name

unsloppify

description

Avoid common AI writing tells (em-dash addiction, negative parallelism "it's not X — it's Y", "delve"/"tapestry"/"landscape", listicle-in-trench-coat, false ranges, throat-clearing transitions, bold-first bullets, gravitas-word inflation, etc.) when drafting, editing, or reviewing prose. Use whenever producing long-form text — blog posts, docs, marketing copy, essays, or rewriting AI-generated drafts to sound human.

AI Writing Tropes

When drafting or revising prose, load references/tropes.md and check the output against every category before returning it.

Categories (full examples in references/tropes.md)

Word choice: magic adverbs, "delve" family, ornate nouns ("tapestry"/"landscape"), "serves as" dodge, gravitas words
Sentence structure: negative parallelism, "Not X. Not Y. Just Z.", rhetorical self-Q&A ("The X? A Y."), anaphora, tricolons, "it's worth noting", superficial -ing tails, false ranges
Paragraph structure: short punchy fragments, listicle in a trench coat, numbered phase labels
Tone: "here's the kicker", "think of it as", "imagine a world", false vulnerability, "the truth is simple", stakes inflation, "let's break this down", vague attributions, invented concept labels, compliment sandwich, false exclusivity, clichéd idioms, fake casual quotes
Formatting: em-dash addiction, bold-first bullets, unicode decoration (→, smart quotes)
Composition: fractal summaries, dead metaphor, historical analogy stacking, one-point dilution, content duplication, signposted conclusion, "despite its challenges"
Engineering / technical jargon (PRs, commits, reports, offerings): performative slang (footgun, blast radius, monkeypatch, red herring, "ships day one", "the knob", "smoke test", "when X lands"), fallback diagnoses (race condition / debounce used as generic labels), technical filler ("at a glance", "in-depth", "deep dive", "under the hood")

How to use

Before producing the final text, scan once against each category above.
For any flagged pattern, rewrite — don't soften, remove.
One trope used once is fine. Multiple tropes, or the same trope repeated, is the failure mode.

Write like a human: varied, imperfect, specific.

AI Writing Tropes

When drafting or revising prose, load references/tropes.md and check the output against every category before returning it.

Categories (full examples in references/tropes.md)

Word choice: magic adverbs, "delve" family, ornate nouns ("tapestry"/"landscape"), "serves as" dodge, gravitas words

Sentence structure: negative parallelism, "Not X. Not Y. Just Z.", rhetorical self-Q&A ("The X? A Y."), anaphora, tricolons, "it's worth noting", superficial -ing tails, false ranges

Paragraph structure: short punchy fragments, listicle in a trench coat, numbered phase labels

Tone: "here's the kicker", "think of it as", "imagine a world", false vulnerability, "the truth is simple", stakes inflation, "let's break this down", vague attributions, invented concept labels, compliment sandwich, false exclusivity, clichéd idioms, fake casual quotes

Formatting: em-dash addiction, bold-first bullets, unicode decoration (→, smart quotes)

Composition: fractal summaries, dead metaphor, historical analogy stacking, one-point dilution, content duplication, signposted conclusion, "despite its challenges"

Engineering / technical jargon (PRs, commits, reports, offerings): performative slang (footgun, blast radius, monkeypatch, red herring, "ships day one", "the knob", "smoke test", "when X lands"), fallback diagnoses (race condition / debounce used as generic labels), technical filler ("at a glance", "in-depth", "deep dive", "under the hood")

How to use

Before producing the final text, scan once against each category above.

For any flagged pattern, rewrite — don't soften, remove.

One trope used once is fine. Multiple tropes, or the same trope repeated, is the failure mode.

Write like a human: varied, imperfect, specific.