The stop slop skill: install, how it works, alternatives
Stop Slop is the most-starred skill for removing AI tells from prose (about 18.9k GitHub stars when we checked). Here's what's in it, how to install it in each tool, where it's strong, where it isn't, and what to use alongside it.
Get stop-slop on GitHub Check its output for free
What's inside
| File | What it does |
|---|---|
SKILL.md | Core instructions and sentence rules: no Wh- starters, no em dashes, no staccato fragments, no lazy extremes, active voice. |
references/phrases.md | Banned phrases: throat-clearing openers, emphasis crutches, business jargon, all adverbs, vague declaratives, meta-commentary. |
references/structures.md | Structural clichés: binary contrasts, negative listings, dramatic fragmentation, rhetorical setups, false agency, narrator-from-a-distance, passive voice. |
references/examples.md | Before/after rewrites. |
Scoring: 1–10 on directness, rhythm, trust, authenticity and density; revise anything under 35/50.
Install
Claude Code
git clone https://github.com/hardikpandya/stop-slop ~/.claude/skills/stop-slop
Restart Claude Code. Ask: "apply stop-slop to this draft".
Codex
git clone https://github.com/hardikpandya/stop-slop ~/.codex/skills/stop-slop
Claude Projects / claude.ai
Upload SKILL.md and the three reference files to the project's knowledge, as the README describes.
ChatGPT, API, other tools
Copy the core rules from SKILL.md into custom instructions or your system prompt. Reference files can be added when needed.
Where it's strong, where it isn't
- Strong: clear ban lists, good before/after examples, a rubric that forces a revision pass. It works well for essays, newsletters and posts.
- Strict by design: banning all adverbs, passive voice and Wh- starters will fight you in technical docs, academic writing and legal text. Loosen those rules if they don't fit your genre.
- Dated in places: the last update was March 2026. It doesn't cover tells that arrived with GPT-6-era models, such as "rather than" corrective disclaimers or the new vocabulary in Wikipedia's 2026 list.
- Self-graded: the agent scores its own output. An independent check catches what it misses, which is what the AI slop detector is for.
Free alternatives and companions
| If you want… | Use |
|---|---|
| Countable caps and a one-line tally for agent pipelines, PRs, commit messages | anti-slop skill (gounslop, free) |
| A rewrite that reads naturally, with 2026 tells, without banning whole word classes | humanizer skill (gounslop, free) |
| Cleanup of AI-written code, not prose | deslop skill |
| An independent score for any text, no agent needed | AI slop detector |
| A finished rewrite of someone else's text | unslop tool |
You can run stop-slop and still check the result: paste the agent's output into the detector and see what slipped through.
The skill and checklists are free. For finished text from ChatGPT or Claude, the unslop tool rewrites with a slop score. After the 3 free daily rewrites: $5 for 50 or Pro $12/month.
FAQ
What is the stop slop skill?
Stop Slop is an open-source Agent Skill by Hardik Pandya (github.com/hardikpandya/stop-slop, MIT license) for removing AI tells from prose. It's a SKILL.md plus three reference files: phrases.md (phrases to remove), structures.md (structural patterns) and examples.md (before/after rewrites).
How do I add the stop slop skill to Claude?
Claude Code: git clone https://github.com/hardikpandya/stop-slop ~/.claude/skills/stop-slop and restart. Claude Projects: upload SKILL.md and the reference files to project knowledge. Elsewhere: paste the core rules from SKILL.md into custom instructions or your system prompt.
How does stop slop score writing?
It rates text 1–10 on five dimensions (directness, rhythm, trust, authenticity, density) and says to revise anything under 35/50.
Is stop slop free?
Yes. It's MIT-licensed: use, modify and share it. Credit Hardik Pandya, the author.
What are the limits of stop slop?
Some rules are strict on purpose (it bans all adverbs, Wh- sentence starters, em dashes and passive voice), which can flatten technical or academic writing. When we checked on November 13, 2026, the last commit was from March 2026, so it predates GPT-6-era tells like "rather than" corrective disclaimers. It also has no automatic checker; the agent scores itself.
What's the best stop slop alternative?
It depends on the job. For agent output in repos, our free anti-slop skill uses numeric caps and a re-count. For rewriting prose to sound natural, the humanizer skill follows Wikipedia's guide plus 2026 tells. To check any text without an agent, use the free AI slop detector.