The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as Grafana dashboard JSON models and server logs, which constitutes a potential indirect prompt injection surface. This is addressed through extensive defensive instructions.
Ingestion points: The skill ingests dashboard JSON models, alert rules, and server logs for diagnosis and engineering as detailed in references/dashboard-engineering.md and references/troubleshooting.md.
Boundary markers: The skill relies on instructional boundaries, requiring agents to treat resources as separately authorized and to verify resource provenance before acting, rather than using technical delimiters.
Capability inventory: Operational tools are restricted to standard management utilities including the grafana CLI and curl for API health and discovery.
Sanitization: Instructions mandate the redaction of sensitive values (tokens, passwords, secureJsonData) and require manual verification of query correctness against live data sources to mitigate exploitation risks.

grafana