skills/grafana/gcx/agento11y/Gen Agent Trust Hub

agento11y

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes gcx agento11y commands to interact with Grafana's Agent Observability service. These commands facilitate listing conversations, managing evaluators, and setting up evaluation rules.- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for creating llm_judge evaluators that ingest untrusted data from LLM generations.
  • Ingestion points: Generations containing untrusted request/response data are retrieved via gcx agento11y generations get and gcx agento11y conversations search as described in SKILL.md.
  • Boundary markers: The evaluator templates in references/evaluator-examples.md interpolate {{assistant_response}} and {{latest_user_message}} directly into the judge's prompt without using protective delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill uses the Bash tool to run CLI commands and the Write tool to manage local YAML manifests.
  • Sanitization: There is no evidence of sanitization or content filtering for the ingested generations before they are processed by the LLM judge.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 02:33 PM
Security Audit — agent-trust-hub — agento11y