
-
-```bash
-bin/update-reference-sdks
-```
-
-This clones/updates:
-- Python SDK from GitHub → `vendor/claude-agent-sdk-python`
-- TypeScript SDK from npm → `vendor/claude-agent-sdk-npm`
-
-## Step 2: Research Official Documentation
-
-Use the `claude-code-guide` agent to check for any new or updated SDK features:
-
-```
-Task(subagent_type: "claude-code-guide", prompt: "What are all the features and options in the Claude Agent SDK? Include configuration options, hooks, permissions, MCP support, and control protocol methods.")
-```
-
-This helps catch features that may be documented but not yet in the SDK source files.
-
-## Step 3: Analyze All Three SDKs
-
-Read and compare the following files:
-
-### TypeScript SDK (Primary Reference)
-- `vendor/claude-agent-sdk-npm/sdk.d.ts` - Complete API surface with all types
-
-### Python SDK
-- `vendor/claude-agent-sdk-python/src/claude_agent_sdk/types.py` - Type definitions
-
-### Ruby SDK (This Repository)
-Key files to check:
-- `lib/claude_agent/options.rb` - Configuration options
-- `lib/claude_agent/messages.rb` - Message types
-- `lib/claude_agent/content_blocks.rb` - Content block types
-- `lib/claude_agent/types.rb` - Additional types
-- `lib/claude_agent/hooks.rb` - Hook types
-- `lib/claude_agent/permissions.rb` - Permission types
-- `lib/claude_agent/control_protocol.rb` - Control protocol
-- `lib/claude_agent/sandbox_settings.rb` - Sandbox config
-- `lib/claude_agent/mcp/server.rb` - MCP server
-- `lib/claude_agent/mcp/tool.rb` - MCP tools
-- `lib/claude_agent/client.rb` - Client class
-- `lib/claude_agent/errors.rb` - Error types
-
-## Step 4: Update SPEC.md
-
-Update the existing SPEC.md file with:
-
-1. **Reference Versions** - Update TypeScript SDK version and Python SDK commit
-2. **Feature Tables** - Update all ✅/❌ markers based on current implementations
-3. **New Features** - Add any new features found in the TypeScript SDK
-4. **Removed Features** - Remove any deprecated features
-
-### Categories to Compare
-
-1. Options/Configuration
-2. Message Types
-3. Content Blocks
-4. Control Protocol
-5. Hooks
-6. Permissions
-7. MCP Support
-8. Sessions
-9. Subagents
-10. Sandbox Settings
-11. Error Handling
-12. Client API
-
-## Guidelines
-
-- **Be thorough** - Check every field and option in each SDK
-- **TypeScript is authoritative** - The TypeScript sdk.d.ts is the most complete reference
-- **Preserve format** - Keep the existing table structure and markdown formatting
-- **Update versions** - Always update the reference version info at the top
-- **Note changes** - If significant changes are found, mention them after updating
-
-## Output
-
-After updating SPEC.md, provide a brief summary of:
-- SDK versions checked
-- Any new features added to the spec
-- Any features removed from the spec
-- Any notable gaps in the Ruby SDK