Install
openclaw skills install @skyzer/deslop-the-copyDeslop, humanize, edit, or audit AI-sounding, robotic, generic, corporate, or ChatGPT-like copy while preserving the writer's voice.
openclaw skills install @skyzer/deslop-the-copyEdit AI-sounding writing into clear human prose without replacing the writer's voice with a different generic voice.
Keep the meaning, factual claims, attitude, and level of formality intact. Make the minimum effective edit. Leave strong human sentences alone.
One rule overrides every stylistic preference: never return an em dash. Use a comma, period, colon, semicolon, parentheses, or a rewritten sentence. This applies to edited copy, explanations, and quoted evidence.
Use edit mode by default when the user wants text rewritten, cleaned up, humanized, or deslopped.
Return the cleaned text. Do not explain the changes unless the user asks. For a sentence or two, use a light touch.
Use detect mode when the user asks to audit, scan, flag, or judge a draft without rewriting it.
For every pattern found:
Do not rewrite the full draft, assign a slop score, or claim that AI wrote it. Named patterns are evidence; authorship detectors are guesses. If quoted source text contains an em dash, replace the character with [em dash] in the quotation.
If the user has not provided text, ask for it. Ask about audience, format, or intended effect only when the missing context could materially change the edit.
Read the whole draft and identify internally:
Do not reorganize the draft unless its structure blocks understanding. Do not invent facts, examples, statistics, sources, opinions, or confidence.
Cut claims that tell the reader something matters instead of showing why:
Keep the concrete fact and let the reader judge its importance.
Cut sentences that only announce importance, simplicity, contrast, scope, or what the next paragraph will explain. Common examples include "That distinction matters," "This matters because," "The practical mental model is simple," "The key point is," "The same principle applies," and "This is why it matters."
Delete the sentence if it adds no fact, argument, or useful tension. If it contains a real point, merge that point into the surrounding sentence.
Replace inflated words such as "vibrant," "groundbreaking," "stunning," "breathtaking," "cutting-edge," and "game-changing" with specific details, or remove them.
Replace "experts agree," "studies show," "industry reports suggest," and similar claims with a named source. If no source exists, remove or flag the claim. Never invent attribution.
Cut trailing clauses built around "highlighting," "underscoring," "showcasing," "reflecting," "emphasizing," or "fostering" when they merely gesture at meaning. State the real consequence when the draft supports one.
Prefer plain language over habitual AI words such as:
Do not replace a precise technical term just because it appears on this list.
Prefer "is," "has," and direct verbs over "serves as," "stands as," "functions as," "acts as," "made a decision," or "has the ability to."
Use active voice when it makes the actor and action clearer. Do not force active voice when the actor is unknown or irrelevant.
Remove templates such as:
State the real point directly.
Cut setups that delay or inflate the point:
Keep a personal admission or aside only when it adds character, context, or honest uncertainty.
Rewrite dramatic noun-phrase reveals such as "The best part: it learns" or "The real problem: distribution." Use colons for lists, labels, explanations, and quotations, not manufactured suspense.
Remove "What if I told you," "Think about it," "Plot twist," and self-answered question-and-answer pairs when a direct statement works better.
Fix stacked fragments such as "That's it. That's the whole thing." and "X. And Y. And Z." Keep fragments that genuinely match the writer's natural cadence and do not create fake drama.
Do not force ideas into neat trios. Remove the third item when it is redundant or exists only for rhythm.
Use the same clear noun consistently instead of rotating among near-synonyms to avoid repetition.
Replace "from X to Y" when X and Y are not endpoints on a meaningful scale. Use "X and Y" or name the actual scope.
Compress empty phrases:
Remove stacked hedges. Keep "I think," "maybe," "probably," and similar language when it expresses real uncertainty or belongs to the writer's voice.
Break repeated sentence shapes, identical paragraph structures, and overly tidy parallelism. Do not replace them with a stack of manufactured punchy fragments. Let rhythm serve the point and the writer.
Delete the cute metaphor, aphorism, or mic-drop line when it adds no information. End on the last concrete point, useful takeaway, or next action.
Remove "In conclusion," "Ultimately," "Overall," and final paragraphs that only repeat what the reader just read.
Remove decorative formatting that does not help the content:
Remove every em dash from edited text. Replace it with punctuation or rewrite the sentence. In detect mode, flag every source occurrence as the "Em dash" pattern and render the character as [em dash] in quoted evidence.
Remove:
After removing slop, prevent the edit from becoming sterile or uniformly polished.
The result should sound like this writer on a good day, not like a universal "human" persona.
Read references/eval.md and check the full output against it. If any check fails, revise and run the audit again.
Before returning, scan the actual response for the Unicode em-dash character. There must be zero occurrences.
For edit mode, return only the cleaned text unless the user asks for commentary.
For detect mode, return the named findings with quoted evidence and concise fixes. If no listed pattern appears, say that plainly without inventing problems.
Preserve meaning above all.