sre-engineer
Warn
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill contains multiple Python scripts that execute shell commands using the
subprocessmodule. - In
references/automation-toil.md, theAutomatedRunbookclass usessubprocess.run(shell=True)to execute arbitrary command strings. This is a dangerous pattern that can lead to remote command execution if the runbook steps are sourced from untrusted data. - In
SKILL.md, theToil Automation Scriptusessys.argvto populate akubectl rollout restartcommand. While passed as a list, the lack of validation on these arguments could lead to unintended behavior if inputs are user-controlled. - The skill uses high-privilege commands including
systemctl restart,iptables(viakubectl exec), andtc(traffic control) for system remediation and chaos experiments. - [EXTERNAL_DOWNLOADS]: Network operations are present for legitimate monitoring purposes.
- The automation script in
SKILL.mdusesurllib.requestto query a Prometheus API. references/automation-toil.mdusescurlto perform service health checks.- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection (Category 8).
- Ingestion points: The skill reads and processes data from Prometheus metrics, Kubernetes logs (
kubectl logs), and service health endpoints. - Boundary markers: None identified in the provided scripts to distinguish between data and instructions.
- Capability inventory: The skill has broad capabilities including file deletion (
find ... -delete), service restarts (systemctl,kubectl), and network configuration changes (iptables,tc). - Sanitization: Minimal sanitization is observed (e.g.,
urllib.request.quotefor URL queries), but command-line arguments and data from logs are interpolated into shell commands without sufficient validation or escaping.
Audit Metadata