Skip to main content

Agent Security & Isolation

Jenkins agents execute untrusted and user-controlled code. If agents are not isolated properly, they become the easiest entry point for attackers.

Why Agent Security Matters​

Agent compromise can lead to:

  • Credential theft
  • Lateral movement to other systems
  • Controller compromise
  • Infrastructure abuse

Assume every build is potentially hostile.

Trust Model: Controller vs Agents​

  • Controller: trusted, sensitive, minimal execution
  • Agents: untrusted, disposable, isolated

Never treat agents as trusted machines.

Isolation Strategies​

Static Agents (Least Secure)​

  • Long-lived
  • Shared state
  • High blast radius

Use only for trusted workloads.

Docker Agents​

  • Container-level isolation
  • Shared host kernel
  • Moderate security

Good baseline isolation.

  • Pod-level isolation
  • Namespace separation
  • Resource limits

Best balance of security and scalability.

One Build per Agent Rule​

Strongly recommended:

  • One executor per agent
  • One pipeline per agent

This prevents:

  • Cross-build contamination
  • Secret leakage
  • State bleed

Credential Exposure Control​

Rules:

  • Never store secrets on agent filesystem
  • Use short-lived credentials
  • Scope credentials per job
  • Avoid global credentials

Prefer runtime injection.

Network Isolation​

  • Separate agent networks
  • Restrict outbound traffic
  • Block access to controller APIs where possible

Agents should not talk freely.

Agent Permissions​

  • No root access
  • No cloud admin permissions
  • Minimal IAM roles
  • No SSH access between agents

Kubernetes-Specific Hardening​

  • Use dedicated namespaces
  • Enable Pod Security Standards
  • Avoid privileged containers
  • Disable hostPath volumes

Common Agent Security Failures​

  • Reusing static agents
  • Running multiple builds per agent
  • Over-privileged cloud IAM roles
  • Agents with SSH access to controller

Best Practices​

  • Ephemeral agents only
  • One build per agent
  • Tight network rules
  • Separate prod and non-prod agents
  • Regularly rotate credentials

Interview Focus Areas​

  • Why agents are untrusted
  • Static vs ephemeral agent risk
  • Isolation strategies
  • Kubernetes vs Docker agent security