field-cage: a local-first alternative for GitHub Actions egress control

<field-cage: a local-first alternative for GitHub Actions egress control>

Field-cage is an MIT-licensed eBPF agent designed to monitor and restrict outbound connections from Linux GitHub Actions runners. It aims to provide a smaller scope compared to StepSecurity Harden-Runner, which is a CI/CD runtime security platform. Field-cage focuses on egress filtering with explicit policies, local reporting, and no hosted backend.

Both tools can observe and block outbound connections not permitted by policy. However, field-cage differs in the infrastructure and security context it provides. Harden-Runner offers a comprehensive platform with network, process, and file activity monitoring, behavioral baselines, and managed threat intelligence. Field-cage, on the other hand, keeps reporting local, has no external service requirement, and does not provide centralized management or enterprise support.

Field-cage uses an allowlist of explicit IPv4, IPv6, and CIDR entries, A/AAAA domain records resolved at startup, and DNS responses observed during job execution. Unknown YAML keys are rejected, and each permitted destination remains visible in code review. Field-cage operates in two modes: audit (records connection attempts without blocking) and block (applies default-deny enforcement to IPv4 and IPv6 connections).

To use field-cage in GitHub Actions, the action uses the takihito/field-cage@v0.1.0 release, verifies its SHA-256 checksum, and starts the agent in the background. It is recommended to initially run in audit mode to review observed connections and then switch to block mode. The action should run near the beginning of the job.

Field-cage is suitable for private repositories that prefer free egress monitoring and blocking, a small, inspectable, fully static Go binary, explicit network policy stored with the repository, and local logs that leave the runner only when configured to do so. It is not intended to replace Harden-Runner, which offers more advanced features such as process execution and file-write monitoring, correlation with workflow steps, historical baselines, anomaly detection, enterprise support, and centralized visibility across repositories and runners.