Trait PermissionPolicy

Source

pub trait PermissionPolicy:
    Send
    + Sync
    + 'static {
    // Required method
    fn decide(&self, request: &PermissionRequest) -> PolicyDecision;

    // Provided method
    fn supports_interaction(&self) -> bool { ... }
}

Expand description

Policy for handling permission requests.

Implement this trait to automatically approve, deny, or filter permission requests before they reach the user. Useful for:

Headless servers: Auto-approve everything in trusted environments
Allowlists: Only prompt for paths/commands outside a safe list
Audit logging: Log all permission requests regardless of decision
Rate limiting: Deny requests that exceed usage thresholds

§Example: Custom Allowlist Policy

use agent_core_runtime::agent::interface::{PermissionPolicy, PolicyDecision};
use agent_core_runtime::permissions::{PermissionRequest, GrantTarget};

struct AllowlistPolicy {
    allowed_paths: Vec<String>,
}

impl PermissionPolicy for AllowlistPolicy {
    fn decide(&self, request: &PermissionRequest) -> PolicyDecision {
        match &request.target {
            GrantTarget::Path { path, .. } => {
                let path_str = path.to_string_lossy();
                if self.allowed_paths.iter().any(|p| path_str.starts_with(p)) {
                    PolicyDecision::Allow
                } else {
                    PolicyDecision::AskUser
                }
            }
            _ => PolicyDecision::AskUser,
        }
    }
}

Required Methods§

Source

fn decide(&self, request: &PermissionRequest) -> PolicyDecision

Decide how to handle a permission request.

Called before the request is sent to the consumer. Return:

Allow or AllowWithGrant to approve immediately
Deny to reject immediately
AskUser to forward to the consumer for user decision

Provided Methods§

Source

fn supports_interaction(&self) -> bool

Whether this policy supports interactive user questions.

Returns false for headless/auto-approve policies, causing UserInteractionRequired events to be auto-cancelled (the tool receives “User declined to answer”).

Returns true (default) for interactive policies where a user can answer questions.