@syedahmedx3/aiguard

A zero-dependency, type-safe, fail-safe closed security guardrail for AI agents and LLM applications. Protect local application execution paths from directory traversal attacks, accidental sensitive resource leakage, or unauthorized system access.

Features

• 🛡️ Fail-Safe Closed Invariants: Any error, edge case, or misconfiguration results in strict access denial.
• 📂 Path Containment Enforcement: Prevents directory traversal (../), null-byte manipulation, and URL-style paths.
• 🔄 Symlink Attack Mitigation: Validates file descriptors atomically post-open to neutralize race conditions (TOCTOU).
• 📜 Policy Engine: Pattern matches against an .aipolicy file using robust glob mechanics via minimatch.
• 📊 Structured Auditing: Built-in, high-performance memory and appending file-based audit logging sinks.
• 🔒 Secret Redaction (Phase 3): Opt-in structural text masking for high-risk tokens (OPENAI_API_KEY, private keys, etc.).
• 🤝 Interactive Consent (Phase 4): Asynchronous hook architecture ("Ask Mode") for real-time human-in-the-loop validation.

Installation

npm install @syedahmedx3/aiguard