The Agent Skills Directory

[SAFE]: The skill is entirely instructional and utilizes authorized monitoring tools to perform diagnostics. Analysis of the scripts and instructions reveals no evidence of data exfiltration, hardcoded credentials, or unauthorized command execution. The logic is strictly focused on interpreting Cassandra metrics via the Netdata MCP integration.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes external data (metrics, logs, and alerts) retrieved from Netdata. This is a characteristic of monitoring skills rather than a specific vulnerability.
Ingestion points: Data entering the agent context via MCP tool calls including query_metrics, list_metrics, and find_anomalous_metrics (referenced in SKILL.md and rule files).
Boundary markers: The instructions do not define explicit boundary markers or provide warnings to ignore instructions that might be embedded in metric metadata or alert descriptions.
Capability inventory: The skill's operational scope is limited to read-only diagnostic queries; it lacks tools for writing to the filesystem, executing arbitrary shell commands, or making outbound network requests to unknown domains.
Sanitization: No specific sanitization or filtering of the monitoring data is performed before it is presented to the agent for analysis.

troubleshoot-cassandra