nl2-nrql

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted metadata such as event names, attribute names, and sample values retrieved from a New Relic account. The instructions explicitly mitigate this risk by defining discovery output as data rather than instructions and directing the agent to ignore embedded directives.
  • Ingestion points: Output from the execute_nrql_query tool during discovery steps (e.g., SHOW EVENT TYPES, keyset(), uniques()) as described in SKILL.md and skill-manifest.md.
  • Boundary markers: The skill contains a specific 'Untrusted input' section in SKILL.md that warns the agent to treat discovery results as hostile text.
  • Capability inventory: Access is limited to read-only New Relic MCP tools: execute_nrql_query, list_available_new_relic_accounts, and convert_time_period_to_epoch_ms.
  • Sanitization: The agent is instructed to quote names and ignore any strings that resemble instructions within the returned data.
  • [SAFE]: The skill does not perform any shell command execution, dynamic code evaluation, or persistence-related activities. It is composed entirely of Markdown instructions and reference files.
  • [SAFE]: Network access and tool usage are restricted to the New Relic MCP environment over authenticated channels. No non-whitelisted domains or unauthorized exfiltration paths were detected.
  • [SAFE]: The skill enforces least-privilege principles by using only the observability:read scope and requiring explicit Account IDs and questions before proceeding, preventing speculative tool usage.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 12:18 PM