crm-maintenance
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Considerations: The skill processes untrusted external data from email threads and calendar events to update HubSpot records.
- Ingestion points: In SKILL.md, the workflow steps involve reading email threads (subject, participants, and content) as well as calendar events (title, attendees, and descriptions).
- Boundary markers: There are no explicit text delimiters or specialized isolation instructions specified when parsing the external text data to prevent the agent from executing instructions embedded in emails.
- Capability inventory: The skill maintains capabilities to write, modify, and log activities within an external CRM platform (HubSpot), which includes creating contacts and updating fields like next steps or deal metrics.
- Sanitization: The reference documentation does not specify sanitization or escaping mechanisms for the ingested text prior to record modification.
- Mitigation: The architecture implements clear user approval gates requiring human confirmation before executing any structural cleanup modifications, which helps manage and mitigate unauthorized automated actions.
Audit Metadata