Crate oxidized_agentic_audit 

oxidized-agentic-audit

Security auditing framework for AI agent skills.

oxidized-agentic-audit scans skill directories for security vulnerabilities including prompt injection, dangerous bash patterns, exposed secrets, unsafe package installations, and more. It runs multiple scanners in parallel and produces reports in human-readable, JSON, or SARIF formats.

Quick start

use std::path::Path;
use oxidized_agentic_audit::{scan::{self, ScanMode}, config::Config, output};

let config = Config::load(None).expect("failed to load config");
let report = scan::run_scan(Path::new("./my-skill"), &config, ScanMode::Skill);

if report.passed {
    println!("Scan passed!");
} else {
    let text = output::format_report(&report, &output::OutputFormat::Pretty);
    print!("{text}");
}

Architecture

The crate is organized around a pipeline:

1. config — load and validate configuration from TOML files.
2. scanners — pluggable scanners::Scanner trait with built-in implementations (prompt, bash patterns, secrets, shellcheck, semgrep, …).
3. scan — orchestrate scanners in parallel and collect results.
4. finding — core data types (finding::Finding, finding::ScanReport).
5. output — format reports as pretty text, JSON, or SARIF.

Scanners

Scanner	External tool	Description
prompt	—	Prompt injection pattern detection
bash_patterns	—	Dangerous bash anti-patterns (Categories A–H)
typescript_patterns	—	Dangerous TypeScript/JavaScript patterns (Categories A–H)
package_install	—	Unsafe package installation patterns
frontmatter	—	SKILL.md frontmatter validation
shellcheck	shellcheck	Shell script linting
secrets	gitleaks	Secret and credential scanning
semgrep	semgrep	Static analysis

Modules

config

Configuration loading and management.

finding

Core data types for scan findings and reports.

output

Output formatting for scan reports.

scan

Scan orchestration.

scanners

Pluggable security scanners.