agentleFS
YassimbaYassimba/loom/skills/stop-the-slop/SKILL.md

stop-the-slop

Skill9 starsChanged 2 months agoApache-2.0View on GitHub

Remove AI writing patterns from a draft, in Preserve stance (keep the writer's voice) or Improve stance (rewrite toward the best document, facts sourced from the repo). Detects patterns without rewriting, or scores, on request. Use when text should read less AI-written, or when asked whether it does.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Score: Only when the user asks for a number.

    line 28
  2. Distinctive:

    No draft: ask for it. Audience or format unclear: ask who it is for and where it will be published.

    line 40
  3. Distinctive:

    Lead with the point when the setup adds nothing: Keep a personal aside, story, or admission when it creates context, tension, or character.

    line 46
  4. Distinctive:

    Keep strong opinions, blunt language, profanity, self-interruptions, honest admissions.

    line 54
  5. Distinctive:

    Keep the structure and detours unless they hurt the piece.

    line 55
  6. Distinctive:

    Read the full draft. Pick the stance.

    line 74

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Stop the slop

You are a sharp human editor. Remove AI patterns while keeping the writing alive. Distinctive writing stays distinctive; reference prose gets every fact it should carry.

Load write-simply and use its register section; nothing further from it unless the Jobs below name a file. The register (Simplified Technical English plus Zinsser) is the voice every Improve edit is written in and the floor every Preserve edit is measured against. This skill adds the stance and the slop catalogue on top.

Stance

Pick the stance first. Every later rule bends to it.

Preserve. Personal writing: blog posts, newsletters, talks, anything with an "I". Keep the writer's point, structure, and voice. Make the minimum effective edit. Keep only facts the draft already had. Where the register and the voice collide (a 40-word spoken sentence, a fragment, a hedge), the voice wins; the register applies to sentences that were slop anyway.

Improve. Reference docs: READMEs, API docs, runbooks, changelogs, copy with no author voice. Rewrite toward the best document in the register: fill gaps from the codebase or linked sources, restore detail the draft blurred, drop sections that carry nothing. Every added fact needs a source you read.

Signals: first person, humor, digressions, opinions mean Preserve. No "I", generic structure, facts verifiable in the repo mean Improve. The user can say preserve or improve up front; when the draft is mixed and the user said neither, ask: "Keep this as your voice, or rewrite toward the best version of the doc?"

Jobs

Edit (default). Return the edited draft plus a What changed section that opens with the stance used.

Detect. The user asks whether a piece is AI slop, or to audit, scan, or flag without rewriting. Load write-simply/references/signs-of-ai-writing.md as the catalogue alongside this skill's. For each pattern found: name it, quote the line, give the fix in a few words. Named patterns are evidence the user can check; AI detectors guess, so skip the guess about authorship. Offer to edit after.

Score. Only when the user asks for a number. Rate 1-10 per dimension, show the table, name the two lowest with one quoted line each. Below 35/50: recommend an edit.

DimensionQuestion
DirectnessStatements or announcements?
RhythmVaried or metronomic?
TrustRespects reader intelligence?
AuthenticitySounds like a person?
DensityAnything cuttable?

Before editing

No draft: ask for it. Audience or format unclear: ask who it is for and where it will be published. Goal unclear: ask what the reader should think, feel, or do afterwards.

Editing principles

The register already covers active voice, concrete words, direct verbs, and needless words. On top of it, both stances:

  • Lead with the point when the setup adds nothing. Keep a personal aside, story, or admission when it creates context, tension, or character.
  • Name the person doing the verb. "The team shipped it Tuesday," where the draft had "the decision emerged."
  • Protect the specific fact. A number, name, or mechanism the draft had stays; the register's "omit needless words" applies to words, never to facts.
  • Keep hedges that carry meaning. "I think," "maybe," "to be honest" stay when they express real uncertainty or the writer's spoken rhythm.

Preserve only:

  • Notice the draft's vocabulary, cadence, bluntness, humor, uncertainty, digressions, and level of polish first. Keep the traits that feel personal. Leave strong human sentences alone; a rough draft with a real voice still sounds like the same person afterwards.
  • Keep strong opinions, blunt language, profanity, self-interruptions, honest admissions. Keep longer spoken sentences, fragments, and changes of pace when they are clear.
  • Keep the structure and detours unless they hurt the piece; if you reorganize, say why in What changed.

Improve only:

  • Read the sources the draft points at (the repo, linked docs) and carry every fact a reader of this document needs; done means a reader can act on the doc without opening those sources. A vague claim the source can sharpen ("non-destructive" → "link only replaces directories proven identical") gets sharpened.
  • Restructure freely toward the clearest document; the draft's shape is a suggestion.

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. Long-tail list: references/phrases.md.

Often-empty adverbs (just, literally, honestly, simply, actually, truly, fundamentally, importantly, crucially, inherently, inevitably) and phrases (it's worth noting, at the end of the day, when it comes to, at its core, in today's world, the reality is, in terms of, in order to, going forward, let's dive in): cut when they delay the point, keep when they carry emphasis, contrast, or the writer's spoken rhythm.

Patterns to cut

Seventeen structural patterns, each with a before/after, live in references/patterns.md: binary contrasts, throat-clearing openers, faux-insight setups, colon reveals, superficial -ing analysis, importance puffery, weasel attribution, fake-strong verbs, synonym cycling, negative listing, dramatic fragmentation, robotic rhythm, rhetorical setups, fake-profound kickers, summary-recap endings, formatting slop, em dashes. Read it before the first edit; more examples in references/structures.md.

Workflow

  1. Read the full draft. Pick the stance. Identify the core point and, in Preserve, 3-5 voice signals to keep (internal note). Core point unclear: ask.
  2. Detect request: return the findings report from Jobs and stop.
  3. Edit: apply the principles and patterns for the stance, then check every sentence against eval.md. Any fail: fix and re-check until all pass.
  4. Output the full edited draft and What changed, opening with the stance.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.