Command line

The writinglint package installs a writinglint command. It lints prose against the ai-style rulepack (by default) and, separately, scores how AI-shaped a document reads.

Install

The owned ONNX parser runs locally without Python:

npm install
npm run cli -- lint README.md

Usage

writinglint essay.txt              # lint one doc (+ score)
writinglint lint posts/*.md        # lint many docs
writinglint score posts/*.md       # just the AI-style score per doc
cat essay.txt | writinglint        # read from stdin
writinglint --json essay.txt       # machine-readable output
writinglint --quiet posts/*.md     # one summary line per doc
writinglint --config writinglint.config.ts essay.txt
writinglint --model /path/to/onnx/model essay.txt
slopsift --json essay.txt           # exits 1 while findings remain

In the workspace, the ONNX bundle is discovered automatically. Override it with --model <dir> or WRITINGLINT_ONNX_MODEL.

Each lint prints its location, rule id, category, and message, followed by an inline colour reproduction of the text with the flagged spans highlighted — and, at the end, the document’s AI-style score.

Configuration

Drop a writinglint.config.ts in your working directory and the CLI picks it up automatically (or point at one with --config). It’s the same defineConfig you’d use as a library:

writinglint.config.ts

import { defineConfig } from 'writinglint-core';
import { recommended } from 'writinglint-rulepack-ai-style';

export default defineConfig({
  extends: [recommended],
  rules: {
    // Writing casual prose? Silence the emoji rule:
    'ai-style/emoji': 'off',
    // Promote the construction that started this project to an error:
    'ai-style/corrective-antithesis': 'error',
  },
});

With no config file, the CLI falls back to the AI-style recommended config. Findings derive their level from confidence: high is an error, medium is a warning, and low is informational.

Exit code & CI

--json emits a structured record per document (score, verdict, and every lint with offsets), which is what you want in CI or an editor integration. Pipe it to jq, gate a commit on it, or feed it to your own reporter.