Lemon Agent № 1 · the taste arm · 24 entries draft

The Prose Failure List

Run it against a draft — yours, or your assistant’s. Every match is a cut or a rewrite, never an addition. The list names failures, not virtues: prose that survives subtraction is yours.

Distilled from the recurring failure modes of AI-assisted prose and the editing habits of one opinionated author. v0 — ids are stable (cite PF-07, PF-16), but entries may still be cut before 1.0: a shorter list you can defend beats a long one you’d shrug at.

How to use it

A match is a signal, not a sentence. Any entry can be overridden on purpose — the failure is doing these things by default, not doing them at all.

In any chat assistant. Paste this prompt, then the list, then your draft — works anywhere that reads text:

Apply the Prose Failure List (below) to my draft as a subtraction pass.
For every match: cite the PF id, quote the offending text, and give the
exact cut or rewrite. Never add content — only cut or tighten. Finish
with the three edits that would improve the piece most.

As a Claude Code skill. Installs a subtraction pass you can run on any file in a project:

git clone https://github.com/chronick/lemon-agent
cp -r lemon-agent/skills/prose-failure-list ~/.claude/skills/

Then ask: “run the prose failure list on drafts/my-post.md”.

In an agent or pipeline. Fetch the raw list at /lists/prose-failure-list.md and the machine catalog at /catalog.json. Position it as an advisory pass — it reports, it never gates: taste advises, it doesn’t block.

A. Diction

  1. PF-01 — Inflated examination verbs. “Delve into,” “unpack,” “dive deep,” “explore” — verbs that gesture at work instead of naming it. Say what actually happened: read, measured, listed, compared.
  2. PF-02 — Negation-elevation. “It’s not just X — it’s Y.” The first clause is scaffolding; assert Y and let it stand.
  3. PF-03 — Rule-of-three padding. Triads where the third item exists for rhythm, not content (“faster, cheaper, and more scalable”). Two real items beat three where one is invented.
  4. PF-04 — Em-dash chaining. Dashes replacing sentence structure. More than one pair in a paragraph means the sentences were never finished.
  5. PF-05 — Superlative gloss. “Seamlessly,” “effortlessly,” “incredibly,” “powerful” — adjectives claiming what the sentence didn’t demonstrate. Show the seam count or drop the seamless.
  6. PF-06 — Throat-clearing opener. Establishing context nobody doubted (“In today’s rapidly evolving landscape…”). Start where the piece starts.
  7. PF-07 — Fake-synthesis adverbs. “Ultimately,” “at the end of the day” — gluing on a conclusion the argument never derived.
  8. PF-08 — Anthropomorphized abstraction. Technologies “empowering,” trends “driving,” ideas “demanding” — agency assigned to non-agents so no one has to name who does what.

B. Structure

  1. PF-09 — Bullet-itis. Connected reasoning chopped into bullets to look organized. If the items share a subject and verb shape, it was a paragraph; restore it.
  2. PF-10 — Uniform rhythm. Every paragraph three to four sentences of similar length. No short sentence ever lands a point. Vary it.
  3. PF-11 — The restating conclusion. A final section that re-says the introduction, slower. End where the argument ends.
  4. PF-12 — Signpost overload. “Let’s take a look at,” “as we’ve seen,” “it’s worth noting that” — narrating the essay instead of writing it.
  5. PF-13 — Colon-title reflex. “X: Why Y Matters in 2026.” One clause, no subtitle, or earn the colon.
  6. PF-14 — Symmetry over weight. Sections sized equally regardless of how much each matters. The interesting part deserves the imbalance.
  7. PF-15 — Preemptive contents-prose. Telling the reader what you’re about to say, in a piece short enough to just say it.

C. Rhetoric & stance

  1. PF-16 — Verdict-free both-sides-ism. “There are advantages and disadvantages” with no ranking — a survey delivered where a stance was owed. Rank them or own that you can’t yet.
  2. PF-17 — “Only time will tell.” And every other closer that hands the question back unanswered.
  3. PF-18 — Unearned “we.” “We all know,” “as a community” — consensus asserted rather than demonstrated. Say I, or name who.
  4. PF-19 — Stakes-free enthusiasm. Everything “exciting” and “fascinating” while nothing is risked, spent, or lost. Excitement is shown by what you did next.
  5. PF-20 — Analogy stacking. A second analogy for a thing the first one — or the literal sentence — already carried.
  6. PF-21 — Vague authority. “Studies show,” “experts agree,” “many argue” — the shape of a citation with no citation inside.
  7. PF-22 — Moralizing coda. The tacked-on lesson (“what this teaches us about…”) after the piece already made its point.

D. Authenticity

  1. PF-23 — Experience-free authority. No “I tried it,” no failure, no cost anywhere — advice with no scar tissue. If you didn’t run it, say so; if you did, the failure is the interesting part.
  2. PF-24 — Number-shaped fog. “Several,” “significantly,” “a large number” where the real count was one lookup away. Numbers carry dates and sources or they’re vibes — the house rule, applied to prose.

Part of Lemon Agent — free tools, honest measurements, agentic services. Companion: The Agent Workflow Failure List (№ 2, the verification arm).

← Lemon Agent · raw markdown