Phil Gearphilgear.dev
securitySecurity
Published: 2025-01-15

Sentinel Security Guard

Automated Supply Chain & Egress Firewall

Zero-trust supply-chain security analyzer and egress boundary guardian for modern web applications and AI agent runtimes.

Analysis Speed< 450ms per repo
CVE Coverage100% OSV Database
SSRF PreventionStrict Whitelist Regex
Pipeline IntegrationGitHub Actions + CLI
#Python#Security#GitHub Actions#CodeQL#Zero-Trust#Egress Filtering#AST Parsing

Architecture Highlights

  • •Static AST parser detecting un-sanitized network outbound calls and SSRF risks.
  • •AI Prompt Injection boundary sanitizer isolating user context from base system instructions.
  • •Automatic package-lock override integrity verification preventing dependency hijacking.
  • •Ephemeral container egress filtering locking outbound packets to verified domain endpoints.

Engineering Hurdles Solved

  • •Distinguishing legitimate third-party API webhooks from stealth data exfiltration attempts in obfuscated JavaScript dependencies.
  • •Preventing indirect prompt injection in autonomous LLM agent execution graphs without breaking reasoning capabilities.

Executive Overview

Sentinel Security Guard enforces zero-trust runtime and build-time guardrails for AI applications, Node.js monorepos, and Python sidecars. It halts untrusted outbound egress requests, isolates untrusted AI inputs, and eliminates supply-chain vulnerabilities before code reaches production environments.

In modern cloud applications and AI agent workflows, software development lifecycles are vulnerable to dependency confusion attacks, compromised transitive npm packages, and indirect prompt injections. Sentinel acts as an immutable boundary guardian, inspecting code syntax trees, build scripts, and network socket allocations.

The Agentic Egress Dilemma

Autonomous AI agents equipped with code execution or terminal tools create severe data exfiltration attack surfaces. If an agent ingests an untrusted web page or user document containing an adversarial prompt injection, it may attempt to transmit local environment variables (AWS_SECRET_ACCESS_KEY, GEMINI_API_KEY) via curl or DNS tunneling. Sentinel enforces strict egress boundary isolation at the OS socket layer.

The 4 Pillars of Sentinel Defense

Sentinel 4-Pillar Zero-Trust Architecture
Fig 1.0 —Sentinel 4-Pillar Zero-Trust Architecture: Static AST code parser, OS-level egress socket firewall, prompt injection isolation mesh, and SHA-512 lockfile integrity auditing.

Continuous CI/CD Enforcement

Sentinel runs both as a lightning-fast local CLI tool for engineers and as a non-bypassable GitHub Actions status check:

  • Pre-Commit Hook: Analyzes staged diffs in under 450ms, warning engineers of exposed credentials or unvetted dependencies before commit creation.
  • Hermetic Build Enforcement: Ensures production builds run in read-only containers (--read-only --network none) with zero access to the internet during asset compilation.

Explore More Systems

View full portfolio of AI architectures, creative technology, and open systems.

All Projects