rudder-profiles-project
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is authored with a strong security posture. It incorporates five mandatory confirmation gates (selecting connections, tables, defining IDs, entity features, and final YAML review) that ensure human-in-the-loop control over the project creation process.
- [DATA_EXPOSURE]: The instructions include specific safeguards against credential exposure, explicitly forbidding the agent from requesting secrets in chat or printing warehouse credentials found in command outputs or configuration files.
- [COMMAND_EXECUTION]: The skill utilizes a specialized tool (
pb) for warehouse operations. It follows a least-privilege logic by performing validation and compilation checks before offering a pilot run, and restricts tool use to groundings discovered through MCP tools. - [INDIRECT_PROMPT_INJECTION]: The skill explicitly treats external warehouse metadata and tool outputs as untrusted. It includes instructions to extract only specific expected fields (e.g., table and column names) and to double-check all values against the discovered state before using them in project files, mitigating potential injection via malicious metadata.
Audit Metadata