victoriametrics-cardinality-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from VictoriaMetrics status endpoints, presenting a risk of indirect prompt injection if external clients populate high-cardinality labels or error messages with malicious payloads.
- Ingestion points: Data enters the agent's context from the
/api/v1/status/tsdb,/api/v1/status/metric_names_stats, and/api/v1/rulesAPI endpoints invoked via curl commands inSKILL.md. - Boundary markers: Absent; the raw strings and structures returned from the time-series database are rendered into the final markdown report without explicit delimiters or behavioral isolation constraints.
- Capability inventory: Uses
Bash(curl:*)andBash(jq:*)across multiple workflow phases to pull and filter time-series metadata. - Sanitization: Absent; no sanitization, escaping, or structural validation is conducted on label values or query text before appending them to the report templates.
Audit Metadata