Mina — Terminal Companion

About Mina — Terminal Companion

You found a command on the web or an AI wrote one for you. Now what? Paste it into Mina. Before anything runs, you get a plain-language explanation of what the command does, what it changes, and whether that change can be undone. Nothing executes until you say so. That explanation is not an LLM guess. Mina's analysis is deterministic and runs entirely on your Mac: no network call, no model, nothing about your command leaves the machine. It parses the shell structure and matches against 679 built-in command definitions, and when it doesn't recognize a program, Mina says so instead of calling it safe. Placeholders like <API_KEY> and YOUR_USERNAME become labeled fields you TAB through. A value used four times is one editable field. Quoting and escaping are handled for you, including inside quoted strings, so you stop hunting through a wall of text to edit one argument. Mina is built for the moment your AI agent has to stop. When Claude, Codex, or a local agent hits a password, a login, an approval it cannot grant, or something it simply cannot do, it can hand the command to Mina through the bundled CLI. You type the secret locally. It never enters the model's context, and Mina's transport keeps it out of the process arguments where ps could read it. Mina also knows what your agent's flags mean: it recognizes the ones that turn the guardrails off, like --dangerously-skip-permissions, --yolo, and --sandbox danger-full-access, and makes you acknowledge them explicitly. Beginners get a Terminal that explains itself. Everyone else gets a review layer for the commands worth reading twice.