Elastic medium stable eql
Execution via GitHub Actions Runner
This rule detects potentially dangerous commands spawned by the GitHub Actions Runner.Worker process or by shell interpreters launched via a runner entrypoint script on self-hosted runner machines. Adversaries who gain the ability to modify or trigger workflows in a linked GitHub repository can execute arbitrary commands on the runner host. This behavior may indicate malicious or unexpected workflow activity, including code execution, reconnaissance, credential harvesting, or network exfiltration initiated through a compromised repository or unauthorized workflow.
Detection Logic
process where event.type == "start" and event.action in ("exec", "exec_event", "start", "ProcessRollup2", "executed", "process_started") and
(
/* Direct child of the GitHub Actions Runner.Worker process */
process.parent.name in ("Runner.Worker", "Runner.Worker.exe") or
/* Child of a shell interpreter launched via a runner entrypoint script
(e.g. /home/runner/runners/<ver>/run/entrypoint.sh or similar paths) */
(
process.parent.name in ("sh", "bash", "zsh") and
process.parent.command_line like "*runner*entrypoint.sh"
)
) and
(
process.name : (
/* Network / download utilities */
"curl", "curl.exe", "wget", "wget.exe",
/* Windows scripting & LOLBins */
"powershell.exe", "cmd.exe", "pwsh.exe", "certutil.exe", "rundll32.exe",
/* Unix shells */
"bash", "sh", "zsh", "dash", "ash", "tcsh", "csh", "ksh", "fish", "mksh", "busybox", "pwsh",
/* File / archive manipulation */
"tar", "gzip", "rm", "sed", "chmod",
/* macOS-specific */
"osascript",
/* Process persistence helpers */
"nohup", "setsid",
/* Scripting runtimes */
"python*", "perl*", "ruby*", "lua*", "php*", "node", "nodejs", "node.exe",
/* Discovery & reconnaissance */
"pgrep", "grep", "find", "printenv", "env", "nmap",
/* Crypto / encoding (potential exfiltration or C2 channel) */
"openssl", "base64", "basez", "base64plain", "base64url", "base64mime", "base64pem", "basenc", "base32", "base16", "xxd",
/* Data manipulation / inspection */
"tr", "cat",
/* Network relay / tunneling */
"nc", "ncat", "netcat", "nc.traditional", "nc.openbsd", "socat", "wg", "wg-quick",
/* Remote access */
"ssh", "ssh.exe", "ftp", "tftp", "scp", "sftp",
/* Kubernetes / infrastructure */
"kubectl", "helm", "docker", "ctr", "crictl",
/* Secret management */
"vault",
/* GitHub CLI */
"gh",
/* AWS CLI */
"aws",
/*Azure CLI */
"az",
/*GCP CLI */
"gcloud",
/* Google Workspace CLI */
"gws"
) or
process.executable : ("/tmp/*", "/private/tmp/*", "/var/tmp/*", "/dev/shm/*", "/run/*", "/var/run/*", "?:\\Users\\*")
) and
not (
/* Runner orchestration layer. Runner.Worker writes every `run:` step to <work>/_work/_temp/<guid>.<ext>
and launches the configured shell on it (bash/sh/pwsh/powershell/cmd/python built-ins all use the
same template), and runs JavaScript actions as node <work>/_work/_actions/<owner>/<repo>/<ref>/...
The exclusion is anchored on the runner `_work/_temp` and `_work/_actions` directories so that
admin-configured job-extension hook scripts on an arbitrary path that merely contains `_temp` are
NOT suppressed. docker and tar are intentionally NOT excluded, because an attacker-controlled
workflow can invoke them directly to stage, archive, or execute payloads. */
process.parent.name in ("Runner.Worker", "Runner.Worker.exe") and
(
(process.name : ("bash", "sh", "dash", "cmd.exe", "python", "python3", "python.exe") and
process.args : ("*/_work/_temp/*", "*\\_work\\_temp\\*")) or
(process.name : ("pwsh", "pwsh.exe", "powershell.exe") and
process.args : ("*/_work/_temp/*.ps1*", "*\\_work\\_temp\\*.ps1*")) or
(process.name : ("node", "node.exe") and
process.args : ("*/_work/_actions/*", "*\\_work\\_actions\\*"))
) and
not process.executable : ("/tmp/*", "/private/tmp/*", "/var/tmp/*", "/dev/shm/*")
) False Positives
- ⚠ Authorized GitHub actions runner with no malicious workflow actions.
Field Validations
Loading…
Comments (0)
Loading comments...