perplexity-search
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface because it retrieves and handles unstructured text data from external web sources.
- Ingestion points: Real-time web data and third-party content are ingested into the skill environment within
scripts/perplexity_search.pyvia Perplexity API search responses. - Boundary markers: Absent. No specific delimiters or protective prompting instructions are used to separate untrusted search results from agent instructions.
- Capability inventory:
scripts/perplexity_search.pycontains local file write operations when utilizing the--outputparameter to save search results, but lacks arbitrary shell execution or dynamic execution capabilities. - Sanitization: Absent. The text retrieved from the external API is handled, printed, or saved directly without validation, filtering, or escaping.
Audit Metadata