astrid-workspace

Operational workspace boundaries for the Astrid secure agent runtime SDK.

Overview

This crate defines where an agent can freely operate and when it needs user approval to "escape" those boundaries. Unlike the WASM sandbox (which is inescapable for security), the operational workspace is a user-controlled boundary that can be relaxed with approval.

Key Concepts

• Workspace: A directory tree where the agent operates without restrictions
• Escape: Operations outside the workspace require user approval
• Modes: Control how strictly boundaries are enforced

Workspace Modes

Usage

use astrid_workspace::{WorkspaceBoundary, WorkspaceConfig, WorkspaceMode};

let config = WorkspaceConfig::new("/home/user/project")
    .with_mode(WorkspaceMode::Guided);

let boundary = WorkspaceBoundary::new(config);

// Check if a path is allowed
match boundary.check("/home/user/project/src/main.rs") {
    PathCheck::Allowed => println!("Path is in workspace"),
    PathCheck::RequiresApproval => println!("Needs user approval"),
    _ => {}
}

Key Types

• WorkspaceBoundary - Evaluates paths against workspace rules
• WorkspaceConfig - Configuration for workspace boundaries
• WorkspaceMode - Safe, Guided, or Autonomous operation
• PathCheck - Result of checking a path (Allowed, RequiresApproval, Denied)
• EscapeRequest - Request to operate outside the workspace
• EscapeDecision - User's response to an escape request

Escape vs WASM Sandbox

License

This crate is licensed under the MIT license.