The Agent Skills Directory

[SAFE]: The skill performs legitimate web search operations using well-known and trusted services including OpenRouter and the LiteLLM library. All external references target reputable development and AI platforms.- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests untrusted data from the web via search results. 1. Ingestion points: Data entering the agent context via the answer and citations fields in scripts/perplexity_search.py. 2. Boundary markers: Absent; results are returned without specific delimiters to distinguish search content from instructions. 3. Capability inventory: The skill itself lacks dangerous write or execution capabilities, though the impact depends on other tools available to the agent. 4. Sanitization: No content filtering or instruction scrubbing is performed on the external search results.- [SAFE]: Credential management is handled through recommended practices. The scripts/setup_env.py script helps users configure an environment file safely, and the core search script masks the API key during configuration checks.

perplexity-search