Expand description
AgentGuard - Linux control plane for AI agent security.
§Scaffold notice
This crate is a scaffold: it defines the security trait surface and
receipt types, but does not implement any OS-level enforcement (BPF LSM,
cgroup v2, Landlock, seccomp, or eBPF). The initialize() method only
sets an internal boolean. No actual sandboxing is applied.
Do not rely on this crate for production agent containment. Until real enforcement is implemented, treat all security decisions as advisory.
§Linux Only
This crate is only available on Linux. It will not compile on other systems.
§Example
ⓘ
use agent_guard::{AgentGuard, Subject, Action, ActionType};
let mut guard = AgentGuard::new();
guard.initialize()?;Structs§
- Action
- An action being attempted.
- Agent
Guard - AgentGuard security manager.
- Security
Decision - Represents a security decision made by the control plane.
- Subject
- The entity a security decision applies to.
Enums§
- Action
Type - Types of actions that can be secured.
- Error
- Errors that can occur in agent-guard operations.
- Security
Mechanism - A security mechanism that was applied.
Traits§
- Control
Plane - Control plane for enforcing security policies on agents.
Type Aliases§
- Result
- Result type alias for agent-guard operations.