The Agent Skills Directory

[SAFE]: The skill utilizes placeholder domains (e.g., company.com) and environment variables (e.g., $DB_HOST, $DB_USER, $MIGRATION_VERSION) for infrastructure interaction. This adheres to security best practices by avoiding hardcoded credentials or environment-specific secrets within the instructions.
[SAFE]: The runbook templates reference multiple well-known services and trusted organizations for incident management, monitoring, and documentation, including Grafana, Sentry, Stripe, PagerDuty, Google, and Atlassian. These references are documented neutrally and are appropriate for the skill's stated purpose of site reliability engineering.
[SAFE]: The skill includes infrastructure management commands (using kubectl, psql, and curl) and local script executions (e.g., ./scripts/smoke-test-payments.sh) that are consistent with the primary use case of incident response and service maintenance.
[INDIRECT_PROMPT_INJECTION]: The skill describes procedures for reading and processing untrusted data, specifically application logs via kubectl logs.
Ingestion points: Application logs are ingested via kubectl logs -n payments -l app=payment-service --tail=500 in the Partial Failures section of Template 1.
Boundary markers: None are present in the provided templates.
Capability inventory: The skill uses kubectl for resource management, psql for database operations, and curl for network requests.
Sanitization: No specific sanitization or filtering of log content is performed before the agent processes it.
Conclusion: While this constitutes a potential ingestion surface for indirect prompt injection, it is the standard and necessary practice for the skill's primary purpose of log analysis and incident triage, and thus does not escalate the verdict.

incident-runbook-templates