audience-behavior-study
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from the public web.
- Ingestion points:
Step 2 — Researchandreferences/research-playbook.mdutilizeWebFetchto retrieve content from external platforms like Reddit, X/Twitter, and Hacker News. - Boundary markers: The instructions do not define specific delimiters or guidance to ignore potential instructions embedded within the fetched external data.
- Capability inventory: The skill utilizes
WebSearch,WebFetch,FileRead, andFileWrite(report generation). - Sanitization: No explicit requirement for sanitizing or filtering the fetched content for malicious instructions before processing is mentioned.
- [SAFE]: The skill implements safety guidelines to prevent common data abuse patterns.
- It explicitly prohibits collecting PII beyond public handles and forbids automated outreach or scraping that violates terms of service.
Audit Metadata