troubleshoot-pgbouncer

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests external data from a Netdata instance via MCP tools such as list_nodes, query_metrics, and list_metrics to perform diagnostics. This constitutes an indirect prompt injection surface where crafted metric metadata could theoretically attempt to influence agent behavior.
  • Ingestion points: Metric names, dimension labels, and node metadata retrieved through MCP function calls in SKILL.md and rules/overview.md.
  • Boundary markers: Not present; the instructions do not implement specific delimiters or 'ignore' instructions for the retrieved metric strings.
  • Capability inventory: The skill is restricted to information retrieval and diagnostic analysis; it lacks capabilities for arbitrary shell execution, file system modification, or unauthorized network exfiltration.
  • Sanitization: No explicit validation, escaping, or filtering of the external metric data is performed within the instruction set.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 11:24 AM
Security Audit — agent-trust-hub — troubleshoot-pgbouncer