Supply Chain Guardian AI

Autonomous, closed-loop CVE detection and remediation for containerized workloads — local CPU inference, cloud LLMs, or bring your own API key.

---

💡 What is Supply Chain Guardian AI?

SupplyChain Guardian AI Explainer Watch the complete explainer covering: Kubernetes DevSecOps Workflow End-to-End Platform Walkthrough

✍️ Building a Zero-Egress AI DevSecOps Pipeline Read about: Lessons from Building SupplyChain Guardian AI

Supply Chain Guardian AI is an all-in-one GitHub Action that automatically secures your containerized applications:

1. 🔍 Scans your container images using Trivy for CRITICAL and HIGH vulnerabilities.
2. 🤖 Patches vulnerable Dockerfiles using AI (Local Ollama, NVIDIA NIM, OpenAI, or DeepSeek).
3. 🧪 Smoke Tests the patched Dockerfile (docker build + runtime health check).
4. ☸️ Validates deployment stability inside a temporary KinD (Kubernetes in Docker) cluster (optional).
5. 🔀 Opens a Pull Request with complete proof and security audit logs.

Push / Cron → Trivy Scan → CVE Found → AI Patches Dockerfile
       → Smoke Test → KinD Validates → Re-scan Confirms → PR Opened

---

📌 Table of Contents

• ⚡ Quickstart
• 📖 Common Workflow Examples
  • 1. Minimal Setup (Default Local Ollama)
  • 2. Recommended Cloud Setup (NVIDIA NIM / DeepSeek / Kimi)
  • 3. Scanning a Specific Dockerfile in a Monorepo
  • 4. Simple App without Kubernetes (KinD Disabled)
  • 5. Full Production Setup (KinD Testing + Auto PR)
• 🤖 Supported AI Providers
• 📋 Complete Inputs & Outputs
• 🛡️ Security & Reliability Features
• ❓ FAQ & Troubleshooting
• 📄 License

---

⚡ Quickstart

Add this step to your GitHub Actions workflow file (e.g. .github/workflows/security.yml):

name: Security Audit
on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

permissions:
  contents: write
  pull-requests: write

jobs:
  scan-and-remediate:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout Code
        uses: actions/checkout@v4

      - name: Run Supply Chain Guardian
        uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
        with:
          dockerfile: './Dockerfile'

---

📖 Common Workflow Examples

1. Minimal Setup (Default Local Ollama)

No API key required! Uses Ollama (llama3.2:1b) directly on the GitHub Actions runner CPU.

- name: Run Supply Chain Guardian (Local AI)
  uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
  with:
    dockerfile: './Dockerfile'

---

2. Recommended Cloud Setup (NVIDIA NIM / DeepSeek / Kimi)

Fast, high-accuracy inference using cloud AI models.

Prerequisite: Add NVIDIA_NIM_API_KEY (or your provider's API key) to your Repository Settings → Secrets and variables → Actions.

- name: Run Supply Chain Guardian (NVIDIA NIM)
  uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
  with:
    dockerfile: './Dockerfile'
    provider: 'openai'
    model: 'moonshotai/kimi-k2.6'  # or 'deepseek-ai/deepseek-v4-flash'
    openai-endpoint: 'https://integrate.api.nvidia.com/v1'
    api-key: ${{ secrets.NVIDIA_NIM_API_KEY }}
  env:
    GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

---

3. Scanning a Specific Dockerfile in a Monorepo

Specify sub-directory paths easily:

- name: Run Supply Chain Guardian (Backend App)
  uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
  with:
    dockerfile: './backend/Dockerfile'
    provider: 'openai'
    model: 'z-ai/glm-5.2'
    openai-endpoint: 'https://integrate.api.nvidia.com/v1'
    api-key: ${{ secrets.NVIDIA_NIM_API_KEY }}

---

4. Simple App without Kubernetes (KinD Disabled)

If your repository does not deploy to Kubernetes, turn off KinD cluster testing by setting kind-enabled: "false".

- name: Run Supply Chain Guardian (KinD Disabled)
  uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
  with:
    dockerfile: './Dockerfile'
    kind-enabled: 'false'
    healthz-port: '5000'
    healthz-path: '/health'
    create-pr: 'true'
  env:
    GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

---

5. Full Production Setup (KinD Testing + Auto PR)

Runs complete scanning, AI remediation, KinD deployment test, and opens an automated security PR.

name: Supply Chain Security Pipeline

on:
  push:
    branches: [main]
  schedule:
    - cron: '0 2 * * *'  # Daily security check at 2:00 AM

permissions:
  contents: write
  pull-requests: write

jobs:
  guardian-security:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout Code
        uses: actions/checkout@v4

      - name: Run Supply Chain Guardian
        uses: barbaria888/SupplyChain-Guardian-AI-Github_Action@v2
        with:
          dockerfile: './backend/Dockerfile'
          provider: 'openai'
          model: 'deepseek-ai/deepseek-v4-flash'
          openai-endpoint: 'https://integrate.api.nvidia.com/v1'
          api-key: ${{ secrets.NVIDIA_NIM_API_KEY }}
          kind-enabled: 'true'
          k8s-manifests: './k8s/'
          healthz-port: '8080'
          healthz-path: '/healthz'
          create-pr: 'true'
          pr-branch: 'security/auto-cve-patch'
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

---

🤖 Supported AI Providers

You can switch between local CPU inference and cloud API providers using the provider and openai-endpoint inputs:

Provider	provider	openai-endpoint	Recommended Models
Local Ollama (Free)	ollama	N/A	llama3.2:1b (default)
NVIDIA NIM	openai	https://integrate.api.nvidia.com/v1	moonshotai/kimi-k2.6, deepseek-ai/deepseek-v4-flash, z-ai/glm-5.2
OpenAI Direct	openai	https://api.openai.com/v1	gpt-4o-mini, gpt-4o
Groq / DeepSeek / Custom	openai	(Your base URL)	Any OpenAI-compatible model

---

📋 Complete Inputs & Outputs

Inputs

Input	Required	Default	Description
dockerfile	No	./Dockerfile	Path to the Dockerfile to scan and remediate
image-ref	No	''	Pre-built image reference (if scanning existing image)
severity	No	CRITICAL,HIGH	Comma-separated Trivy vulnerability severities
provider	No	ollama	AI engine: ollama or openai
model	No	(auto)	Model tag or identifier for the provider
api-key	No	''	API Key for cloud providers (NVIDIA / OpenAI / Custom)
openai-endpoint	No	https://integrate.api.nvidia.com/v1	OpenAI-compatible endpoint URL
trivy-version	No	0.63.0	Trivy CLI version
kind-enabled	No	true	Set to "false" to skip KinD Kubernetes integration tests
kind-config	No	.kind/cluster-config.yaml	Path to custom KinD cluster config
k8s-manifests	No	k8s/	Directory containing Kubernetes manifests to test in KinD
healthz-port	No	18080	Container host port mapped during health check
healthz-path	No	/	Endpoint path probed during container smoke test
create-pr	No	true	Set to "true" to open a PR on successful patch
pr-branch	No	auto-patcher/cve-remediation	Branch name used for the automated PR
pr-labels	No	security,automated-patch	Comma-separated labels applied to the PR
enforce-non-root	No	true	Enforces non-root execution (USER instruction) in patched Dockerfiles
policy-preset	No	strict	Enforcement mode: strict (fail on unresolved CVE) or lax (warn only)

Outputs

Output	Description
vulnerabilities-found	"true" if Trivy detected matching vulnerabilities
patch-applied	"true" if AI successfully generated a patch
smoke-test-passed	"true" if the patched Dockerfile built and booted cleanly
kind-validation-passed	"true" if KinD Kubernetes deployment test succeeded

---

🛡️ Security & Reliability Features

• 🔒 Zero Egress Option: Use provider: 'ollama' for 100% offline, privacy-first patching directly inside the GitHub Actions runner.
• 🛡️ 3-Layer Hallucination Defense:
  1. Syntax Whitelist: Inspects generated Dockerfile instructions (FROM, RUN, COPY, etc.) to prevent LLM hallucinations.
  2. Smoke Test Gate: Verifies docker build and runtime container stability (with dummy DB env injection for app boots).
  3. KinD Cluster Verification: Ensures the container deploys into Kubernetes without CrashLoopBackOff.
• 📜 Full Auditability: Every run generates a patch_audit.log artifact containing full LLM prompts, raw outputs, and scan reports.

---

❓ FAQ & Troubleshooting

Why is the PR creation step failing?

Ensure your workflow file includes the required permissions:

permissions:
  contents: write
  pull-requests: write

Also ensure you pass the GITHUB_TOKEN environment variable:

env:
  GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

Why did the KinD validation step fail with ErrImageNeverPull?

Starting in v2.4.0, all discovered Kubernetes manifests automatically have their container image: replaced with the locally built patched image (guardian-scan:patched) and imagePullPolicy: Never. Make sure you are using @v2 (or @v2.4.0+). If your repository does not use Kubernetes, simply set kind-enabled: "false".

---

License

MIT License — Created by @barbaria888.