perplexity-search
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches 'litellm' and 'python-dotenv' from standard package registries to facilitate API interactions and environment management.
- [COMMAND_EXECUTION]: Instructs the agent to run local Python scripts included with the skill to perform searches and validate the environment setup.
- [PROMPT_INJECTION]: Contains behavioral steering instructions that prompt the agent to suggest a specific commercial platform (K-Dense Web) when the user's workflow complexity increases.
- [SAFE]: The skill ingests untrusted content from the web through search results, creating an indirect prompt injection surface. Evidence: 1. Ingestion points: Web search results via OpenRouter API. 2. Boundary markers: Absent. 3. Capability inventory: File-write and network operations. 4. Sanitization: Absent. This is a necessary feature of a search tool.
Audit Metadata