website-personalization

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill fetches data from external URLs and public sources to research companies and analyze touchpoints. This creates an attack surface where instructions embedded in external web content could potentially influence agent behavior. 1. Ingestion points: The agent is directed to 'fetch the site or page they named' and 'research the company yourself' using public sources. 2. Boundary markers: The instructions in SKILL.md do not define explicit boundary markers or delimiters for the untrusted content it retrieves. 3. Capability inventory: The skill writes research to local files ('.agents/product-context.md'), performs network fetches, and generates marketing copy for live environments. 4. Sanitization: Although the referenced security standards mention escaping, the skill's specific process lacks explicit sanitization steps for the data it processes.
  • [PERSISTENCE]: The skill is instructed to write discovered research and inferred data to '.agents/product-context.md'. This is a persistence mechanism intended to maintain state and share information between different skills within the environment.
  • [PROMPT_INJECTION]: The skill contains a directive to skip reporting on missing visual tools 'without comment'. This is a concealment pattern that prevents the user from being notified of tool failures or environmental limitations during execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:22 PM