Claude Skill

deslopify

Deslopifies AI-sounding prose and code diffs. Use when the user asks to make text sound human, remove ChatGPT-style filler, rewrite without fluff, clean up AI comments, or strip restatement comments from a diff without changing code.

LLM Mart · 0 points · 0 views 0 listing impressions 0 install-command copies
Download kriscard-skills-skills_productivity_deslopify-08395f4.zip · 3 KB

Install

skills CLI npx skills add https://github.com/kriscard/Skills/tree/main/skills/productivity/deslopify
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install kriscard-skills@llmmart
Git git clone https://github.com/kriscard/Skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole kriscard/skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Deslopify

Strip the patterns that mark text as AI-generated. Preserve all factual content, author stance, intended tone, and useful depth — change the delivery, not the message.

The Slop Patterns

Scan for these before rewriting. More slop patterns justify a stronger edit; mostly-clean text gets a light pass. Treat a pattern as slop when it adds no meaning, obscures the point, or makes the reader decode the prose. Keep intentional voice that still communicates clearly.

Opening filler (cut 80% of the time — start with the actual point):

  • "In today's fast-paced world..."
  • "It's important to note that..."
  • "As we navigate the complexities of..."

Hollow transitions:

  • "Furthermore,", "Moreover,", "Additionally," — usually just "Also," or delete
  • "In conclusion," — end without announcing you're ending
  • "It's worth noting that..." — just say the thing

Emphasis inflation:

  • "Importantly,", "Crucially,", "Significantly," — earn emphasis with content
  • "It goes without saying..." — then don't say it

AI vocabulary (replace with the plain word):

  • "leverage" → "use"
  • "utilize" → "use"
  • "delve into" → "look at" or "explore"
  • "in terms of" → restructure the sentence
  • "moving forward" → cut
  • "at the end of the day" → cut

Excessive hedging:

  • "It seems that", "One might argue", "It could be said" — commit to the claim or cut it

False ranges:

  • "from architecture to debugging" implies a scale or complete span that does not exist — list the topics directly
  • Keep ranges whose endpoints define a real scale, sequence, or scope

Empty synonym cycling:

  • Switching among "the service", "the system", "the platform", and "the solution" only to avoid repetition makes identity less clear
  • Pick one accurate term and repeat it unless the terms name real distinctions

Vague attribution:

  • "Experts believe", "studies show", "industry reports suggest" — name the source when the attribution supports the claim
  • Never remove or weaken an attribution because its source is missing. Preserve the attributed wording and flag it for sourcing; do not turn attributed belief into a fact, a hedge, or the author's own claim.
  • Example: keep "Experts believe the migration will cut incidents by 30%" in the revised text and note Source needed for "Experts believe". Do not rewrite it as "The migration may cut incidents by 30%."
  • Keep attribution that is already specific enough to evaluate, including a named source, a cited study, a bounded group such as "three interviewees," or a deliberately anonymous source whose role and reason for anonymity matter.

Sycophantic or chatbot phrasing:

  • "Great question!", "You're absolutely right", "Of course!", "I hope this helps", "Let me know if..." — respond directly
  • Keep genuine warmth or praise when it is part of the author's intended relationship with the reader

Nominalization (convert to verb):

  • "make a decision" → "decide"
  • "provide assistance" → "help"
  • "conduct an analysis" → "analyze"
  • "have a discussion" → "discuss"

Generic conclusions:

  • "The future looks bright" or a recap that adds no decision, implication, or next step — cut it
  • Keep conclusions that synthesize evidence, state a decision, or tell the reader what follows

Harmful over-compression:

  • Dropped articles, verbless fragments, unexplained abbreviations, and symbol-speak such as "invalid date → exit 2, no write" make the reader reconstruct the sentence
  • Restore the connective meaning with complete clauses: "An invalid date exits with code 2 and writes nothing."
  • Keep concise labels, headings, interface copy, and established notation when their context makes them clear

Mannered metaphor:

  • Stock aphorisms, personified code, and decorative phrases such as "a dial worth turning" can hide the mechanism — name the literal action or property
  • Keep metaphors that clarify a hard idea or belong to the author's deliberate creative voice

Bullet lists where prose would be clearer: not every list needs to be a list. Three related sentences often read better as a paragraph.

En-dash used as em-dash: – should be — (or just a comma)

What Good Looks Like

  • Short sentences land hard. Longer ones build context. Vary them.
  • Prefer active voice when naming the actor improves the sentence: "We shipped X" rather than "X was shipped by the team." Passive voice remains useful when the actor is unknown, irrelevant, or intentionally de-emphasized.
  • Specific beats general: "increased conversion by 18%" not "improved results"
  • The first sentence is the most important — if it's filler, cut it

Process

  1. Read the full text.
  2. Identify which slop patterns appear; mention only the main classes in the note.
  3. Rewrite: cut filler, verb-ify nominalizations, and clarify passive constructions only when naming the actor helps.
  4. Present the revised version or cleaned diff.
  5. Add a 2–3 line note on the key changes made, not a list of every micro-change.

Output Criteria

Prose mode: return the revised text plus a 2–3 line note naming the main classes of edits. Do not list every micro-change.

Code comment mode: return a cleaned diff or edit only comments, then report what was removed. Be surgical: no code, control-flow, formatting, or behavior changes.

Code Comment Mode

When the input is a diff or the user asks to remove AI comments from code:

What to remove:

  • Comments restating what code clearly does (// increment counter above counter++)
  • Overly verbose explanations of obvious code
  • Generic placeholders (// TODO: implement, // Handle error)
  • Comments with AI phrasing ("This function...", "The following code...")
  • Inline docs inconsistent with the file's existing comment style

What to keep:

  • Doc comments for public APIs (JSDoc, docstrings)
  • Comments explaining why (business logic, workarounds, edge cases)
  • Comments matching the file's existing style and density
  • License headers and attribution

Hard constraint: DO NOT modify any code. Only remove comments. No refactoring, no style changes, no removing try/catch or null checks. If you spot code slop, note it in the summary but leave it.

Hard Rules (Prose Mode)

  • Never change factual content — only how it's expressed
  • Never add new claims or examples not in the original
  • Preserve the author's stance, intended audience, and recognizable voice
  • Preserve useful connective explanation; concise does not mean compressed beyond easy reading
  • If the text is mostly fine, say so — don't invent problems to fix
Files (skills)
  • SKILL.md 6.8 KB
    ---
    name: deslopify
    description: >-
      Deslopifies AI-sounding prose and code diffs. Use when the user asks to make
      text sound human, remove ChatGPT-style filler, rewrite without fluff, clean up
      AI comments, or strip restatement comments from a diff without changing code.
    ---
    
    # Deslopify
    
    Strip the patterns that mark text as AI-generated. Preserve all factual content, author stance, intended tone, and useful depth — change the delivery, not the message.
    
    ## The Slop Patterns
    
    Scan for these before rewriting. More slop patterns justify a stronger edit; mostly-clean text gets a light pass. Treat a pattern as slop when it adds no meaning, obscures the point, or makes the reader decode the prose. Keep intentional voice that still communicates clearly.
    
    **Opening filler** (cut 80% of the time — start with the actual point):
    
    - "In today's fast-paced world..."
    - "It's important to note that..."
    - "As we navigate the complexities of..."
    
    **Hollow transitions**:
    
    - "Furthermore,", "Moreover,", "Additionally," — usually just "Also," or delete
    - "In conclusion," — end without announcing you're ending
    - "It's worth noting that..." — just say the thing
    
    **Emphasis inflation**:
    
    - "Importantly,", "Crucially,", "Significantly," — earn emphasis with content
    - "It goes without saying..." — then don't say it
    
    **AI vocabulary** (replace with the plain word):
    
    - "leverage" → "use"
    - "utilize" → "use"
    - "delve into" → "look at" or "explore"
    - "in terms of" → restructure the sentence
    - "moving forward" → cut
    - "at the end of the day" → cut
    
    **Excessive hedging**:
    
    - "It seems that", "One might argue", "It could be said" — commit to the claim or cut it
    
    **False ranges**:
    
    - "from architecture to debugging" implies a scale or complete span that does not exist — list the topics directly
    - Keep ranges whose endpoints define a real scale, sequence, or scope
    
    **Empty synonym cycling**:
    
    - Switching among "the service", "the system", "the platform", and "the solution" only to avoid repetition makes identity less clear
    - Pick one accurate term and repeat it unless the terms name real distinctions
    
    **Vague attribution**:
    
    - "Experts believe", "studies show", "industry reports suggest" — name the source when the attribution supports the claim
    - Never remove or weaken an attribution because its source is missing. Preserve the attributed wording and flag it for sourcing; do not turn attributed belief into a fact, a hedge, or the author's own claim.
    - Example: keep "Experts believe the migration will cut incidents by 30%" in the revised text and note `Source needed for "Experts believe".` Do not rewrite it as "The migration may cut incidents by 30%."
    - Keep attribution that is already specific enough to evaluate, including a named source, a cited study, a bounded group such as "three interviewees," or a deliberately anonymous source whose role and reason for anonymity matter.
    
    **Sycophantic or chatbot phrasing**:
    
    - "Great question!", "You're absolutely right", "Of course!", "I hope this helps", "Let me know if..." — respond directly
    - Keep genuine warmth or praise when it is part of the author's intended relationship with the reader
    
    **Nominalization** (convert to verb):
    
    - "make a decision" → "decide"
    - "provide assistance" → "help"
    - "conduct an analysis" → "analyze"
    - "have a discussion" → "discuss"
    
    **Generic conclusions**:
    
    - "The future looks bright" or a recap that adds no decision, implication, or next step — cut it
    - Keep conclusions that synthesize evidence, state a decision, or tell the reader what follows
    
    **Harmful over-compression**:
    
    - Dropped articles, verbless fragments, unexplained abbreviations, and symbol-speak such as "invalid date → exit 2, no write" make the reader reconstruct the sentence
    - Restore the connective meaning with complete clauses: "An invalid date exits with code 2 and writes nothing."
    - Keep concise labels, headings, interface copy, and established notation when their context makes them clear
    
    **Mannered metaphor**:
    
    - Stock aphorisms, personified code, and decorative phrases such as "a dial worth turning" can hide the mechanism — name the literal action or property
    - Keep metaphors that clarify a hard idea or belong to the author's deliberate creative voice
    
    **Bullet lists where prose would be clearer**: not every list needs to be a list. Three related sentences often read better as a paragraph.
    
    **En-dash used as em-dash**: `–` should be `—` (or just a comma)
    
    ## What Good Looks Like
    
    - Short sentences land hard. Longer ones build context. Vary them.
    - Prefer active voice when naming the actor improves the sentence: "We shipped X" rather than "X was shipped by the team." Passive voice remains useful when the actor is unknown, irrelevant, or intentionally de-emphasized.
    - Specific beats general: "increased conversion by 18%" not "improved results"
    - The first sentence is the most important — if it's filler, cut it
    
    ## Process
    
    1. Read the full text.
    2. Identify which slop patterns appear; mention only the main classes in the note.
    3. Rewrite: cut filler, verb-ify nominalizations, and clarify passive constructions only when naming the actor helps.
    4. Present the revised version or cleaned diff.
    5. Add a 2–3 line note on the key changes made, not a list of every micro-change.
    
    ## Output Criteria
    
    **Prose mode:** return the revised text plus a 2–3 line note naming the main classes of edits. Do not list every micro-change.
    
    **Code comment mode:** return a cleaned diff or edit only comments, then report what was removed. Be surgical: no code, control-flow, formatting, or behavior changes.
    
    ## Code Comment Mode
    
    When the input is a diff or the user asks to remove AI comments from code:
    
    **What to remove:**
    
    - Comments restating what code clearly does (`// increment counter` above `counter++`)
    - Overly verbose explanations of obvious code
    - Generic placeholders (`// TODO: implement`, `// Handle error`)
    - Comments with AI phrasing ("This function...", "The following code...")
    - Inline docs inconsistent with the file's existing comment style
    
    **What to keep:**
    
    - Doc comments for public APIs (JSDoc, docstrings)
    - Comments explaining why (business logic, workarounds, edge cases)
    - Comments matching the file's existing style and density
    - License headers and attribution
    
    **Hard constraint: DO NOT modify any code.** Only remove comments. No refactoring, no style changes, no removing try/catch or null checks. If you spot code slop, note it in the summary but leave it.
    
    ## Hard Rules (Prose Mode)
    
    - Never change factual content — only how it's expressed
    - Never add new claims or examples not in the original
    - Preserve the author's stance, intended audience, and recognizable voice
    - Preserve useful connective explanation; concise does not mean compressed beyond easy reading
    - If the text is mostly fine, say so — don't invent problems to fix
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.