Because the model is trained mostly on bash. On Windows, Cursor's agent runs commands in PowerShell, but the model still reaches for bash habits and Linux tools. Rules telling it to use PowerShell help, but don't stop it every time.
Cursor's own forum has many reports of it: agents ignore the rule to use PowerShell commands and default to Linux, endless syntax errors, and agents that fail the first command they try with bash syntax. Cursor staff suggest adding a rule, while noting it won't guarantee a fix.
.cursor/rules asking for PowerShell commands on Windows.On Windows, Cursor runs commands in PowerShell, so Check has PowerShell's own parser read each command before it runs. A program the command calls that isn't on your machine is stopped, and so is a command that doesn't parse. The agent gets PowerShell's own words back, so the next try is a fix, not a guess.
Cursor has no hook before a file edit, so in Cursor, Check checks commands. When Check can't be sure, for example a command that loads a module or defines a function, it lets the command run.
100 free blocks, then A$0.10 for each action Check stops, from a prepaid balance. Allowed actions are always free. No subscription. Windows and Linux; macOS isn't supported yet. Get your key at the console, then run this in your own terminal:
npx @golproductions/check@latest --installOr add Check as a plugin for Claude Code, Gemini CLI, Copilot CLI or Codex: the guide. Everything Check checks, and what it lets through: What Check does.
100 free blocks. Get your GOL API Key at the console.