"Ratchet," a 2-day-old compliance checker for AI coding agents, is the #15 fastest-growing repo in this run's GitHub star-velocity set

A two-day-old GitHub project called Ratchet just collected 411 stars at roughly 200 a day, making it the 15th fastest-growing repository in the current velocity snapshot. That pace matters because it signals a nerve touched, not just novelty buzz.

Ratchet is a compliance hook for AI coding agents. You configure budgets: how many files the agent can touch, how many dependencies it can add, how many lines it can change. The tool sits in the PostToolUse phase and checks every edit against those limits in real time. If the agent violates a rule, Ratchet reports the problem back to the session immediately, forcing the agent to course-correct before the pull request ships.

The author frames this as fixing the “open loop” problem. Teams write CLAUDE.md files or system prompts that say “don’t add dependencies without asking” or “keep changes under 50 lines.” The agent reads the rule, then ignores it three edits later because nothing enforces it. Ratchet closes that loop. It turns guardrails from suggestions into hard stops.

This maps directly to the kind of proof-of-done logic we already build into systems like the Omni Command Centre, where you want agents to follow policy without constant human babysitting. The difference between a rule written in a markdown file and a rule that actually fires mid-task is the difference between aspiration and control.

The velocity here tells you something practical. If 400 developers star a compliance tool in 48 hours, a lot of shops are already running agents in production and discovering that vibes-based governance does not scale. Ratchet is narrow, early, and specific. It does one thing. The fact that it is growing this fast means the thing it does was missing. You can review the original repo to see how the budgets and hooks are structured.