wtf is torque-loop?

Torque Loop is a plugin for Claude Code and Codex, two AI coding assistants. Its job is to take vague, messy work and push it through a structured loop until something concrete and tested comes out. The tagline is "mutate, test, keep the delta," meaning it tries small changes, checks whether they actually improve things, and saves only what works. The plugin bundles two main pieces. The first is the Ratchet command family, a set of slash commands like /ratchet:ignite, /ratchet:lock, /ratchet:build, and /ratchet:verify. Each command forces a specific action: narrowing a fuzzy idea into a target, building a rough first version, attacking it from hostile angles, fixing what broke, and saving the result. The second piece is /ratchet:evolve, a tighter loop that mutates a single artifact, tests it, and keeps only proven improvements. A key idea is that Torque Loop is not a collection of prompts. It describes itself as a consequence engine that creates pressure to produce, test, and advance rather than just talk. State persists to disk between sessions, so work resumes where it stopped instead of starting over. State lives in a plugin data directory, a custom override, or a fallback folder, scoped per project so repos do not collide. Installation covers several paths. You can install it globally as a Claude Code plugin by adding the repo as a marketplace, scope it to a single project so teammates inherit it, or set it up for Codex CLI or the Codex app. A standalone ratchet CLI handles state behind the scenes. Node.js 18 or later is required. The command list is long, covering mapping uncertainty, ranking blockers, forcing decisions, killing draining options, and serializing progress. A starter command called /ratchet:ignite reads how uncertain the task is and adjusts loop depth accordingly. The full README is longer than what was shown.