vu1nz-gh-actions

GitHub Actions security scanner — detect CI/CD workflow vulnerabilities + AI code review via Claude

A standalone, open-source security scanner for GitHub Actions workflows. Finds script injection, unpinned actions, pull_request_target abuse, secret leaks, permission issues, and supply chain risks in your CI/CD pipelines. Optionally runs AI-powered code review via Claude.

Part of the vu1nz security ecosystem.

---

Features

• 17 security checks — comprehensive coverage of OWASP CI/CD top 10 risks
• AI code review — optional Claude-powered analysis that explains vulnerabilities in plain English
• Reusable GitHub Action — plug into your existing CI/CD pipeline in minutes
• PR comments — automatically posts findings as comments on pull requests
• Configurable severity thresholds — fail builds on critical, high, or medium findings
• JSON reports — machine-readable output for integration with other tools

---

Quick Start

CLI

# Install
pip install vu1nz-gh-actions

# Scan a repository
vu1nz actions scan owner/repo

# Scan with Claude AI code review
export ANTHROPIC_API_KEY=sk-ant-...
vu1nz actions scan owner/repo --claude

# Machine-readable JSON output
vu1nz actions scan owner/repo --json

# Custom GitHub token
vu1nz actions scan owner/repo --token ghp_...

GitHub Actions (CI/CD Integration)

name: Actions Security Scan

on:
  push:
    branches: [main]
  pull_request:

jobs:
  scan-actions:
    runs-on: ubuntu-latest
    steps:
      - uses: profullstack/vu1nz-gh-actions/.github/actions/vu1nz-actions-scan@v1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

With Claude AI Review

      - uses: profullstack/vu1nz-gh-actions/.github/actions/vu1nz-actions-scan@v1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
          claude_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          claude_model: claude-sonnet-4-20250514
          fail_on: high

---

17 Security Checks

#	ID	Severity	Category	Description
1	script_injection_github_context	CRITICAL	script_injection	User-controlled GitHub context (${{ github.event.* }}) interpolated into run: commands
2	pull_request_target_checkout	CRITICAL	context	pull_request_target checking out PR HEAD — attacker runs code with secrets
3	secret_exposure_run_command	CRITICAL	secrets	Secrets inlined in run: commands (captured in logs)
4	untrusted_checkout_self	HIGH	context	PR head checked out without explicit ref on pull_request_target
5	pull_request_target_usage	HIGH	context	pull_request_target trigger used (runs with repo token + secrets)
6	secret_exposure_pr_context	HIGH	secrets	Secrets exposed to PR context via inherit/secrets: write
7	unpinned_action	HIGH	pinning	Third-party action by mutable ref (tag/branch) instead of commit SHA
8	matrix_injection	HIGH	script_injection	Dynamic matrix values from untrusted context
9	untrusted_workflow_call	HIGH	supply_chain	External workflow by mutable ref instead of commit SHA
10	self_hosted_runner	HIGH	supply_chain	Non-ephemeral self-hosted runner in workflow
11	dangerous_deployment_env	HIGH	deployment	Deployment environment without required reviewers
12	write_all_permissions	MEDIUM	permissions	Workflow has write-all permissions
13	unpinned_docker	MEDIUM	pinning	Container image by tag instead of digest (@sha256:)
14	dangerous_artifact_mismatch	MEDIUM	supply_chain	Artifact upload/download version mismatch
15	deploy_on_push_to_main	MEDIUM	deployment	Deploys on push to main without PR/CI checks
16	missing_permissions_top	LOW	permissions	No top-level permissions: block defined
17	secret_exposure_run_command_high	HIGH	secrets	Secrets via env: in workflow steps

---

Output

Human-readable (default)

Scanning profullstack/vu1nz-actions-test...
Found 4 workflow(s). Scanning...

[CRITICAL] Script Injection via GitHub Context
  File: ci.yml (line 12)
  Detail: run: echo "Running tests for PR ${{ github.event.pull_request.title }}"
  Fix: Pass context values as env vars instead of inline

[HIGH] Unpinned Action
  File: deploy.yml (line 15)
  Detail: uses: some-org/deploy-action@v1
  Fix: Pin to commit SHA: some-org/deploy-action@abc123def456...

JSON (--json flag)

{
  "repository": "owner/repo",
  "workflow_count": 4,
  "total_jobs": 7,
  "findings": [...],
  "severity_summary": {
    "critical": 3,
    "high": 9,
    "medium": 6,
    "low": 3
  }
}

---

Requirements

• Python 3.10+
• GitHub token with actions: read and contents: read permissions
• (Optional) Anthropic API key for Claude AI code review

---

License

MIT — see LICENSE.