kickoff
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches and processes content from external, untrusted URLs to infer product details and audience segments. This creates a surface for indirect prompt injection attacks where a website could contain hidden instructions designed to manipulate the agent. The skill implements a mandatory 'Input Hygiene' step (referencing
../shared/references/input-hygiene.md) to detect and remove instruction-style sentences and PII from the ingested content. - [EXTERNAL_DOWNLOADS]: The skill automatically downloads images (logos and product photos) from user-provided URLs to the
.agents/assets/source/directory. While this is a core functionality for brand onboarding, it involves downloading content from external web resources based on potentially untrusted input.
Audit Metadata