analyzing-customers
Pass
Audited by Gen Agent Trust Hub on Mar 22, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted web data which presents an indirect prompt injection surface.
- Ingestion points: The skill performs web searches on G2, Capterra, Amazon, Reddit, and Twitter to collect user reviews and feedback (SKILL.md).
- Boundary markers: There are no explicit delimiters or specific instructions provided to isolate the fetched external content from the agent's core processing logic.
- Capability inventory: The skill is limited to using the WebSearch tool; no file-system access or remote code execution capabilities were detected in the skill definitions.
- Sanitization: The skill instructions do not specify any sanitization or filtering of the content retrieved from external sources before it is processed.
- [EXTERNAL_DOWNLOADS]: Fetches and analyzes data from well-known external web platforms.
- Evidence: The Research Methods section identifies G2, Capterra, and Amazon as sources for automated review mining (SKILL.md).
Audit Metadata