stream-aggregation-helper

Pass

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill's instructions are functional and intended to guide the agent through technical phases. No behavior-overriding patterns or jailbreak attempts were found. The skill maintains an attack surface for indirect prompt injection by processing external data from VictoriaMetrics APIs.
  • Ingestion points: Data is ingested from user-provided endpoints via curl in Phase 2a (TSDB status and query APIs).
  • Boundary markers: None present in the instructions to delimit or ignore instructions within the API response.
  • Capability inventory: The skill uses curl and jq to fetch and parse external metrics and configuration data.
  • Sanitization: No explicit sanitization of input data is performed before interpretation by the agent.
  • [DATA_EXFILTRATION]: The skill handles authentication securely using environment variables (VM_AUTH_HEADER) rather than hardcoded secrets. Network communication is restricted to the user-defined $VM_METRICS_URL, and the skill performs standard analysis operations consistent with its stated purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 29, 2026, 07:06 AM
Security Audit — agent-trust-hub — stream-aggregation-helper