The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains scripts that execute high-privilege system commands across multiple files.
Evidence: Use of iptables for network partitioning and tc (traffic control) for latency injection in references/incident-chaos.md.
Evidence: Use of systemctl restart and kubectl delete for automated remediation in references/automation-toil.md.
Evidence: The AutomatedRunbook class in references/automation-toil.md executes command strings using subprocess.run with shell=True, which is a risky pattern that can lead to command injection if input is not strictly controlled.
[EXTERNAL_DOWNLOADS]: The skill relies on external libraries that are not part of the standard Python distribution.
Evidence: Import of the numpy library in references/automation-toil.md for capacity planning calculations.
Evidence: Import of the grafana_dashboard library in references/monitoring-alerting.md for dashboard generation.
[PROMPT_INJECTION]: The skill implements an automated decision-making loop based on external data, creating a surface for indirect prompt injection.
Ingestion points: The get_error_rate function in SKILL.md fetches and parses JSON data from a remote Prometheus API (http://prometheus:9090).
Boundary markers: No specific delimiters or warnings are used to instruct the agent to ignore potentially malicious content within the monitoring data.
Capability inventory: The skill possesses extensive system-level capabilities, including the ability to restart deployments (kubectl rollout restart) and modify network rules.
Sanitization: The skill lacks sanitization or integrity verification for the data retrieved from the monitoring API before using it to trigger operational actions.

sre-engineer