plato-sandbox — Isolated Execution Environment

Run agent-generated code in an isolated sandbox with resource limits, policy enforcement, and monitoring. Currently simulated (tracks resources without OS-level isolation), with architecture ready for real process-based sandboxing.

Part of the Plato ecosystem.

What This Gives You

• Configurable resource limits — max memory (MB), max CPU time (ms), max output (bytes)
• Policy enforcement — filesystem, network, subprocess, command allowlists
• Resource monitoring — track memory peak, CPU time, output size during execution
• Isolation levels — Simulated (current), Process, Thread (architecture-ready)
• Violation reporting — detailed limit violations with actual vs. allowed values

Quick Start

use plato_sandbox::*;

let config = SandboxConfig::new(256, 10_000); // 256MB, 10s CPU
let mut sandbox = Sandbox::new(config);
sandbox.set_policy(SandboxPolicy::restrictive()); // no fs, no net, no subprocess

// Check policy before execution
assert!(!sandbox.check_policy("network"));

// Execute code
let result = sandbox.execute("print('hello')", "python")?;
println!("Output: {}", result.output);
println!("Duration: {}ms", result.duration_ms);

Custom Policy

let policy = SandboxPolicy {
    allow_filesystem: true,
    allow_network: false,
    allow_subprocess: false,
    allowed_commands: vec!["python".into(), "node".into()],
};

How It Fits

Guards agent code execution. Before any agent-generated code runs, the sandbox checks plato-policy, enforces resource limits, and reports results. Used by plato-shell when agents need to compute.

Installation

[dependencies]
plato-sandbox = "0.1"

License

MIT