llm-models

[Skill Scanner] Pipe-to-shell or eval pattern detected This skill documentation is coherent: its capabilities (calling many LLMs via a single CLI/gateway) match its stated purpose and examples. There are two operational risks to note: (1) installation via `curl | sh` runs a remote installer without local review, and (2) all prompts and auth flow through the inference.sh/OpenRouter intermediary rather than directly to provider APIs, which centralizes access and increases exposure of prompts and tokens to that service. There is no sign of intentionally malicious code in the provided text (no obfuscation, hardcoded secrets, or remote exfiltration domains). I rate the package as largely benign but with moderate operational/trust risk because of the distribution/install pattern and the intermediary data flow. LLM verification: This documentation describes a legitimate-seeming CLI (infsh) that proxies access to many LLMs via inference.sh/OpenRouter. The file itself contains no explicit malicious code, hard-coded credentials, or obfuscated constructs. However, there are notable supply-chain and privacy risks: the provided installer invocation uses an unsafe pipe-to-shell pattern and the workflow routes user prompts and credentials through a third-party gateway without documenting retention or storage policies. Treat the