product-marketing-context
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill features an 'Auto-draft' workflow that extracts information from external sources such as landing pages and 'About Us' pages, as well as local project files like the README. This creates a potential vector for indirect prompt injection where malicious instructions embedded in those untrusted sources could be incorporated into the foundation context file used by over 60 other marketing skills.
- Ingestion points: Instructions in SKILL.md Step 1 mandate reading the 'README, landing page, trang Ve chung toi' and product descriptions.
- Boundary markers: The skill does not implement technical delimiters (like XML tags) for the ingested data, but it includes a procedural mitigation requiring a 'user review + edit' step before the file is finalized.
- Capability inventory: The agent writes the synthesized data into a persistent markdown file at
.agents/product-marketing-context.mdwhich serves as the 'source of truth' for the wider skill ecosystem. - Sanitization: No automated sanitization or filtering of the external content is performed; the system relies on manual verification.
Audit Metadata