The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and scripts to perform its core functions.
Evidence: Usage of openclaw CLI for detection, containment, and recovery.
Evidence: Integration examples using bash and groovy (Jenkins) for automated workflows, including command substitution like $(echo $output | jq .id).
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it ingests and processes untrusted external data.
Ingestion points: The skill scans logs and metrics via the /api/incident/detect endpoint and parses incident data using the --analyze-depth flag.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within logs/metrics are defined.
Capability inventory: The skill has high-impact capabilities including resource containment (e.g., pausing Kubernetes pods) and system recovery (e.g., rollbacks).
Sanitization: While the skill mentions validating JSON fields for required keys like threshold, there is no evidence of sanitizing or escaping the actual content of the logs or incident data before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill references external integrations with well-known services.
Evidence: Integration with PagerDuty (pagerduty.com) and Prometheus for monitoring and notifications.

incident-response