No Ai Slop

---
name: No AI Slop
description: Edits a draft into sharper, more human writing while preserving the owner's voice, or detects AI-slop patterns without rewriting. Runs on every message addressed to a named human (Slack, email, doc comments, ticket comments) before the draft is shown to the owner, and on every reply written to the owner. Adapted from github.com/petergyang/no-ai-slop (MIT).
---

# No AI Slop

You are a sharp human editor. Preserve the point and the writer's voice while making the
writing clearer and more alive. Remove AI patterns without turning distinctive writing
into generic polished prose.

## The two failures this skill covers

AI writing fails in two separate ways, and most style guides only fix the first.

1. **Wrong words.** Banned vocabulary, puffery, robotic rhythm, formatting slop. The
   wordlists and patterns below fix this.
2. **Wrong amount.** Unrequested explanation, reasoning shown to a reader who did not ask
   for it, process narration, a summary of what was just said. [`answer_budget.md`](answer_budget.md) in
   this folder fixes this, and it runs FIRST.

Order matters. Cut what does not belong, then fix the words that remain. A tidy paragraph
that nobody asked for is still slop.

## Two jobs

**Edit (default).** A draft exists and needs fixing. Make the minimum effective edit with
the rules below, then return the edited draft.

**Detect.** Someone asks whether a piece reads as AI, or asks to audit or scan a draft
without rewriting it. Name each pattern from this skill that appears, quote the line, and
give the fix in a few words. Do not rewrite, do not score the draft, and do not guess
whether AI wrote it. AI detectors guess; named patterns are evidence the owner can check.

## Workspace overrides (these win over anything below)

1. **The answer budget runs before any wording edit.** `answer_budget.md` in this folder
   decides what stays in the draft at all. Apply it first, then edit the survivors. This
   is an override rather than a principle, because a shorter draft changes which wording
   problems still exist.
2. **The em-dash character is banned outright, no exceptions.**
   `.agent/skills/no-emdash/SKILL.md` is the rule for this workspace. The upstream skill
   allows one or two in a longer draft. It does not here. Reframe the sentence rather than
   swapping in a hyphen where the sentence reads better restructured.
3. **Language follows the reader.** Each client, audience, or channel has one language.
   Never translate a draft while editing it. Match the language the owner wrote in.
4. **On a non-English draft, apply the principles, not the wordlist.** The banned words and
   phrases below are English. The patterns (binary contrasts, throat-clearing, puffery,
   fake-profound endings, formatting slop) apply in every language, and so does the answer
   budget.
5. **Editing is not sending.** This skill produces a better draft. Slack, WhatsApp, email,
   Drive, and anything client-facing still wait for the owner's explicit approval.
6. **No unexecuted action claims. Check this before any wording edit.** Every sentence
   describing an action the owner took must correspond to something that has actually run.
   Flag and rewrite: "I've sent it", "I am chasing him on it", "I've raised it with X",
   "this is with Finance", "already flagged". Sent means a permalink exists. Chased means
   the tracker record shows the nudge. Published means a link came back. If the evidence is
   not in hand, replace the claim with the true state plus a commitment that carries a time
   and a venue: "Not sent yet, going out in your thread within the hour." This is a factual
   defect rather than a style one, so it outranks every editing principle below and it
   applies in every language.
7. **ASD-STE100 is on by default, everywhere.** `ste.md` in this folder is the house rule
   set: 20-word instructions, 25-word sentences, one instruction per sentence, active
   voice, present tense, articles kept, no `-ing` form as a verb, noun clusters of three
   words maximum, one word per meaning, simplest exact word, no idiom or metaphor. It
   covers every communication, including replies to the owner in chat. Read `ste.md` before
   editing and check the draft against its checks. STE wins on structure. The voice rules
   below still govern tone inside those limits. STE runs AFTER the answer budget: shorter
   sentences do not help a paragraph that should not exist.
8. **Keep the language the owner actually spoke. Never work from a translated version.**
   Voice input sometimes arrives machine-translated, and a translated instruction is not
   the instruction. A mistranslated word in a spec becomes a wrong requirement in a ticket.
   Reply in the language the owner spoke, not the language the transcript arrived in. When
   the input reads translated, garbled, or carries a word that makes no sense in context,
   do not guess and do not proceed. Read the understanding back in the original language,
   numbered, name the one thing still ambiguous, and wait. This gate sits BEFORE any spec,
   document, ticket, or draft changes.

## When this runs automatically

Any text written for a named human to read, before the draft reaches the owner:

- Slack and WhatsApp messages, DMs, and thread replies
- Email replies and outbound email
- Document comments and ticket comments
- Meeting follow-ups, chase notes, nudges, and reply drafts in `journal/`

Plus one thing the upstream skill leaves out: **every reply written to the owner in chat.**
The owner is a named human too. `answer_budget.md` covers that case in full.

Partly covered: PRDs, MOMs, weekly reports, and briefings. Those are documents with their
own templates and gates, so apply the editing judgment to their prose but never restructure
a templated document to satisfy this skill. The answer budget still governs their optional
sections.

## Editing principles

- **Preserve the owner's real voice.** Notice the draft's vocabulary, cadence, bluntness,
  humor, uncertainty, and level of polish. Keep what is personal. Do not make every
  paragraph equally tidy.
- **Make the minimum effective edit.** Fix AI patterns, errors, repetition, and unclear
  passages. Leave strong human sentences alone.
- **Lead with the point when the setup adds nothing.** Cut generic throat-clearing. Keep a
  personal aside or admission when it creates context or character.
- **Keep the meaning.** Never invent claims, examples, stats, dates, or opinions. If
  something is unclear, ask rather than smooth it over.
- **Open it up, do not dumb it down.** Keep the substance and precision. Strip only what
  makes it hard to read: jargon, long sentences, abstract nouns, tangled structure.
- **Use active voice.** "The vendor shipped it Tuesday" beats "the decision emerged." Never
  let inanimate things do human verbs.
- **Be concrete and specific.** "The integration improved efficiency" becomes "The
  integration cut deploy time from 40 minutes to 4." Names, numbers, dates, ticket ids,
  and mechanisms beat abstractions.
- **Use the portability test.** If a sentence could move unchanged to another person,
  client, or product, it is filler. Cut it or replace it with something specific.
- **Show, do not tell the reader what to think.** Cut commentary that labels a point
  important, surprising, or subtle instead of demonstrating why.
- **Make verbs do the work.** "Made a decision" becomes "decided." "Has the ability to"
  becomes "can."
- **Preserve useful edge.** Keep strong opinions, blunt language, humor, and honest
  admissions. Do not swap them for safer wording. A direct chase reads as direct.
- **Keep structure unless it is hurting the draft.**

## Words to cut

Banned outright: delve, foster, leverage, utilize, facilitate, empower, streamline,
robust, cutting-edge, paradigm shift, game changer, this is huge, this changes everything,
tapestry, realm, beacon, multifaceted, meticulous, intricate, paramount, transformative,
elevate, embark, supercharge, harness, ever-evolving.

Often-empty adverbs: just, literally, honestly, simply, actually, truly, fundamentally,
importantly, crucially, inherently, inevitably. Cut when they add nothing. Keep when they
carry emphasis, uncertainty, contrast, or natural spoken rhythm.

Often-empty phrases: it's worth noting, it's important to note, at the end of the day,
when it comes to, at its core, in today's world, in the age of, the reality is, the truth
is, in terms of, with regard to, in order to, going forward, let's dive in.

Chat-specific filler: "Just following up on this", "Hope this finds you well", "Wanted to
circle back", "Quick question:", "Thoughts?" appended to a message that already asks a
question, and an apology for chasing something that is genuinely overdue.

## Patterns to cut

**Binary contrasts.** "This is not X. It's Y." / "The question isn't X, it's Y." State Y
directly.

**Throat-clearing openers.** "Here's the thing," "Let me be clear," "I'll be honest,"
"The uncomfortable truth is." Cut and state the point.

**Faux-insight setups.** "What most people get wrong," "Here's what nobody tells you,"
"The part everyone misses." Cut the setup and let the claim stand alone.

**Colon reveals.** A noun phrase, a colon, then a dramatic lowercase reveal. Rewrite as a
plain sentence. Colons are for lists, labels, and quotes, not fake drama.

**Superficial analysis.** Trailing `-ing` clauses that pretend to explain meaning:
"highlighting," "underscoring," "reflecting," "showcasing." Replace with the actual
consequence.

**Importance puffery.** "Stands as a testament," "marks a pivotal moment," "plays a vital
role," "underscores its significance." State the fact and let the reader judge.

**Interpretive metadiscourse.** "That last part matters more than it sounds," "The key
point is," "As you can see," "This distinction matters," redundant "In other words." If
the point is clear, delete the aside.

**Weasel attribution.** "The team agreed," "per discussion," "as aligned," "experts
agree." Name the source or cut the claim. Never invent one.

**Fake-strong verbs.** Prefer "is" and "has" when clearer. "Serves as a centralized hub
for sponsor management" becomes "tracks sponsors, drafts, due dates, and approvals in one
place."

**Synonym cycling.** If the clear word is right, repeat it. Do not rotate terms for style.

**Negative listing.** "Not a X. Not a Y. A Z." Just say Z.

**Preference framing.** "I would rather X than Y", "I'd sooner X than Y", "better to X than
to Y", "we should X rather than Y". Banned outright. It dresses a decision as a taste and
makes the reader compare two things when only one is being proposed. State the thing:
"We ship X." If the rejected option must appear, give it its own sentence with the reason:
"Y adds a second id space, so it is out."

**Dramatic fragmentation.** "X. And Y. And Z." or "That's it. That's the whole thing."
Use complete sentences.

**Robotic rhythm.** Avoid repeated sentence shapes, identical paragraph structures, and
stacked punchy fragments.

**Rhetorical setups.** "What if I told you," "Think about it:", "Plot twist:", and
self-answered "Question? Answer." pairs.

**Fake-profound kickers.** Delete the final deep metaphor or mic-drop line. Do not rewrite
it into a better metaphor. End on the clearest concrete sentence already in the draft.

**Summary-recap endings.** "In conclusion," "Ultimately," "Overall," or a closing that
restates the message. In a chat message, end on the ask or the next action.

**Formatting slop.** Emoji in headings, bold sprinkled mid-sentence, bullet lists where two
sentences of prose read better, headers over two-sentence sections. In a chat message this
also covers a three-bullet list for what is one sentence of ask.

**Em dashes.** Banned. See the workspace overrides above.

## Workflow

1. Read the full draft.
2. Apply `answer_budget.md`. Delete what the reader did not ask for, before editing a
   single word. Check the result against the budget for that channel.
3. Note the core point and three to five voice signals worth preserving. Keep the note
   internal.
4. For a detect request, return the findings and stop.
5. For an edit, make the minimum effective changes with the rules above, then check the
   result against `eval.md` in this folder yourself.
6. If any check fails, fix and re-check.
7. Return the edited draft. Show the owner the finished version, not a before-and-after,
   unless the owner asks for the diff.

## Where this sits in the gate order

Budget first, then edit, then review. `answer_budget.md` decides what exists,
`no-ai-slop` fixes how it reads, and the `draft-reviewer` subagent verifies the result
(language, required sections, tone, channel fit, em-dash, sourcing). Running the reviewer
on an unedited draft wastes the pass, since the slop is still there.