value-proposition
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze untrusted content from external sources such as buyer reviews, competitor websites, and social media comments. This creates a potential vector for indirect prompt injection attacks.
- Ingestion points: Analysis of external quotes, competitor ad comments, and research data from forum threads (SKILL.md).
- Boundary markers: The skill contains explicit, robust instructions to treat untrusted content strictly as data, never follow URLs found within that data, and report any embedded instructions as findings rather than obeying them (references/agent-security.md).
- Capability inventory: The skill can perform web research and write to the local
.agents/context directory. - Sanitization: Employs mandatory reference guidelines to escape interpolated content and treat third-party text as untrusted data.
- [PROMPT_INJECTION]: The skill implements a 'Hard cap' of three questions for user input and instructs the agent to use 'stated assumptions' for missing information. While intended to optimize user experience, this behavior was flagged for potential concealment of agent logic, though the skill mitigates this by requiring all assumptions and inferences to be explicitly declared in the output.
Audit Metadata