generating-mermaid-diagrams

Pass

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes two Python scripts that use the subprocess module for legitimate functionality. query-org-metadata.py executes Salesforce CLI (sf) commands to retrieve object metadata and record counts for grounding diagrams. mermaid_preview.py uses subprocess.Popen to manage its own lifecycle when running the preview server in background mode.
  • [EXTERNAL_DOWNLOADS]: The mermaid_preview.py script serves an HTML template for local diagram rendering that fetches the Mermaid JavaScript library from the official JSDelivr CDN (cdn.jsdelivr.net). This is a well-known service used for serving open-source libraries.
  • [DYNAMIC_EXECUTION]: The mermaid_preview.py script implements a local HTTP server that dynamically generates HTML to render diagrams. It initializes the Mermaid renderer with securityLevel: 'loose', which allows HTML rendering within diagram labels for visual formatting.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The query-org-metadata.py script retrieves schema information (object names, record counts, sharing models) from the active Salesforce organization.
  • Boundary markers: The skill provides instructions on how to use these outputs but does not define explicit delimiters for the metadata returned to the agent.
  • Capability inventory: The skill has the ability to write files to /tmp, execute CLI tools via subprocess, and host a local web server for previews.
  • Sanitization: The scripts parse structured JSON output from the Salesforce CLI and provide basic escaping for diagram code when generating the preview interface.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 29, 2026, 10:02 PM