crm-automation

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from external sources and using it within AI-driven workflows.
  • Ingestion points: External data enters the system through webhook triggers (e.g., event: form_submission in SKILL.md), CRM event triggers (hubspot_deal_updated), and integration triggers (calendly.booking_created, linkedin.connection_accepted).
  • Boundary markers: The templates lack explicit boundary markers or instructions to the AI to ignore instructions potentially embedded within the lead data or CRM records (e.g., in the Compare these contacts prompt).
  • Capability inventory: The skill utilizes tools to create and update records in HubSpot, Salesforce, and Pipedrive, and has the capability to trigger email sequences and post to Slack notifications.
  • Sanitization: No specific sanitization, validation, or escaping of external lead data is defined before it is interpolated into AI prompts for lead scoring or deduplication logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 06:06 PM
Security Audit — agent-trust-hub — crm-automation