Guizang's agent skill turns your repo's recent updates into a finished promo video — real components, code-rendered animation, an original code-written soundtrack — no video editor involved.
- agent-skill
- video
- promo
- marketing
anti-ai-slop
A skill that strips the 41 patterns that mark text as AI-written, then runs four gates — ending with a second reviewer — before handing it back.
anti-ai-slop rewrites a draft and then makes it pass four gates. Gate 1 removes invisible characters: zero-width spaces, hidden tag characters, bidi controls, odd spacing. Gate 2 is a deterministic Python linter that scores the draft out of 100 against everything in the catalogue a script can find and requires 90 with no serious findings — same score every run, code and quotes skipped. Gate 3 is a second agent that sees only the original and the rewrite and hunts what scripts cannot: made-up facts, invented time frames, dropped caveats, synonym cycling, changed meaning. Gate 4, optional, is another model — Gemini 3.8 Flash by default through LiteLLM, swappable for any provider — reporting a human score and its cost. Failing drafts loop back for up to two more passes; anything unresolved is listed for you, not hidden. The catalogue runs from negative parallelism and throat-clearing openers through contrast-on-every-claim and stacked hedges to em-dash density and Title Case headings.
What it does
- 41 pattern categories in five groups: sentence patterns, bulk-only habits (contrast on every claim, self-answered rhetorical questions), words and phrases (“delve”, “seamless” and about 120 more AI terms; “at its core” and about 100 more empty phrases), rhythm and formatting, and four second-reviewer-only catches
- slopcheck.py exits 0 on a pass and 1 on a fail, so the same gate runs in a pre-commit hook or CI; sanitize.py --fix cleans invisible characters; grade.py scores a rewrite against its original
- Line-level escape hatch: <!-- slop-ignore W1 --> keeps a flagged line (the linter's word list flags even “our leverage ratio fell to 2.1x”); everything it knows lives in rules.json, and tests keep the clean corpus above 95
- Two modes: by default the draft is assumed machine-written and the cuts are hard; say you wrote it yourself and it keeps your voice, fixing only the obvious tells
When to use it
Anyone publishing agent-drafted prose — blog posts, release notes, docs, newsletters — who wants the AI tells gone without a human editor pass; and CI setups where a slop-score gate belongs next to lint.
Getting started
git clone https://github.com/misbahsy/anti-ai-slop.git cp -r anti-ai-slop/skills/anti-ai-slop ~/.claude/skills/ # checks need Python 3.8+; optional fourth gate: pip install litellm python3 skills/anti-ai-slop/scripts/slopcheck.py draft.md # exit 0 = pass
Things to know
The linter is deterministic but pattern-based — the rules are English-centric, other languages will under-detect, and formal registers can trip false positives (the escape comment exists for that). The optional model review sends your draft to the provider you configure and reports each run's cost. Inspired by and crediting no-ai-slop, stop-slop, slop-cop and clearmode. MIT.
Repository
misbahsy/anti-ai-slopCharted
Related on the map
geo-sleuth
SkillAn agent skill that geolocates a photo with nothing readable in it — OSM geometry, elevation skylines and satellite tiles turn “where was this taken?” into a measured answer with an error radius.
- agent-skill
- geolocation
- osint
- openstreetmap
Birdview
SkillA skill that makes AI coding agents map the architecture and declare the blast radius before they touch a line of code — rendered as one standalone HTML page.
- skill
- architecture
- code-visualization
- change-scope