anti-slop

What problem does it solve? Coding agents that repeatedly extend their own code accumulate structural rot — god-functions grown one branch at a time, copy-paste duplication, and dead code from abandoned iterations — while every individual edit still passes tests. Research (SlopCodeBench) shows quality-aware prompting does not stop this degradation; only a measured gate with a failing exit code does. ## Core Features & Use Cases - Measured quality gate: The slopcheck.py script computes structural erosion (complexity mass via CCN×√NLOC), structural clone duplication, and dead-code counts across ~20 languages via lizard, then fails with exit code 1 when any gate is breached. - Ratcheting baseline with anti-gaming policy: A once-per-project baseline advances automatically on every passing check, refuses to be overwritten mid-task, and ignores looser tolerance flags unless a human sets an environment override. - Extend-vs-restructure decision rules: Reference guides with worked examples help decide when to restructure before adding a feature, plus a prose-slop sweep for AI-signature vocabulary in PRs and comments. - Use Case: An agent adds a third language to a CLI tool. Instead of grafting another elif into main(), it extracts a dispatch table, then runs slopcheck.py check . before declaring done — the gate confirms erosion and duplication did not rise. ## Quick Start Ask the agent to run the anti-slop baseline on your project, make its code changes, and then verify the quality gate passes with slopcheck before declaring the task done.