turbo

SUSPICIOUS. The skill’s purpose and requested credential are broadly coherent with hosted code generation, and the implied upstream service matches official Cerebras documentation. However, the key execution path goes through an unverified MCP wrapper that both forwards prompts/API credentials and writes remote-generated code directly to disk, while explicitly hiding the generated code from Claude. That reduced transparency and combined network-plus-write behavior make the skill medium risk despite no clear evidence of malware.