| name | unslop |
| description | Edit or review prose to remove AI-generated writing patterns. Use only for writing, documentation, copy, reports, or messages when the user asks to unslop or deslop text, remove AI tone, or clean up prose. Do not use for source code. |
Unslop
Edit text to remove AI patterns while preserving its substance and intended
voice.
Editorial invariants
- Preserve substance. Keep each fact, name, number, date, quote, citation, ranking, qualification, and intended conclusion. Unsupported promotional evaluation and significance inflation are not protected substance when the user asks to unslop.
- Remove unsupported promotion. Remove praise, sales framing, and claims of importance that the source does not support. Keep concrete facts and substantive opinions, but do not invent evidence or a factual explanation to replace the removed language.
- Do not invent details. Never add facts, sources, statistics, quotes, anecdotes, or personal experiences. Mark a hypothetical as hypothetical.
- Match the writer. A supplied writing sample or established document style takes priority over the default rules below. Preserve deliberate quirks when they do not obscure meaning.
- Respect the register. Do not add opinions, first-person language, humor, or casual phrasing to reference, legal, scientific, or other neutral prose.
- Protect literal text. Do not rewrite quotations, proper names, code, commands, link targets, data, or examples that discuss a watched phrase unless the user asks.
- Treat patterns as evidence, not proof. Fix a pattern when it is repeated, formulaic, or wrong for the context. Do not flatten a human choice merely because it appears in this catalog.
Prose process
- Scan for the patterns below.
- Identify the target voice, register, audience, and output format.
- Rewrite the passage around its main points instead of replacing watched words one at a time.
- Compare the rewrite with the source. Restore any lost fact, qualification, or intended conclusion. Remove every unsupported addition, promotional evaluation, and inflated significance claim.
- Self-audit with the quick checks, then score. Fix remaining tells.
Match voice and register
When the user supplies a writing sample, first note its sentence length, vocabulary, paragraph openings, punctuation, repeated phrases, transitions, and level of formality. Match those habits. The sample overrides generic preferences such as the em-dash rule.
Without a sample, match the context:
- Personal writing, essays, and opinion. Preserve mixed feelings, humor, first person, asides, and uneven rhythm when they fit.
- Technical prose. Keep established domain terms, symbols, commands, and concrete mechanisms. Do not replace precise terminology with casual synonyms.
- Scientific prose. Use "we" for the authors' work when appropriate, retain necessary qualifications, and name or cite specific researchers when the source identifies them instead of invoking unnamed experts.
- Reference, legal, and factual prose. Stay neutral. Clarity matters more than personality.
- Email, chat, and social posts. Match the platform. Do not add Markdown to plain-text contexts or turn a short message into a document with headings.
Add personality only when it fits
Removing patterns is half the job. Sterile writing can be just as generic, but personality must come from the source or the requested voice.
- Keep substantive opinions. Preserve the writer's position instead of neutralizing it into a list of pros and cons. Unsupported praise and importance claims remain removable.
- Vary rhythm. Short sentences. Then longer ones that take their time. Mix it up.
- Keep complexity. Preserve uncertainty, mixed feelings, and unresolved tension instead of forcing a clean verdict.
- Use "I" or "you" when the writer and context call for it. First and second person are tools, not defaults.
- Let some mess remain. Do not polish away a deliberate aside, self-correction, or unusual detail.
- Preserve specificity. Keep concrete details from the source. Do not replace them with a generic reaction or invent new details to make the passage feel personal.
Patterns to detect and fix
Content
- Significance inflation. "pivotal moment", "testament to", "evolving landscape", "setting the stage for", "indelible mark", "deeply rooted". Remove importance that the source facts do not support. Keep a sourced evaluation or reputation claim, and do not invent support.
- Notability name-dropping. Listing media outlets without context. Preserve the named outlets, but add what they said only when the source provides that context.
- Superficial -ing phrases. "highlighting...", "ensuring...", "reflecting...", "showcasing...", "fostering...". Delete empty interpretation or replace it with facts already present in the source.
- Promotional language. "nestled", "vibrant", "breathtaking", "groundbreaking", "renowned", "stunning", "must-visit". Remove sales framing and unsupported praise. Keep concrete facts, sourced reputation, and evaluations supported by the source. Do not preserve unsupported promotion only because the genre uses it.
- Vague attributions. "Experts believe", "Industry reports suggest", "Some critics argue". Name the source when the input provides one. Otherwise keep the qualification on a substantive claim or flag the missing source in review mode. Remove unsupported promotion or significance, and do not invent an authority.
- Formulaic challenges. "Despite challenges... continues to thrive." Keep
specific challenge and outcome facts from the source. Remove unsupported
success framing instead of inventing evidence for it.
- Vague declaratives. "The reasons are structural", "The implications are significant", "The stakes are high". Replace an announcement with its specific meaning when the source supplies it. Otherwise remove empty significance in a rewrite or flag it in a review. Do not invent an explanation.
- Lazy extremes. "every", "always", "never", "everyone", "nobody" doing vague work. Keep a supported quantifier. If the source does not support it, flag the claim in review mode rather than silently changing it.
Language
- AI vocabulary. Additionally, crucial, delve, enduring, enhance, fostering, garner, interplay, intricate, landscape (abstract), pivotal, showcase, tapestry (abstract), testament, underscore, vibrant. Replace with plain words.
- Copula avoidance. "serves as", "stands as", "boasts", "features". Just say "is" or "has".
- Formulaic negative parallelisms. "It's not just X, it's Y." "The answer isn't X. It's Y." "X isn't the problem. Y is." State the point directly when the contrast exists only to manufacture a reveal. Keep a contrast that carries a real distinction. See references/structures.md for the full catalog.
- Negative listing. "Not a X... Not a Y... A Z." Collapse the runway only when X and Y add no substantive claims. Otherwise preserve the distinctions in a less theatrical sentence.
- Forced rule of three. Use the natural number of items. Do not add or remove an item merely to produce or avoid a group of three.
- Synonym cycling. Protagonist, main character, central figure, hero all in one paragraph. Pick one, repeat it.
- False ranges. "from X to Y" where X and Y aren't on a meaningful scale. List topics directly.
- False agency. Inanimate things doing human work can hide who acted: "the decision emerges", "a complaint becomes a fix". Name the actor when the source identifies them and the actor matters. Keep ordinary shorthand such as "the data shows" when it is clear and natural.
Structure and drama
- Dramatic fragmentation. "[Noun]. That's it. That's the [thing]." Staccato stacks of punchy fragments. Manufactured profundity. Use complete sentences and trust the content.
- Rhetorical setups. "What if [reframe]?", "Here's what I mean:", "Think about it:", "And that's okay." Remove the setup, but preserve any conclusion or distinction it carries.
- Throat-clearing openers. "Here's the thing:", "The uncomfortable truth is", "Let me be clear", "It turns out". Any "here's what/this/that" construction is announcement before the point. Cut it and state the point. Full list in references/phrases.md.
- Meta-commentary. "The rest of this essay explains...", "Let me walk you through...", "As we'll see...", "But that's another post". The writing should move, not announce its own structure.
- Wh- sentence starters as a crutch. "What makes this hard is..." becomes "The constraint is..." or, when the source supplies it, the specific constraint. Lead with the subject or the verb.
- Pull-quotes. If a sentence sounds like it was written to be quoted, rewrite it.
- Narrator-from-a-distance. "People tend to..." can create a detached lecturer voice. In personal or instructional prose, use a closer perspective only when the source supports it. Do not change "nobody" to "you" or otherwise alter scope and agency.
Style
- Em dash overuse. Remove repeated or formulaic em dashes. Prefer periods or commas when they read cleanly. If a supplied writing sample or established house style uses em dashes deliberately, match that style instead of banning them.
- Colon overuse. Colons are fine before a list or example. Not as mid-sentence connectors. "If you're coming from traditional automation: instead of registering event handlers, you describe conditions" adds nothing with the colon. Rewrite to let the point stand on its own without comparison framing. "Describing when the scheduler should fire works best as plain English." Same meaning, no crutch punctuation.
- Boldface overuse. Don't bold every proper noun or acronym.
- Inline-header lists. The tell is a bold label and colon that restates the line: "Performance: Performance improved...". Convert those to prose. A bold lead-in that ends in a period, names the item, and is followed by genuinely new detail ("Schema in TypeScript. Tables live in one file.") is fine, not a tell.
- Title case headings. Use sentence case unless the document's established style requires title case.
- Decorative emojis. Remove from headings and bullets.
- Mismatched quotation marks. Match the source format or house style. Do not normalize curly or straight quotes without a reason.
Communication artifacts
- Chatbot phrases. "I hope this helps!", "Let me know if...", "Of course!", "Certainly!", "Found the smoking gun!" Remove.
- Cutoff disclaimers. "While specific details are limited..." State the known limit directly. Remove only unsupported guesses or commentary that adds no claim.
- Sycophantic tone. "Great question! You're absolutely right!" Respond directly.
Filler
- Filler phrases. "In order to" becomes "To". "Due to the fact that" becomes "Because". "It is important to note that" gets deleted. "At its core", "At the end of the day", "When it comes to", "The reality is" get deleted too.
- Emphasis crutches. "Full stop.", "Let that sink in.", "Make no mistake", "This matters because". Remove empty emphasis, but retain any causal explanation or conclusion attached to it.
- Excessive hedging. "could potentially possibly be argued that it might" becomes "may".
- Generic conclusions. "The future looks bright." Prefer specific plans or facts already present in the source. Remove unsupported optimism while preserving a concrete intended conclusion.
Jargon
- Abstract metaphor nouns. Outside established domain usage, substrate, wedge, vector, locus, vantage, nexus, harness, bedrock, scaffolding, modality, paradigm, and gold-plating often have a plainer concrete word. "Substrate" may become "base". "Wedge in" may become "add". "Vector" may become "way" or "method". Preserve terms such as "API surface" and "language primitive" when they carry their accepted technical meaning.
- Business jargon. "Navigate challenges" becomes "handle". "Unpack" becomes "explain". "Lean into" becomes "accept". "Double down" becomes "commit". "Circle back" becomes "return to". Table of replacements in references/phrases.md.
Plain speech
- Say the concrete thing. Don't wrap a simple point in abstract framing, and don't describe how something feels instead of what it does. "the database stays close at hand", "SQL you can read", "types that follow your schema" name a feeling. Use a mechanism or number only when the source provides one: "
.toSQL() returns the exact string sent to the database", "a column rename fails the build". If a concrete restatement would require new facts, keep any substantive point and remove unsupported evaluation, or flag it in review mode.
- Shorten or split dense sentences. If the reader has to backtrack to parse a sentence, break it in two. Drop a clause only when it adds no claim.
- Active voice. Prefer it when the source identifies the actor: "the file is parsed by the loader" becomes "the loader parses the file". Do not invent an actor for "queries are validated." Passive is fine when the actor is unknown or genuinely doesn't matter.
- Cut adverbs, or use a stronger verb. "runs quickly" may become "is fast." Replace "significantly improves" with a measured delta only when the source provides one. Remove an adverb only when doing so preserves the claim. Frequent offenders: "really", "just", "literally", "genuinely", "honestly", "simply", "actually", "fundamentally", "crucially".
- Prefer the plain word. "utilize" becomes "use", "leverage" becomes "use", "facilitate" becomes "help", "numerous" becomes "many", "in the event that" becomes "if". The fancier synonym is rarely clearer.
Composition
- Invented concept labels. Do not coin labels such as "the supervision paradox" or "the acceleration trap" to make an observation sound established. Use a name only when the text defines and needs it.
- Fractal summaries. Do not preview, explain, and recap the same point at the document, section, and paragraph levels. Keep the version that helps the reader act or understand.
- Dead metaphor repetition. A useful metaphor can explain one point. Repeating it across the piece turns it into scaffolding for ideas that should stand on their own.
- Historical analogy stacking. Do not add companies or past technology shifts as borrowed authority. When editing an existing stack, preserve its substantive comparisons while removing repetitions that make the same claim.
- One-point dilution. Cut sections that restate the thesis with new wording but add no fact, condition, example, or consequence.
Check for false positives
Do not flag one isolated feature as proof of AI writing. Preserve:
- Deliberate repetition, fragments, contrasts, and punctuation that establish the writer's voice.
- Necessary passive voice when the actor is unknown, irrelevant, or deliberately omitted.
- Domain terminology, legal qualifications, safety notices, and supported uncertainty.
- Quotations, titles, proper names, and text used as an example.
- Useful objections and alternatives that the document names and answers.
- Specific odd details, dated references, mixed feelings, and genuine asides.
Quick checks
Before delivering prose:
- Any adverbs propping up weak verbs? Cut them or fix the verb.
- Any passive voice without a reason? Use the actor as the subject when the
source identifies one.
- Inanimate thing doing a human verb ("the decision emerges")? Name the actor when known and relevant.
- Any "here's what/this/that" throat-clearing? Cut to the point.
- Any formulaic "not X, it's Y" contrasts? State Y directly unless the distinction carries meaning.
- Three consecutive sentences match length? Break one.
- Every paragraph ends with a punchy one-liner? Vary it.
- Repeated or formulaic em dashes? Rewrite them unless they belong to the writer's established style.
- Vague declarative ("The implications are significant")? Use the specific
implication from the source, or remove unsupported significance.
- Meta-joiners ("The rest of this essay...")? Delete. Let the text move.
- Sounds like a pull-quote? Rewrite it.
Scoring
Rate 1-10 on each dimension:
| Dimension | Question |
|---|
| Directness | Statements or announcements? |
| Rhythm | Varied or metronomic? |
| Trust | Respects reader intelligence? |
| Authenticity | Sounds human? |
| Density | Anything cuttable? |
Below 35/50: revise. Keep the score internal unless the user asks for it.
Return the result
- Pasted text. For a rewrite request, return the final rewrite. Add a short change summary only when it helps the user assess the edit. For a review request, return the findings and examples without rewriting unless asked.
- File. Edit only the prose in scope. Preserve code blocks, frontmatter, data, links, and unrelated content, then report the file and the main changes.
- Embedded use. When another task invokes this skill for a message, document, commit message, or pull request description, return only the final text.
Before delivery, ask: "Did the rewrite lose a fact, name, number, qualification, or intended conclusion; add a fact; or retain unsupported promotion or significance?" Treat any such result as an error.
References
- references/phrases.md: throat-clearing openers, emphasis crutches, business jargon table, adverb guidance, meta-commentary, vague declaratives.
- references/structures.md: binary contrasts, negative listing, false agency, narrator-from-a-distance, rhythm patterns.
- references/examples.md: before/after transformations.
- UPSTREAM.md: source material, licenses, and adaptation notes.