Crate agentshield 

AgentShield — Security scanner for AI agent extensions.

Offline-first, multi-framework, SARIF output. Scans MCP servers, OpenClaw skills, and other agent extension formats for security issues.

Quick Start

use std::path::Path;
use agentshield::{scan, ScanOptions};

let options = ScanOptions::default();
let report = scan(Path::new("./my-mcp-server"), &options).unwrap();
println!("Pass: {}, Findings: {}", report.verdict.pass, report.findings.len());

Modules

adapter

analysis

baseline

Baseline schema for tracking known findings across scan runs.

certify

DSSE envelope and in-toto attestation for scan results.

config

doctor

egress

Egress policy module for runtime network enforcement.

error

fix

ir

Unified Intermediate Representation for agent extension analysis.

output

parser

rules

ux

Structs

ScanOptions

Options for a scan invocation.

ScanReport

Complete scan report.

Functions

render_report

Render a scan report in the specified format.

render_report_with_experimental_risk

Render a scan report with the opt-in experimental informational risk index.

scan

Run a complete scan: detect framework, parse, analyze, evaluate policy.

scan_with_path_filter_overrides

Run a complete scan with optional include/exclude patterns layered on top of the scan configuration.