no-connect-multitouch
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill hardcodes a roster of employees with their full names, corporate email addresses, HubSpot user IDs, and Nooks internal identifiers. It also hardcodes a specific Nooks server UUID (
8c1d2738-2529-4248-b268-342859275d11). While used for authentication logic, hardcoding these identifies specific internal infrastructure and organizational roles. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external CRM sources to generate outbound communications, creating a path for untrusted data to influence agent behavior.
- Ingestion points: Prospect and account data retrieved via
hubspot_get_contact,hubspot_search_contacts, and NookslistTasks/listCalls(referenced inSKILL.md). - Capability inventory: The skill has the ability to write to the user's environment by using
create_draftin Gmail (referenced inSKILL.md). - Boundary markers: None; the instructions do not specify delimiters or warnings to ignore instructions embedded within the CRM prospect data.
- Sanitization: The skill uses
check_my_copyfor brand voice alignment, but lacks specific sanitization or filtering for injection attacks within the ingested content.
Audit Metadata