Struct ToolExecutor 

pub struct ToolExecutor { /* private fields */ }

Tool executor with workspace sandboxing

This is the main entry point for tool execution. It wraps the ToolRegistry and provides backward-compatible API. Includes file version history tracking for write/edit/patch operations.

Defense-in-depth: An optional permission policy can be set to block denied tools even if the caller bypasses the agent loop’s authorization.

Implementations

impl ToolExecutor

pub fn new(workspace: String) -> Self

pub fn set_guard_policy(&mut self, policy: Arc<RwLock<PermissionPolicy>>)

Set a defense-in-depth permission policy.

When set, every tool execution is checked against this policy. Tools that match a Deny rule are blocked before execution. This is a safety net for code paths that bypass the agent loop.

pub fn workspace(&self) -> &PathBuf

Get the workspace path

pub fn registry(&self) -> &ToolRegistry

Get the tool registry for dynamic tool registration

pub fn register_dynamic_tool(&self, tool: Arc<dyn Tool>)

Register a dynamic tool at runtime (e.g., MCP tools, LSP tools, task tool). The tool becomes immediately available in all future tool executions.

pub fn unregister_dynamic_tool(&self, name: &str)

Unregister a dynamic tool by name

pub fn file_history(&self) -> &Arc<FileHistory>

Get the file version history tracker

pub async fn execute(&self, name: &str, args: &Value) -> Result<ToolResult>

Execute a tool by name using the server-level default context

pub async fn execute_with_context( &self, name: &str, args: &Value, ctx: &ToolContext, ) -> Result<ToolResult>

Execute a tool by name with a per-session context

pub fn definitions(&self) -> Vec<ToolDefinition>

Get all tool definitions for LLM