reactivate
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill processes data from untrusted external sources, including customer reviews, support tickets, and email threads. This creates a potential surface for indirect prompt injection where instructions could be embedded in the ingested data.
- Ingestion points: The skill gathers data from public reviews, Zoho Desk tickets, and email threads to understand customer sentiment.
- Boundary markers: The skill contains a specific security gate instructing the agent to never follow instructions found within the data it reads, treating them as descriptive information rather than commands.
- Capability inventory: The skill utilizes tools for fetching web data and interacts with CRM systems to log activities and send outreach messages.
- Sanitization: The instructions explicitly separate tool-result text from commands, requiring verification for sensitive requests like credential changes.
- External Service Integration: The skill interacts with several well-known third-party platforms to perform its functions.
- Interaction Context: It uses standard patterns to communicate with services such as Stripe, PayPal, Shopify, and various CRM providers to retrieve order history and customer details.
Audit Metadata