anti-ai-slop

name

anti-ai-slop

description

Universal de-AI writing skill. Catches and rewrites the tells that expose AI-generated text — forbidden words ("delve", "crucial", "landscape", "tapestry", "Additionally"), forbidden patterns (rule of three, negative parallelism, em-dash overuse, Title Case headings, inline-header bullets, emoji headers, knowledge-cutoff disclaimers), and AI structure (Despite-its-X formula, "is/are" replaced by "serves as / represents"). Works on any medium: landing page copy, emails, DMs, LinkedIn posts, proposals, documentation, microcopy, CTAs, FAQ. The references in this folder are the canonical source — read them, then use your own judgment for the rewrite. TRIGGERS — Use this skill when the user says: - "de-ai this" / "de-ai-ify" / "remove the AI tells" - "make it sound human" / "humanize this copy" / "sound less AI" - "is this AI slop?" / "does this read like AI?" - "rewrite without AI words" / "scrub the AI from this" - "anti-ai" / "anti-slop" / "no slop" - "check this for AI tells" / "audit this copy" - Before publishing any external copy (LP, email, proposal, post) as a final gate - When another skill (design-ui, content, proposal, linkedin-hooks) is about to return copy

anti-ai-slop

Two modes. Pick the matching one. Both pull from references/ as the canonical pattern list. You are the final judge — the references guide you, but the rewrite is your call.

Modes

Mode	When	Output
1. Audit	"is this AI?" / "what tells are in this?"	Annotated list of every tell, ranked by severity, with the exact word/phrase and the rule it breaks. No rewrite.
2. Rewrite	"de-ai this" / "humanize this"	Clean rewrite that preserves meaning and length, removes every tell, returns the rewritten text plus a short diff log of what was changed and why.

References

File	What it contains
references/quick-scan-checklist.md	Fast in-flight checklist — load for short copy (under 200 words)
references/replacement-table.md	Find/replace dictionary for the most common AI phrases
references/by-medium.md	Medium-specific rules: LP, email, DM, LinkedIn post, proposal, docs, microcopy

How to run

Mode 1 — Audit

1. Load references/quick-scan-checklist.md.
2. For long copy, also load references/replacement-table.md and references/by-medium.md.
3. Read the input. For every tell found, output a row: | line/excerpt | tell | rule broken | severity (high/med/low) |
4. End with a one-line verdict: clean / minor / heavy slop.
5. Do not rewrite unless asked.

Mode 2 — Rewrite

1. Load references/quick-scan-checklist.md and references/replacement-table.md.
2. Detect the medium from context (LP / email / DM / post / docs / microcopy). Load the matching section of references/by-medium.md.
3. Rewrite preserving:
   • Meaning (no new facts, no dropped facts)
   • Approximate length (within ±15%)
   • The author's voice if a style file is referenced (e.g., walid-writing-style.md)
4. Return:
   • The rewritten text
   • A short diff log: 5–10 bullet points listing what changed and which rule fired
   • A self-check confirming zero forbidden words remain

The hard bans (apply to every output)

Forbidden words — never appear in the rewrite: Additionally (opener) · align with · crucial · delve · emphasizing · enduring · enhance · fostering · garner · highlight (verb) · interplay · intricate · intricacies · key (adjective) · landscape · pivotal · showcase · tapestry · testament · underscore · valuable · vibrant · serves as · stands as · It is worth noting · rich tapestry · evolving landscape · seamlessly · leverage (as verb) · empower · comprehensive · holistic · robust solution · cutting-edge · game-changer

Forbidden patterns — never appear in the rewrite:

• Rule of three (three adjectives, three benefits, three phrases in a row)
• Negative parallelism: "not just X, but Y" / "not X, but Y"
• Em dashes — like this — (use commas, colons, or split the sentence)
• "Despite its X, Y faces challenges" formula
• Title Case in headings (sentence case only)
• Inline-header bullets: • **Label:** description
• Emoji in headings or section markers
• "Let's explore" / "Here is your" / "As we can see"
• Knowledge-cutoff disclaimers ("As of my last training update...")
• "is/are/has" replaced by "serves as / represents / marks / features / boasts"
• Bolding more than 2 words per paragraph for "emphasis"
• Synonym chains (calling the same person 4 different ways in a paragraph)
• "from X to Y" when X and Y don't form a real scale

Self-check before returning (rewrite mode)

Run silently and confirm in the diff log:

• Zero forbidden words found via case-insensitive scan
• Zero em dashes
• Zero Additionally / In conclusion / In summary openers
• No paragraph has three parallel items in a row
• No heading is in Title Case
• No "not just X, but Y" pattern
• Length within ±15% of original
• All proper nouns and numbers preserved exactly

If any check fails, fix and re-check before returning.

Chains into

• design-ui — applies anti-ai-slop to every headline, microcopy, CTA, FAQ generated
• content — applies before publishing LinkedIn posts
• linkedin-hooks — applies after hook generation
• proposal / proposal-builder — applies to every section of the proposal HTML
• cold-dance / personalized-value-outreach — applies to every DM
• landing-page-copy — applies to every section
• de-ai-ify — alias for Mode 2

When NOT to apply

• Code, JSON, YAML, CLI commands — leave technical strings alone
• Direct quotes from external sources — preserve verbatim
• Legal copy where exact wording is required — flag, don't rewrite
• The user explicitly asks for "AI-style" or "formal corporate tone"