Security Checks

This skill provides instructions for performing security audits on the VAK codebase.

Security Implementation Status

ID	Feature	Status	Implementation
SEC-001	Supply Chain Hardening	✅ Complete	`.github/workflows/security.yml`
SEC-002	License Compliance	✅ Complete	`deny.toml`, CI workflow
SEC-003	Unsafe Rust Audit	✅ Complete	`#![deny(unsafe_code)]` + docs
SEC-004	Prompt Injection Protection	✅ Complete	`src/reasoner/prompt_injection.rs`
SEC-005	Rate Limiting	✅ Complete	`src/kernel/rate_limiter.rs`

Prerequisites

•cargo-audit (cargo install cargo-audit --locked)
•cargo-deny (cargo install cargo-deny --locked)
•cargo-geiger (cargo install cargo-geiger --locked)
•cargo-sbom (cargo install cargo-sbom)
•clippy (rustup component add clippy)

CI/CD Integration

Security checks run automatically via .github/workflows/security.yml:

•On push: main, develop branches
•On PR: main branch
•Scheduled: Weekly on Sundays at midnight UTC

Instructions

Vulnerability Scan (SEC-001)

To scan dependencies for known vulnerabilities:

bash

cargo audit

With strict mode (deny warnings):

bash

cargo audit --deny warnings

Generate JSON report:

bash

cargo audit --json > security-audit-report.json

License & Ban Check (SEC-002)

To check license compliance and banned crates:

bash

cargo deny check

The project uses deny.toml with the following allowed licenses:

•MIT, Apache-2.0, BSD-2-Clause, BSD-3-Clause
•ISC, Zlib, 0BSD, MPL-2.0, CC0-1.0, Unlicense

Unsafe Code Audit (SEC-003)

The project uses #![deny(unsafe_code)] by default. To audit for unsafe usage in dependencies:

bash

cargo geiger --all-features --all-targets

Current Status:

•Main crate: Denies unsafe code
•WASM skills: Documented unsafe blocks with // SAFETY: comments

Security Lints (Clippy)

To run security-focused Clippy lints:

bash

cargo clippy --all-targets --all-features -- \
  -D clippy::unwrap_used \
  -D clippy::expect_used \
  -D clippy::panic \
  -W clippy::cognitive_complexity \
  -W clippy::too_many_arguments

Generate SBOM

To generate a Software Bill of Materials:

bash

cargo sbom > sbom.json

Prompt Injection Protection (SEC-004)

The project includes built-in prompt injection detection:

rust

use vak::reasoner::PromptInjectionDetector;

let detector = PromptInjectionDetector::new(DetectorConfig::default());
let result = detector.analyze(input_text);
match result.action {
    RecommendedAction::Allow => { /* proceed */ }
    RecommendedAction::Block => { /* reject input */ }
    _ => { /* log and review */ }
}

Detection categories:

•Instruction override attempts
•Role manipulation
•Context injection
•Unicode smuggling
•Encoding tricks

Rate Limiting (SEC-005)

Per-resource rate limiting is implemented in src/kernel/rate_limiter.rs:

•Per-agent+action+resource key limits
•Token bucket algorithm with configurable refill
•Burst allowance support

Examples

Check Specific Advisory

bash

cargo audit --deny RUSTSEC-2023-0001

Check Licenses Only

bash

cargo deny check licenses

Check Bans Only

bash

cargo deny check bans

Full Security Audit Script

bash

#!/bin/bash
set -e
echo "Running vulnerability scan..."
cargo audit --deny warnings
echo "Running license check..."
cargo deny check
echo "Running unsafe audit..."
cargo geiger --all-features || true
echo "Running security lints..."
cargo clippy --all-targets --all-features -- -D warnings
echo "All security checks passed!"

Guidelines

•Zero Tolerance: Critical and High vulnerabilities must be fixed immediately.
•Deny Unsafe: All unsafe blocks must have a // SAFETY: comment explaining why it is safe.
•License Compliance: Ensure all dependencies are compatible with the project license (MIT/Apache-2.0).
•No Panics: Production code should not use unwrap(), expect(), or panic!(). Use Result instead.
•Default Deny: Policy engine fails closed - no policy = no access (POL-007).
•Audit Trail: All security-relevant operations are logged to the immutable audit log.

Related Files

•.github/workflows/security.yml - CI security workflow
•deny.toml - cargo-deny configuration
•src/reasoner/prompt_injection.rs - Prompt injection detector
•src/kernel/rate_limiter.rs - Rate limiting implementation
•src/policy/enforcer.rs - Policy enforcement with fail-closed