writinglint-rulepack-craft
Better-writing rules (formerly writinglint-rulepack-conviction). A rulepack for
WritingLint that catches the
science of bad writing — first the tells of a writer who doesn't quite believe
they're allowed to say the thing, then the music of the prose itself:
| Rule | Catches | Default |
|---|---|---|
hedge-opener |
"It might seem counterintuitive, but …" — apologising for the claim before making it | error |
verdict-echo |
"better parsers to build better guardrails" — the same evaluative word spent twice | error |
verdict-word |
"unbiased", "properly", "robust" — a conclusion with the evidence missing | warn |
parenthetical-hedge |
"(not that important)" — a sentence undercut from inside a parenthesis | error |
qualifier-softener |
"pretty good" — a dimmer switch on your own claim | warn |
apology-reflex |
"sorry for the long post" — apologising to a reader who hasn't objected | error |
uniform-rhythm |
metronome prose — every sentence near the same length (Gary Provost: "vary the sentence length, and I create music") | warn |
The structural rules (hedge-opener, qualifier-softener) match on the
dependency graph, so any words can fill the construction's slots: "This could
sound naive …" fires; "It seemed late" and "pretty flowers" don't.
Every rule here was earned by a real draft. The messages don't just name the problem — they say what to do instead, because the fix is always one of two moves: lead with the claim, or replace the verdict with the event that produced it ("like what?").
Use
import { defineConfig } from 'writinglint-core';
import { recommended } from 'writinglint-rulepack-craft';
export default defineConfig({
extends: [recommended],
// rules: { 'craft/verdict-word': 'error' }, // promote / demote / disable
});Or layer it with the ai-style pack — they flag different failure modes (models overclaim; people underclaim) and compose cleanly.
License
MIT.