Main application settings. Settings are loaded from multiple sources in order: 1. Default values 2. JSON config file (.clawcode.json)

Result of handling a built-in slash command.

Response from tool execution. Attributes: content: The response content metadata: Optional metadata (as JSON string) is_e

Represents a tool call from the LLM. Attributes: id: Unique call identifier name: Name of the tool to call input: Tool in

LLM provider configuration.

Central orchestrator for the Claude Code compatible plugin system. Responsibilities: - Discover and load plugins from user / project / extra

Complete response from LLM. Attributes: content: The text content thinking: The thinking/reasoning content tool_calls: Li

Execution context for tool calls. Provides access to services and information during tool execution.

Language Server Protocol configuration.

Configuration for a language server.

HUD-aligned snapshot for slash handlers (built by ChatScreen).

Metadata about a tool. Attributes: name: Unique tool identifier description: Human-readable description parameters: JSON

Represents a tool call from the LLM. Attributes: id: Unique call identifier name: Name of the tool to call input: Tool in

Failed to materialize a plugin directory.

Token usage information.

Domain model for a conversation message. Attributes: id: Unique message identifier session_id: Parent session ID role: Me

Persist and load plan artifacts from `.claw/plans/` with fallback reads.

Options object for build system

OpenAI GPT and O series reasoning provider. Supports GPT-4, GPT-4 Turbo, GPT-4o, GPT-3.5 Turbo, and O series reasoning models. Also supports

MCP protocol error.

Metadata about a tool. Attributes: name: Unique tool identifier description: Human-readable description parameters: JSON

Main chat screen showing conversation interface.

Mock LLM provider for testing. Simulates LLM responses without calling real APIs.

A directory (or file leaf) in the scanned project tree.

Base exception for provider errors.

Plain text content.

AI Agent implementing ReAct loop for tool-augmented LLM interaction. The agent manages: - Conversation history management - Streaming LLM

Extension package entry. Attributes: name: Package name description: Package description homepage_url: Package home page

Thread-safe registry of running and finished background processes.

A tracked background process with output buffering.

Scanned project directory tree with architecture classification.

Action result Attributes: status: Action status - ["ok", "warning", "error"] message: Action status explanation needs_res

Persistent mapping between architecture layers and project directories.

Extension manager using pip as package manager and PyPi.org as packages source.

Aggregate state for the Code Awareness panel.

Static config statistics shown in the HUD.

Maps logical operations to registered ``BaseTool`` instances and runs them.

Event from agent processing. Attributes: type: Event type message: Associated message error: Error if any session

Aggregate HUD state that can be rendered to a multi-line string.

A permission request for tool execution. Attributes: tool_name: Name of the tool being called description: Description of what th

A modal dialog for requesting permission to execute a tool. Users can choose to: - Allow once (a) - Always allow for session (y) - De

Shell configuration for bash tool. On Windows, defaults to ``pwsh``/``powershell`` when on PATH, else ``cmd.exe``. PowerShell is invoked via

Plain text content.

A color theme for the ClawCode TUI.

Agent configuration.

Anthropic Claude LLM provider. Supports: - Claude 3.5 Sonnet, Haiku - Claude 3 Opus - Tool calling - Streaming responses - Ex

Todo item shown in the HUD, aggregated from runtime tool events.

Plugin configuration (Claude Code compatible). ``data_root_mode`` selects where plugin data lives (mirrors Claude Code's ``~/.claude`` layout

Extension manager without installation capabilities.

Streaming progress / result chunk.

Result of inspecting a user message for slash handling.

Thread-safe iteration counter for parent/child agent turns. Mirrors reference ``IterationBudget``: caps total LLM-call iterations; optional r

Raised when a website policy file is malformed.

One subagent (Claude Code frontmatter + body prompt).

JSON-RPC 2.0 Request. Attributes: jsonrpc: Protocol version (always "2.0") id: Request ID method: Method name par

Run commands on the host; optional persistent ``bash -l`` session (Claw-aligned).

Response to a permission request. Attributes: granted: Whether permission was granted session_scoped: Whether permission applies

Domain model for a chat session. Attributes: id: Unique session identifier parent_session_id: Parent session ID (for sub-tasks)

Service for managing conversation sessions.

Response from tool execution. Attributes: content: The response content metadata: Optional metadata (as JSON string) is_e

Rich renderables inside the conversation (Markdown, tool lines, etc.).

Large UI colors for the chat screen subtree (programmatically applied).

Single file change item grouped by architecture context.

Tool for searching content in files. Supports: - Regular expression search - Case-sensitive/insensitive search - Directory specificat

Service for managing conversation messages. Provides CRUD operations and handles ContentPart serialization.

Service for managing chat sessions. Provides CRUD operations and publishes events for session changes.

Extended thinking content.

Hybrid monitor: event-driven queue plus adaptive heartbeat polling.

Coder :class:`Agent` with a dedicated Claw-mode entrypoint. The typical agent ``run_conversation`` pattern maps conceptually to this class plus

Closed-loop learning service for DeepNote + ECAT integration.

Exception raised when patch application fails.

Bootstrap services and run workflows.