attribution
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references an external GitHub repository for tool integration guides and a central tools registry. These resources are located at
https://github.com/coreyhaines31/marketingskills, which is not an organization within the recognized trusted vendor scope. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from multiple external sources, including CRM systems (HubSpot, Salesforce), advertising platforms (Google Ads, Meta), and third-party booking webhooks (SavvyCal, Calendly, Stripe). This introduces a potential surface for indirect prompt injection where malicious instructions embedded in marketing data or customer records could attempt to influence agent behavior.
- Ingestion points: Webhook payloads (SavvyCal/Calendly), CRM lead records (HubSpot/Salesforce), and platform-reported conversion data.
- Boundary markers: The skill suggests a structured 'Attribution Readout' format for reporting and emphasizes 'Anonymity guards' to prevent PII leakage.
- Capability inventory: The skill provides JavaScript code snippets and implementation instructions for users to deploy. The agent itself does not possess direct file-write or network-access tools in the provided instructions, but assists the user in configuring them.
- Sanitization: The instructions explicitly include email normalization (trimming and lower-casing), identity verification logic to prevent PII smuggling (isAnonymousDistinctId), and guidelines for safe webhook diagnostics that avoid logging sensitive payloads or raw exceptions.
Audit Metadata