unsloppify is a CLI prose linter for markdown. It fixes
the mechanical AI tells, flags the phrases that need judgment, and
ships the same standards as a Claude Code skill and output style.
In today's fast-paced world, teams need a seamless clear way to delve into review copy before release.
Other anti-slop skills hand a draft to Claude and ask it to look for
the tells. That works, but it costs tokens and the obvious cases slip
past. unsloppify splits the work three ways.
In today’s fast-paced world, we wanted to delve into examine the vibrant tapestry range of seamless unbroken integrations. Let that sink in.
We wanted to examine the range of unbroken integrations.
Three-tier rule engine. Auto-fixes em dashes, smart quotes,
and a curated set of jargon swaps. Exits with code 1 on findings, so it drops into any CI or pre-commit pipeline
without ceremony.
A two-phase Claude Code workflow: the CLI removes the obvious punctuation tells first, then a judgment pass handles vocabulary, openers, and structures. Scored against a five-dimension rubric.
A Claude Code output-style that patches Claude’s own voice so the slop never appears in the first place. The preventative half of the prescription.
Every rule lives in one of three tiers. The tier decides whether a regex can fix it, flag it, or whether the call belongs to a human or an LLM with judgment.
CLI: uvx, pipx, pre-commit, GitHub Actions, any language-agnostic CI runner.
Agents: Claude Code (plugin marketplace + Agent Skills) and any other agent supporting Agent Skills.
A sample of the bundled rules. Run unsloppify --list-rules for the full set, or browse the YAML at src/unsloppify/rules/.
| 01 | -- | em dash -> double hyphen |
| 02 | " " | smart quotes -> straight |
| 03 | ... | ellipsis char -> three dots |
| 04 | \u200B | zero-width space -> stripped |
| 01 | delve into | cliche AI verb |
| 02 | In today’s fast-paced world | error: cut the entire phrase |
| 03 | Let that sink in | error: emphasis crutch |
| 04 | Here’s the thing | throat-clearing |
| 05 | deep dive | -> examination (fixable) |
| 06 | lean into | -> embrace (fixable) |
| 07 | circle back | -> revisit (fixable) |
| 08 | tapestry | almost never the right word |
| 01 | delve | the single most-flagged AI marker |
| 02 | seamless | show the seams aren’t there |
| 03 | vibrant | pick a more specific word |
| 04 | pivotal | key, important, or be specific |
| 05 | meticulous | careful or thorough |
| 06 | leverage | use “use” |
| 07 | foster | build, create, or encourage |
| 08 | paramount | critical, or be specific |
| 01 | It’s not X, it’s Y | binary contrast |
| 02 | Not just X, but Y | additive hedge |
| 03 | The X doesn’t do Y, it does Z | false agency |
| 04 | Short. Punchy. Sentences. | parataxis cliche |
| 05 | Whether you’re a beginner or expert | audience-pandering opener |
One CLI, one Claude Code Plugin or Agentic Skill, one output-style. Pick the rail that suits your pipeline.
Try it instantly with no install. Requires uv.
uvx unsloppify draft.md Install once with pipx, then call unsloppify from anywhere.
pipx install unsloppify Adds the /unsloppify skill and the unsloppified output-style.
plugin marketplace add petems/unsloppify plugin install unsloppify@unsloppify Works with Claude Code, Codex, Cursor, and any other agent supporting Agent Skills.
npx skills add petems/unsloppify Block slop from ever landing in git.
repos:
- repo: https://github.com/petems/unsloppify
rev: v0.1.0
hooks:
- id: unsloppify Render findings as inline PR annotations with the GitHub reporter.
name: prose
on: [pull_request]
jobs:
unsloppify:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- run: uvx unsloppify docs/ --format github + In today's fast-paced world, teams need a seamless way to review copy.
Formulaic opener. Remove the throat-clearing phrase.
AI-coded vocabulary. Replace it with the concrete behavior you mean.
text, json, and github annotations for CI/unsloppify, a two-phase workflow with a five-dimension scoring rubric (Clarity, Specificity, Agency, Rhythm, Honesty)unsloppified, which patches Claude's own responses so the slop never appears in the first place--rules path/to/rules.yaml