what is superloopy fr?

Superloopy is a command layer for Codex and Claude Code that turns a short instruction, typed as loopy followed by a task, into a guided loop where the AI agent plans the work, proves each step with a real file or command result, and only reports the task done once that evidence exists. The idea is to stop an agent from just claiming success in a confident sentence without anything to back it up. At its center is a distinction between two kinds of proof. Command-backed criteria are checked by literally rerunning the command again at the end and requiring it to still pass, so a stale or made-up result cannot slip through. Manual criteria, ones without a command to rerun, are instead checked by requiring a real evidence file to exist, with the final judgment left to a human or reviewing agent rather than an automatic rerun. Beyond the core loop, the project ships several optional add-on skills. One helps diagnose problems with the installation itself. Another runs a structured research process that produces a list of claims and a cited writeup. Others handle cloning an existing website with visual proof it matches, building frontend UI across web, desktop, and mobile targets, converting a presentation into a single web page, and rewriting Korean text to sound less like AI output. For bigger tasks, it can also hand off pieces of work to six optional named subagents, each responsible for one lane of the job. The tool itself is lightweight: a single small command-line program with no runtime dependencies, and all its state kept inside a project's own .superloopy folder rather than anywhere external. It works with both the Codex and Claude Code coding assistants, and its README is available in English, Korean, Chinese, Japanese, and Spanish.