The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple templates and examples that utilize subprocess.run to execute system-level utilities for automation and monitoring.\n
Evidence: Use of kubectl rollout restart and kubectl exec in SKILL.md and references/automation-toil.md; use of df, find, and curl in references/automation-toil.md; and tc and iptables in references/incident-chaos.md.\n
Context: These commands are used appropriately within the context of SRE tasks like self-healing and chaos engineering.\n- [PRIVILEGE_ESCALATION]: Several scripts provided as references perform operations that typically require elevated privileges on a production system.\n
Evidence: Scripts in references/automation-toil.md and references/incident-chaos.md interact with systemctl, iptables, and tc to manage services and network configurations.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where the agent ingests external operational data, creating a potential surface for indirect prompt injection.\n
Ingestion points: Monitoring metrics from Prometheus (SKILL.md), system logs (references/monitoring-alerting.md), and incident descriptions (references/incident-chaos.md).\n
Boundary markers: Absent; there are no specific instructions or delimiters to isolate ingested data from the agent's core instructions.\n
Capability inventory: The skill provides capabilities for arbitrary command execution via subprocess.run, Kubernetes management via kubectl, and service control via systemctl across SKILL.md and references/automation-toil.md.\n
Sanitization: Absent; while URL encoding is used for internal API queries, there is no evidence of sanitization for broader data ingestion like log content or incident reports.

sre-engineer