product-deep-research
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill demonstrates safe credential management by instructing the agent to use the '$SCAVIO_API_KEY' environment variable for authentication rather than hardcoding secrets.
- [SAFE]: Network communication is restricted to the vendor's official domains (mcp.scavio.dev, api.scavio.dev) and well-known, trusted retailers and social services (Amazon, Google, Reddit, etc.) for its primary research purpose.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it retrieves and summarizes content from numerous untrusted third-party sources.
- Ingestion points: Procedures in 'SKILL.md' (Steps 4, 5, 6, 7) fetch data from various social platforms and arbitrary URLs via an 'Extract' tool.
- Boundary markers: Absent; the instructions do not explicitly tell the agent to ignore instructions embedded in the external data.
- Capability inventory: The skill is limited to making network API calls to the Scavio service and presenting a report; it lacks dangerous capabilities like shell access or file system modification.
- Sanitization: Absent; there is no mention of filtering or sanitizing the retrieved text before processing.
Audit Metadata