The Agent Skills Directory

[SAFE]: No security issues detected. The skill performs legitimate web-search operations through a reputable third-party API (OpenRouter).
[CREDENTIALS_UNSAFE]: The skill correctly handles credentials by requiring the user to set an OPENROUTER_API_KEY environment variable. It does not hardcode any secrets and provides a clear setup guide using placeholders.
[COMMAND_EXECUTION]: The scripts/search wrapper and scripts/search.mjs utilize standard Node.js and Bash patterns for script execution and argument parsing. It does not use eval() or exec() with untrusted input.
[EXTERNAL_DOWNLOADS]: The skill makes a network request to https://openrouter.ai/api/v1/chat/completions, which is a well-known service for LLM API access. This is documented neutrally as a functional requirement.
[DATA_EXFILTRATION]: No sensitive data access or exfiltration patterns were identified. Network operations are limited to the intended API provider for processing user queries.
[PROMPT_INJECTION]: The query building logic (buildSotaQuery, buildRecentQuery) uses template literals to wrap user input with research instructions. While this creates a surface for indirect prompt injection, it is used for functional steering and is considered a low-risk, standard practice for LLM agent skills.

perplexity-search