skills/metabase/agent-skills/rde/Gen Agent Trust Hub

rde

Pass

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from various untrusted sources, including warehouse tables, user-uploaded CSV files, and external business logic artifacts such as code, spreadsheets, or documents (as seen in playbooks/extract-business-logic.md).
  • Ingestion points: Untrusted data enters the agent's context through tools like mb query, mb table get, and the WebFetch tool used for fetching business logic artifacts.
  • Boundary markers: The skill uses instruction-based constraints (e.g., 'read nothing else', 'follow it') and structured JSON delimiters for command inputs, though explicit safety boundary markers for data interpolation are not consistently applied across all playbooks.
  • Capability inventory: The agent has access to Bash, Write, Edit, and WebFetch, allowing it to execute commands, modify the filesystem, and make network requests based on its analysis of potentially malicious data.
  • Sanitization: The skill emphasizes using the --json flag and jq for parsing CLI outputs, which helps in structured data handling, but the natural language processing of external documents lacks formal sanitization.
  • [DYNAMIC_EXECUTION]: The skill dynamically generates an executable bash script (./.scratch/q) to facilitate warehouse queries.
  • Evidence: In references/state.md, the skill uses cat > ./.scratch/q <<'SH' and chmod +x ./.scratch/q to create and set permissions for a helper script that wraps mb query calls. This script is then executed throughout the playbooks.
  • [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to interact with the system and the mb (Metabase) CLI. This includes creating directories (mkdir -p ./.scratch), managing files, and executing complex query commands.
  • [EXTERNAL_DOWNLOADS]: The skill recommends the installation of external software and scripts.
  • Evidence: README.md and SKILL.md suggest installing @metabase/cli via npm and adding the skill itself using npx skills add metabase/agent-skills. These resources are provided by the vendor (Metabase) and are documented neutrally.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 8, 2026, 10:48 PM