datahub-evals

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses pip install to install the vendor's own package, acryl-datahub-cloud. It follows best practices by recommending a specific version (2.1.4rc1) and suggesting the use of a Python virtual environment to prevent dependency conflicts.
  • [COMMAND_EXECUTION]: The skill executes the acryl-datahub-cloud CLI and the claude agent binary. These tools are used for their intended purpose: managing evaluation data and running sub-agents to answer questions.
  • [INDIRECT_PROMPT_INJECTION]: The skill acknowledges that evaluation questions fetched from DataHub are untrusted text. It proactively mitigates the risk of indirect prompt injection by instructing the runner to use --strict-mcp-config and --allowedTools to narrow the tool surface available to the answering agent.
  • [DATA_EXFILTRATION]: While the skill communicates with DataHub, it uses established authentication methods (environment variables or config files) and only sends evaluation results. No unauthorized data exposure or exfiltration patterns were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 07:07 AM
Security Audit — agent-trust-hub — datahub-evals