Is Stop Slop official Anthropic software?

No. It is an MIT-licensed personal project by Hardik Pandya, a design executive at Pocket FM who previously worked at Google and Atlassian. It does not live in the anthropics/skills repo.

Does it only work in Claude Code?

No. It is plain markdown, so the README covers Claude Code, Claude Projects, custom instructions, and pasting SKILL.md into any system prompt, and the launch post adds Cursor via .cursorrules.

Is it still maintained?

The last commit landed March 17, 2026, and 49 issues and PRs sat open as of September 1, 2026. The skill still functions — it is instructions, not code — but fixes and pattern additions have stalled.

Will it make my writing pass AI detectors?

It makes no such claim. It removes a catalogued set of phrases and structures readers associate with AI prose; none of its documentation mentions detector scores.

Skill review · by Hardik Pandya

Stop Slop Claude Skill

The stop slop skill for Claude is hardikpandya/stop-slop, a single SKILL.md plus three reference files that teach Claude Code, Claude Projects, or any LLM to strip recognizable AI writing patterns from prose. Hardik Pandya, SVP of Design at Pocket FM with earlier stints at Google and Atlassian, published it on January 11, 2026 under MIT; it sits at roughly 16,700 stars with 1,200 forks. The skill bans catalogued filler phrases, formulaic structures like "Not X. But Y.", every adverb, and every em dash, then adds a self-graded 50-point rubric that triggers a rewrite below 35. One thing to know before installing: the repo has had no commits since March 17, 2026, and carries 49 open issues and pull requests, so treat it as a finished opinion you copy into your skills folder, not maintained software. It is a community project, not an Anthropic release.

View the sourceMIT licensed · 16,666★ · facts checked September 1, 2026

By Robert MacKelfresh

Founder, Automation Squad · September 1, 2026

Install

1Clone the repo: git clone https://github.com/hardikpandya/stop-slop
2Claude Code, all projects: copy the folder to ~/.claude/skills/stop-slop (or .claude/skills/ inside one project). Claude then triggers it on drafting and editing tasks automatically.
3Claude.ai and Claude Desktop: zip the folder and upload it under Settings > Capabilities > Skills (paid plans).
4Claude Projects: upload SKILL.md plus the three references/ files to project knowledge.
5Anywhere else: paste SKILL.md into the system prompt, or into .cursorrules for Cursor — it is plain markdown with no code or dependencies.

What it does

Banned phrases: references/phrases.md catalogs throat-clearing openers, emphasis crutches, business jargon, vague declaratives, meta-commentary, and all adverbs for removal from drafts.
Structural rules: references/structures.md flags binary contrasts ("Not X. But Y."), negative listings, dramatic fragmentation, rhetorical setups, false agency, passive voice, and narrator-from-a-distance voice.
Sentence-level rules: No Wh- word sentence openers, no em dashes anywhere, no staccato fragments, no lazy absolutes ("always", "never"), and active voice with a human subject in every sentence.
Self-scoring rubric: Rates each draft 1-10 on directness, rhythm, trust, authenticity, and density; anything under 35/50 gets revised before delivery, backed by a 12-item pre-delivery checklist.
Before/after examples: references/examples.md holds worked transformations that Claude loads on demand when editing, per the progressive-disclosure skill format.

Common questions

Is Stop Slop official Anthropic software?: No. It is an MIT-licensed personal project by Hardik Pandya, a design executive at Pocket FM who previously worked at Google and Atlassian. It does not live in the anthropics/skills repo.
Does it only work in Claude Code?: No. It is plain markdown, so the README covers Claude Code, Claude Projects, custom instructions, and pasting SKILL.md into any system prompt, and the launch post adds Cursor via .cursorrules.
Is it still maintained?: The last commit landed March 17, 2026, and 49 issues and PRs sat open as of September 1, 2026. The skill still functions — it is instructions, not code — but fixes and pattern additions have stalled.
Will it make my writing pass AI detectors?: It makes no such claim. It removes a catalogued set of phrases and structures readers associate with AI prose; none of its documentation mentions detector scores.

Before you install

Unofficial. A personal project by one author; no Anthropic affiliation despite how directory listings frame it.
Stale at verification: last commit March 17, 2026 — about five and a half months before this review — with 49 open issues and PRs, several of them spam, and no maintainer responses visible.
The rules are absolutes, not settings: every adverb and every em dash is banned. That is one designer's taste, and an open PR reports that mechanical em-dash removal can introduce comma splices.
The SKILL.md trigger is broad ("drafting, editing, or reviewing text"), so once installed it can rewrite prose where you wanted the original voice unless you scope or rename it.
The 35/50 quality gate is the model grading its own output, not an external measurement.
No plugin or marketplace manifest ships in the repo; install is a manual copy. Listings on skill directories are third-party mirrors, not the author's.
It removes stylistic tells only. Nothing in the docs claims, or shows evidence of, beating AI-content detectors.

Facts come from the skill's own repository and docs on the date above, via the GitHub API — not from memory, and never from a sponsor.

Worth comparing

blader/humanizer — Siqi Chen's skill with the same de-AI-ify goal; larger (about 39,500 stars) and pushed more recently (August 2026).
stephenturner/skill-deslop — Same idea aimed specifically at scientific writing; about 380 stars.
adenaufal/anti-slop-writing — A universal anti-slop system prompt for Claude Code, Codex CLI, Gemini CLI, Copilot, and Cursor rather than a Claude-format skill; about 120 stars.

← All MCP servers & skills