If you discover a security vulnerability in crew-runner, please report it by emailing security@alteriom.net or opening a private security advisory on GitHub.

Do NOT open public issues for security vulnerabilities.

• Description of the vulnerability
• Steps to reproduce
• Potential impact
• Suggested fix (if available)

• Initial response: Within 48 hours
• Patch timeline: Within 7 days for critical issues
• Public disclosure: After patch is released

Current Status (v1.0.0): ⚠️ No authentication implemented

Crew-runner v1.0.0 does NOT implement authentication. It should only be deployed:

• Behind a firewall
• On internal networks only
• With access restricted to trusted backend services

Planned (v1.1.0): API key authentication for all execution endpoints.

• Deploy behind reverse proxy (Caddy, nginx)
• Use internal Docker networks
• Pin image versions (not :latest)
• Regularly update dependencies
• Monitor logs for suspicious activity
• Use secrets management for credentials

• Expose port 8081 publicly
• Use :latest tag in production
• Commit API keys or credentials
• Run as root (container uses non-root appuser)
• Disable health checks

Environment Variables:

# Required
OLLAMA_BASE_URL=http://ollama:11434
LITELLM_API_KEY=sk-...
BACKEND_URL=http://backend:3007

# Optional
MAX_SESSIONS=2
WORKER_HEARTBEAT_INTERVAL=60

Secrets Storage:

• Use Docker secrets or environment files (.env)
• Never commit .env files to git
• Rotate API keys monthly
• Use separate credentials per environment (dev/staging/prod)

Non-Root User:

# Crew-runner runs as non-root user 'appuser'
USER appuser

Multi-Stage Build:

• Build dependencies isolated from runtime
• Minimal runtime image (python:3.12-slim)
• No build tools in final image

Health Checks:

healthcheck:
  test: ["CMD", "curl", "-f", "http://localhost:8081/health"]
  interval: 30s
  timeout: 10s
  retries: 3

Internal Network Only:

# docker-compose.prod.yml
services:
  crew-runner:
    networks:
      - internal  # Not exposed to public network

Port Binding:

# Bind to localhost only (not 0.0.0.0)
ports:
  - "127.0.0.1:8081:8081"

• Crew-runner uses pure Python/FastAPI (no subprocess spawning)
• CrewAI execution is in-process
• No SIGCHLD handler needed
• Verified: Zero zombie processes under load

• Planned for v1.1.0
• Current mitigation: Deploy behind reverse proxy with rate limiting

• Pydantic validates request structure
• No additional prompt sanitization
• Planned: Input filtering for v1.1.0

Execution Logging:

logger.info(
    "execute: starting execution=%s tenant=%s agents=%d tasks=%d",
    execution_id, tenant_id, len(agent_configs), len(task_configs)
)

Recommended Log Retention:

• Production: 30 days minimum
• Audit compliance: 90 days

Automated Updates:

• Dependabot enabled (GitHub)
• Security advisories monitored
• Critical updates: Deployed within 24 hours

Dependency Pinning:

# requirements.txt
fastapi>=0.115,<1.0  # Allow minor updates, not major
pydantic>=2.0,<3.0   # Pin major version

In Case of Security Incident:

1. Immediate: Stop affected containers
2. Assess: Review logs, identify scope
3. Patch: Deploy security fix
4. Notify: Inform affected parties
5. Document: Post-mortem report

Emergency Contacts:

• Email: security@alteriom.net
• GitHub: @Alteriom/security-team

• ✅ API key authentication
• ✅ Rate limiting (per-client)
• ✅ Input sanitization
• ✅ Request signing (HMAC)

• ✅ JWT token support
• ✅ Role-based access control (RBAC)
• ✅ Audit trail export
• ✅ Security headers (CORS, CSP)

• ✅ OAuth2 integration
• ✅ mTLS support
• ✅ Encryption at rest
• ✅ Security scan integration (Trivy)

This security policy is part of the crew-runner project and is subject to the same license terms.

Last Updated: April 14, 2026