Please report security vulnerabilities by emailing security@coding-swarm.local (replace with actual email) or by opening a GitHub advisory if hosted on GitHub.

Running autonomous AI agents with --dangerously-skip-permissions inside Docker containers carries inherent risks:

1. Container Escape: While we use standard Docker isolation, container escapes are theoretically possible. Do not run agents on sensitive production hosts without additional sandboxing (e.g., gVisor).
2. Resource Exhaustion: Agents run in infinite loops. Use Docker resource limits (--cpus, --memory) to prevent them from consuming all host resources.
3. Network Access: By default, agents may have network access. You can restrict this by passing --network none in config.docker.extra_args, but this will prevent pip install or other network-dependent tasks.

We recommend running the swarm in an ephemeral VM or a dedicated sandbox environment.