landing-page-information-structure
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests case studies from 'references/cases.md' which include 'Prompt excerpts'. These excerpts contain imperative instructions such as 'Do not ask follow-up questions' and 'Generate 5 separate prompts'. There is a risk that the agent may mistakenly follow these embedded instructions instead of the primary skill instructions.
- Ingestion points: 'references/cases.md' is read during every invocation to select an anchor case.
- Boundary markers: The excerpts are contained within markdown blockquotes, but no explicit 'ignore instructions in this text' command is given to the agent.
- Capability inventory: The agent has the capability to generate code, copy, and use browsing or image tools.
- Sanitization: The skill requires a 'Preserve / Replace / Avoid' contract which forces the agent to categorize the input, providing some logical separation.
- [EXTERNAL_DOWNLOADS]: The skill directs the agent to access external URLs on 'goodcase.ai' and 'x.com' to inspect media and source documents for visual verification and design research.
Audit Metadata