perplexity-search
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFECREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious behaviors or suspicious patterns were found in the provided scripts or documentation.
- [CREDENTIALS_UNSAFE]: The skill manages OpenRouter API keys using environment variables and a local .env file template, which is the recommended practice for local development. Evidence:
scripts/setup_env.pyandassets/.env.example. - [EXTERNAL_DOWNLOADS]: The skill relies on LiteLLM, a well-known and widely used library for interfacing with LLM providers. Evidence:
SKILL.md. - [PROMPT_INJECTION]: The skill processes untrusted web search results, presenting a surface for indirect prompt injection. Ingestion point:
scripts/perplexity_search.py(vialitellm.completion). Boundary markers: None identified. Capability inventory: Script capabilities are limited to local search and file output; no dangerous system-level capabilities are exposed to the model's instructions. Sanitization: None; results are returned as provided by the model. This is characteristic of search-oriented functionality. Evidence:scripts/perplexity_search.py.
Audit Metadata