The Agent Skills Directory

[SAFE]: The skill uses environment variables for secure API key management, supported by a dedicated setup script (scripts/setup_env.py) and a .env.example template. It also provides documentation on excluding sensitive files from version control using .gitignore.\n- [SAFE]: External dependencies are restricted to reputable, well-known packages such as litellm and python-dotenv. All network communication is directed to legitimate services (OpenRouter and Perplexity).\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a standard attack surface for indirect prompt injection due to its core function of retrieving external web content.\n
Ingestion points: Real-time web search results are ingested via the completion function in scripts/perplexity_search.py through the Perplexity API.\n
Boundary markers: The script returns search results as plain text strings without implementing specific delimiters to segregate potentially untrusted web content from agent instructions.\n
Capability inventory: The skill is capable of performing network requests and writing search results to the local filesystem if the --output parameter is specified.\n
Sanitization: There is no explicit sanitization or filtering of the content returned by the search API, as it is intended for research and informational purposes.

perplexity-search