landing-page-information-structure-by-197q59i
Pass
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and analyze content from external websites (goodcase.ai, x.com) to derive design patterns and code structures.
- Ingestion points: Multiple external URLs in
references/cases.mdserve as source data for the agent's workflow. - Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious instructions embedded in the external content (e.g., within page metadata or video descriptions).
- Capability inventory: The agent is tasked with generating implementation-ready code (React, Vite, Tailwind), which could be manipulated if the external evidence sources are compromised.
- Sanitization: There is no mention of sanitizing or validating the data extracted from these external sites before it is interpolated into the code generation prompt.
- [EXTERNAL_DOWNLOADS]: The
references/cases.mdfile contains numerous links to external media and social media platforms. The agent is explicitly instructed to access these URLs to "inspect finished media" as a required step in the workflow. - [COMMAND_EXECUTION]: The skill generates implementation-ready frontend code (React, TypeScript, Framer Motion) based on patterns derived from untrusted external sources. Although the agent does not execute this code, the generated output could harbor malicious logic if the agent's source material is intentionally deceptive.
Audit Metadata