web-search
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
infshCLI to run search and extraction apps on the inference.sh platform. The allowed tools are restricted to the vendor's specific binary. - [EXTERNAL_DOWNLOADS]: The skill references and suggests installation of the
inference-shCLI and related tools from the official vendor repository. These are verified vendor resources. - [DATA_EXFILTRATION]: User queries and extracted URLs are sent to external search providers (Tavily and Exa) as part of the core functionality. This is expected behavior for search-related tasks.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection when extracting content from untrusted websites. Ingestion points: web page content via
tavily/extractandexa/extract(SKILL.md). Boundary markers: None present to delimit extracted content. Capability inventory: Execution of platform apps viainfsh(SKILL.md). Sanitization: No explicit sanitization of extracted web content is defined.
Audit Metadata