amazon-returns-recovery
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill utilizes a fixed JavaScript template to override the window.open behavior in the browser. This modification ensures that live chat sessions remain within the agent's tracked context rather than opening in a new window, which is a necessary adaptation for its execution environment.
- [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading Amazon order histories and live chat transcripts. These external inputs could potentially harbor malicious instructions. The skill addresses this vulnerability surface through its 'Iron Law' and 'Boundaries' sections, which mandate explicit user confirmation for every action and require a report of findings to the user before any interaction with Amazon support occurs.
Audit Metadata