drunkrhin0/antislop

Antislop is a small project aimed at stripping the most recognizable AI writing patterns out of text. It does this through two skills you plug into an AI coding agent, such as Claude Code or opencode. Any agent that reads the SKILL.md or AGENTS.md convention can pick them up. The README frames the project as a late shower thought from a frustrated LinkedIn rant about machine-written copy. The two skills do different jobs. The first, called antislop, is a writing style skill that stays on in the background and quietly steers everything the agent writes away from common AI tells. The second, called antislop-audit, is a detection skill that takes a piece of text and returns a score out of 100, a violations table with severity and excerpt, and a plain summary of what to fix first. The README lists what the audit looks for. It flags a list of banned vocabulary, including filler verbs, inflated adjectives, and other words associated with AI prose, plus banned filler phrases like it's worth noting and at its core. It catches em-dashes used as authority props, scare quotes that hedge instead of commit, random bolding that decorates rather than emphasizes, and ambiguous bolded bullets where the body does not support the claim. It also flags the rule of three, synonym cycling, vague attributions, and chatbot artifacts such as I hope this helps or Great question. The score is split into bands. From 85 to 100 the text reads like a person, from 65 to 84 there is some slop that targeted edits will fix, from 40 to 64 a significant rewrite is needed, and below 40 the text reads like unreviewed AI output. The audit gives no exceptions based on the writer's intent. Installation is short. The simplest path is to run a single openskills command that pulls the skills from GitHub, but you can also ask the agent to install it directly, or copy the antislop and antislop-audit folders into your skills directory by hand. For chat tools that do not support skills, you can paste the SKILL.md contents at the start of a conversation. The project credits two earlier MIT projects, a Reddit thread on copywriting, and Wikipedia's signs of AI writing as inspiration, and is itself MIT-licensed.