Skip to content

Security: vivekyarra/argus-agent

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
2.1.x ✅ Actively supported
< 2.0 ❌ No longer supported

Reporting a Vulnerability

If you discover a security vulnerability in ARGUS, please report it responsibly:

  1. Do NOT open a public GitHub issue for security vulnerabilities
  2. Email: security@argus-agent.dev (or open a private advisory)
  3. Include a detailed description with reproduction steps
  4. Allow up to 72 hours for an initial response

Security Architecture

ARGUS implements defense-in-depth with the following security controls:

Input Validation (OWASP A03:2021 — Injection)

  • All WebSocket payloads are validated with Pydantic V2 strict models
  • Command text is sanitized (null bytes, ANSI escapes, control characters removed)
  • URL schemes are restricted to http:// and https:// only
  • Base64 payloads are validated for size and encoding correctness
  • Image dimensions are capped at 4096×4096 pixels

Authentication (OWASP A07:2021 — Identification & Auth Failures)

  • Optional API key authentication on WebSocket connections
  • Constant-time key comparison using HMAC to prevent timing attacks
  • Keys sourced from Google Cloud Secret Manager in production

Rate Limiting (OWASP A04:2021 — Insecure Design)

  • Per-client IP rate limiting with configurable window
  • Token-bucket algorithm with automatic cleanup of expired entries
  • Default: 60 requests/minute per client

Injection Prevention (OWASP A03:2021 — Injection)

  • Zero shell=True in the entire codebase
  • URL opening uses webbrowser.open() instead of subprocess
  • No string interpolation in system commands
  • PyAutoGUI actions use coordinate-based interaction only

Security Headers (OWASP A05:2021 — Security Misconfiguration)

  • Content-Security-Policy: script-src 'strict-dynamic' 'self'
  • X-Frame-Options: DENY
  • X-Content-Type-Options: nosniff
  • Referrer-Policy: strict-origin-when-cross-origin
  • Permissions-Policy: camera=(), microphone=(), geolocation=()

Secrets Management

  • Production credentials loaded via Google Cloud Secret Manager
  • .env files used only for local development
  • No secrets committed to version control
  • .env listed in .gitignore

Container Security

  • Docker image runs as non-root user (appuser)
  • Minimal base image (python:3.11-slim)
  • No unnecessary system packages installed
  • Health check endpoint for container orchestration

Data Protection

  • Screenshots are stored in Google Cloud Storage with default encryption
  • Firestore data is encrypted at rest by Google Cloud
  • Action logs use append-only JSONL (no mutation of existing records)

Dependency Management

  • Dependencies pinned with minimum version constraints
  • Security linting via ruff with flake8-bandit rules enabled
  • Type checking via mypy --strict to catch unsafe patterns

Security Testing

The test suite includes dedicated security tests:

  • tests/test_security.py — Rate limiting, auth, sanitization, URL validation
  • tests/test_vulnerabilities.py — ReDoS resistance, command injection prevention, path traversal blocking, XSS payload rejection
  • All tests run in CI before deployment

Acknowledgments

This security policy follows the OWASP Top 10 (2021) framework for web application security risk assessment.

There aren't any published security advisories