No AI Slop
github.com/petergyang/no-ai-slop- Category
- AI Agents
- Rank
- No. 2047Tools index
Previous survey · No. 2055 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Productivity & Collaboration
- Interfaces
- Agent Skill / Plugin
- Builder
- petergyang
- GitHub
- 12.1k stars
- Latest release
- v1.0.6
- Date
About
An installable agent skill that edits a piece of writing to remove more than 20 recurring AI-generated stylistic tics, such as the "It's not X, it's Y" construction and manufactured-insight openers. It is designed to strip those patterns without flattening the vocabulary, cadence, and imperfections that make the text sound like its author. It installs globally into Claude Code, Codex, or other coding agents via a natural-language prompt or `npx skills add`.
What it does
A set of plain-language editing instructions for an AI assistant. Paste a draft and it acts as a human editor: it finds named patterns of machine-sounding prose, makes the smallest fix that removes each one, then grades its own result against a pass/fail checklist and repairs anything that fails. You get back the full edited text plus a short note on what changed. A second mode only flags problems: it quotes each offending line and suggests a fix in a few words, without rewriting or guessing who wrote it.
Why it's ranked here
The rules are specific enough to apply mechanically. Each pattern comes with a before-and-after example, a banned-word list, and an explicit test: if a sentence could move unchanged to another product or company, it is filler. The detect mode refuses to score a draft or guess at AI authorship, and says named patterns are evidence a person can check. MIT licensed, with a separate self-check list of about thirty questions the assistant must pass before returning an edit.
What's good
It protects the writer as hard as it cuts. Phrases like "I think" or "to be honest" stay when they carry real uncertainty, and profanity, humor, digressions and long spoken sentences are kept when they belong to the author. It forbids inventing claims, stats or sources; when a vague attribution has no source, it asks instead. The flag-only mode is useful on its own, since it quotes lines rather than issuing a verdict.
Tradeoffs
It is a prompt, not software. Nothing in the repository counts patterns or verifies the output; the quality depends on how faithfully the host model follows the instructions and its own checklist. The readme advertises more than twenty patterns but lists ten, so the full set lives only in the instructions. The packaging script validates files for a ChatGPT and Codex plugin, but there is no test suite for the editing behavior itself.
How to use it well
Run it on drafts you wrote yourself and want to keep sounding like you: newsletters, posts, product copy. Use the flag-only mode first on a draft you are unsure about, then ask for the edit once you agree with the findings. If the audience or goal is unclear it will ask one question, so state who the piece is for up front. It does not fact-check, research, or draft from scratch beyond a satire mode that generates deliberately bad posts.
Technical notes+
The skill is two Markdown files: skills/no-ai-slop/SKILL.md holds the editing principles, a banned-word list, about twenty named patterns (binary contrasts, colon reveals, interpretive metadiscourse, negative listing, fake-profound kickers and others) and a six-step workflow whose step four runs skills/no-ai-slop/eval.md, a pass/fail checklist grouped into editing principles, words, patterns and a final read. The em dash rule is soft: none in short copy, one or two allowed in longer drafts. scripts/build_plugin.py checks required fields in .codex-plugin/plugin.json, caps starter prompts at three entries of 128 characters, copies the skill, license, privacy and terms files into a dist folder, zips it, and byte-compares the packaged skill files against the originals. PRIVACY.md states there is no server, account or data collection.
Observed
- License
- MIT
- Form
- Agent skill made of Markdown instructions (SKILL.md plus eval.md checklist); no runtime code for the editing itself
- Install surface
- npx skills add petergyang/no-ai-slop --skill no-ai-slop --global --yes, or a natural-language install prompt to a coding agent
- Packaging
- ChatGPT and Codex plugin manifest in .codex-plugin/plugin.json, built and validated by scripts/build_plugin.py (Python)
- Modes
- Edit (default) and Detect (quote patterns without rewriting); also a satire draft prompt
- Data handling
- PRIVACY.md: no external server, no account, no data collection
- Tests
- No test suite in the files read; build script validates package contents only
Read from README.md, LICENSE, skills/no-ai-slop/SKILL.md, skills/no-ai-slop/eval.md, .codex-plugin/plugin.json, scripts/build_plugin.py, PRIVACY.md.
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.