Install
Add this folder as a skill.Who it is for
If you want an agent to check the boilerplate, structure, and rhythm that jump out in prose—instead of rewriting large stretches first—Stop Slop fits. Officially it is “remove AI traces from prose”: it teaches the model to catch predictable wording, structure, and rhythm.
The first pass is better as diagnosis only. Officially it scores Directness, Rhythm, Trust, Authenticity, and Density from 1 to 10 each, and a rewrite is needed only below 35/50. If you only want to polish tone and do not accept sentence rules such as “ban all adverbs” or “no dashes”, read the limits first, then decide.
The public page writes usage for Claude Code, Claude Projects, custom instructions, and the API. The first try uses Claude Code only: add this folder as a skill. Do not copy a second set of core rules in parallel, and do not read this roster as an exclusive support surface.
First try in 3 minutes
Use only this official install note: add the stop-slop folder as a skill.
Add this folder as a skill.
Prerequisite: you already use Claude Code, and you have a short passage that can be scored, not an empty repository. This page does not write a concrete skill-directory path; do not turn it into skills add or a copy of skills/stop-slop on your own.
After install, do a check that does not change the original: confirm the agent can see SKILL.md and the phrases, structures, and examples under references. That is the install-success signal.
Then give it a short passage and ask only for the five scores and whether the total is below 35/50. Do not let it overwrite files yet. Seeing the five scores and whether a rewrite is needed is a successful first try.
If the total is below 35/50, have it name the boilerplate it will drop and the sentences it will change; allow the rewrite only after you confirm. The public material does not give a command for auto-save or bulk repo edits.
Usage examples
Example 1: Score first, then decide whether to change. Give the agent an existing draft and score Directness, Rhythm, Trust, Authenticity, Density from 1 to 10 each. Below 35/50, go to rewrite; otherwise stop at the score. This is a reversible check. The public notes give no standard sample essay; the score is only responsible for the text you provided.
Example 2: Clean a draft you have already decided to change. After you confirm a rewrite is needed, have the agent work against banned wording, structural boilerplate, and sentence rules: drop throat-clearing openings, emphasis crutches, business boilerplate, vague assertions, and meta-comments; avoid binary contrasts, dramatic fragments, fake-outs, and passive voice; and follow sentence rules such as no Wh- openings, no dashes, and active voice. Have it name the sentences it will change, then replace after you confirm. The public material points the details to references/phrases.md and references/structures.md, but does not write them as a second install path.
Limits
This skill checks and rewrites prose traces. It is not a grammar proofreader and not a general translator. The sentence rules are hard: ban all adverbs, no dashes, require active voice. If your style depends on those devices, do not treat this as a must-follow publishing standard.
Install is only “add this folder as a skill”. The page also writes uploading SKILL.md and the reference files to Claude Projects, copying SKILL.md core rules into custom instructions, and putting SKILL.md into the system prompt on API calls; those are other uses, not the main path for the first try. Do not write Claude Code as the only usable client.
A rewrite is needed only below 35/50; a rewrite changes the original and must be confirmed first. The public material has no uninstall steps and no copyable command-line installer.
Repository address:
https://github.com/hardikpandya/stop-slop