education-cloud-multi-campus-configure

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (PDF, URL, CSV) which constitutes a potential attack surface for indirect prompt injection.
  • Ingestion points: Step 1 uses the Read tool for files and fetches content from user-provided URLs.
  • Boundary markers: Step 3 requires a mandatory visual confirmation of the parsed tree using a specific visualization template before any record processing occurs.
  • Capability inventory: The skill has the capability to write, update, and delete Salesforce records (Account, BusinessProfile, RecordType) via the headless-360 MCP tool and execute shell commands via the sf CLI.
  • Sanitization: The workflow enforces human verification at multiple stages (Steps 3 and 6) and prohibits silent data source substitution to maintain data integrity.
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to invoke the sf CLI (Salesforce CLI). This is a standard administrative tool for the vendor's ecosystem and is used here for legitimate environment verification and record management.
  • [EXTERNAL_DOWNLOADS]: The skill fetches data from remote URLs to parse organizational structures. It includes a specific safety constraint that prevents the agent from using alternative third-party sources (like search results) if the primary URL fails, ensuring that only user-authorized sources are processed.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 08:41 AM
Security Audit — agent-trust-hub — education-cloud-multi-campus-configure