ecommerce-content-marketing
Pass
Audited by Gen Agent Trust Hub on Apr 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill is designed to fetch and analyze content from external, untrusted sources including Amazon reviews, Reddit discussions, and social media platforms. Content from these sources could be crafted by an attacker to include hidden instructions that influence the agent's behavior. * Ingestion points: The workflow in
SKILL.md(specifically Step 2: Customer Insight Mining and Step 4: Competitor Content Analysis) relies onweb_searchandweb_fetchto retrieve data from external sites. * Boundary markers: The skill does not instruct the agent to use specific delimiters or to ignore instructions embedded within the fetched data. * Capability inventory: The agent uses search and fetch tools to collect and process this information. * Sanitization: The instructions do not define any sanitization or validation steps for the external data before it is processed by the agent.
Audit Metadata