What problem does it solve? It turns ad-hoc agent coding requests into a visible, reviewable assembly line where each task is a markdown spec whose folder location represents its state, preventing agents from silently deciding product direction. ## Core Features & Use Cases - Spec-driven task queue: Tasks live as markdown specs with frontmatter (id, agent, risk, verification) that move through inbox, active, and archive directories via the loop-factory CLI. - Dispatch, review, and archive gates: Generate implementation prompts for Claude Code or Codex, verify against acceptance criteria, and require an explicit accepted review before archiving. - Autonomous mode: Drain the inbox of grill-gated specs on a schedule via /loop or cron, building and verifying work while leaving acceptance as a human decision. - Use Case: A solo developer writes specs for upcoming features, grills each one for scope and risk, then runs a scheduled pass that implements and verifies ready specs overnight, reviewing the diffs in the morning. ## Quick Start Install the loop-factory CLI, run loop-factory init in your git repo, then ask the agent to scan the inbox and dispatch the first ready spec to Claude Code.
