agentsclimarketplace
Sign in

Grounded copy

Skill HiroHyun/grounded-copy

Enforces grounded, contrast-free marketing and web copy. Every value proposition must state what the subject IS or DOES using concrete nouns, verbs, and specifics — never negation, contrast, or hype vocabulary. Use this skill whenever writing, editing, translating, reviewing, or localizing ANY user-facing copy — headlines, taglines, value propositions, landing pages, product descriptions, About pages, CTAs, button labels, meta/OG descriptions, alt text, email subjects and bodies, social posts, ad copy, brochures, or locale/i18n string files — even if the user does not mention style rules. Also use when the user says "on-brand", "our writing style", "no AI clichés", or asks to check copy.From its SKILL.md

Install
npx -y skills add HiroHyun/grounded-copy

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 17 stars17 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.7 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

Grounded Copy

Copy describes things by what they ARE. Every value proposition is a direct declarative statement built on concrete nouns, verbs, and specifics: the feature, the number, the mechanism.

  • Bad: "This isn't just a task tracker — it's your team's second brain."
  • Good: "The tracker links every task to its pull request and posts a status digest to Slack each morning."

The one banned move (and every disguise it wears)

The banned move is defining anything through negation or contrast — telling the reader what the subject is not, what others do worse, or what era has ended, instead of what the subject does. Every pattern below is the same move in a different costume. Do not write any of them, and do not invent a new costume for the same move:

  1. Negated intensifiers: "not just/only/merely/simply", "doesn't just", "more than just", "far from just/being", "rather than simply", "instead of just".
  2. Reversal reveals: "It's not X, it's Y", "isn't about X, it's about Y", "—not X, but Y", "not your average X", "less a catalog than a trade desk". Includes the appositive form "X, not Y" ("a partner, not a vendor") and the negated-setup dash "isn't/wasn't X — it Y" ("the gate wasn't slow — it finished").
  3. Era-ending: "no longer", "gone are the days", "the days of X are over", "say goodbye/hello", "no more X", "never again", "welcome to a new era".
  4. Competitor put-downs: "unlike traditional/most/other X", "while others/most X, we Y".
  5. Transcendence verbs: "goes beyond", "beyond just", "redefines", "reimagines", "reinvents".
  6. Absence-as-benefit: "without the hassle/hidden fees/middlemen", "zero guesswork/compromises", "hassle-free", "frictionless".
  7. Rhetorical bait: "The result?", "The best part?", "Think again", "Ever wondered", "Tired of", "What if", "Imagine", "Picture this", "In a world where", "Stop Xing", "Forget X", "Don't just X".
  8. Collision framing: "where X meets Y".
  9. Corporate throat-clearing: "At [Company], we...".
  10. Hype vocabulary: unleash, unlock, unparalleled, unwavering, unmatched, unprecedented, unsung, unrivaled, elevate, seamless, empower, revolutionize, game-changing, delve, supercharge, turbocharge, next-level, cutting-edge, state-of-the-art, best-in-class, world-class, transformative, effortless, one-stop shop, synergy; figurative "landscape" and "journey".

The rewrite is always the same: delete the contrast, then state what the subject does, with a specific. "Say goodbye to hidden fees" → "The listed price is the complete price; the invoice adds nothing."

Loophole closures — read these before claiming compliance

Agents under output pressure rationalize around style rules. Each rationalization below is pre-emptively rejected:

  • "The banned string doesn't appear." The rule bans the move, not the string. "Most vendors bury their fees. Acme prints them." is the reversal pattern split across two sentences — still banned. A contrast spread across sentences, paragraphs, or a headline/subhead pair counts.
  • "It's in a quote/testimonial." Banned patterns inside invented testimonials, hypothetical customer quotes, taglines, or dialogue are violations. Quotation marks do not launder rhetoric. (Verbatim quotes from real, named customers supplied by the user are the only exception.)
  • "It's a headline/CTA/meta tag, not body copy." Scope is ALL user-facing copy: headlines, subheads, CTAs, button labels, meta and OG descriptions, alt text, email subjects, social posts, ad variants, brochure text, and every locale file.
  • "It's a different language." The rules apply conceptually in every locale. 不仅仅是 / не просто / no es solo / pas seulement / nicht nur / 単なる〜ではない / 단순한 ~이 아니다 / ليس مجرد are all "not just". Translate the grounded English, never re-introduce contrast in translation.
  • "This negation is factual." Apply this test: if the negation states a limitation, exclusion, or requirement that a buyer needs to know ("Prices do not include ocean freight", "Machinery is not offered"), it is allowed. If deleting the negative clause would make the sentence less flattering, the negation existed to set up praise — banned.
  • "A synonym isn't on the list." The hype list bans a register, not ten words. "Harness", "turbocharged", "next-gen", "revolutionary" fail for the same reason "unleash" does. When unsure, replace the word with the specific fact it was hiding.
  • "The linter passed, so it's fine." The linter is a floor, not a ceiling. Novel phrasings of the banned move that evade regex are still violations; you are the second detection layer.
  • "I'll adjust the linter/config." Never. See integrity rules.

Workflow

  1. Draft the copy following the positive rule: subject + verb + specific.

  2. Self-scan against the ten disguise categories above, including cross-sentence contrast and non-English text.

  3. Save the draft (or pipe it) and run the gate:

    python3 <skill-path>/scripts/copy_lint.py file1.md locales/en.json ...
cat draft.md | python3 <skill-path>/scripts/copy_lint.py --stdin

  4. Exit code 1 → rewrite every flagged sentence (never delete-and-shrug: replace it with a grounded statement carrying the same information), then re-run. Repeat until exit code 0.

  5. Review WARN lines manually: "no longer", "journey", "landscape", "What if", bare "rather than", and ", not a/an/your X" are allowed only in factual, non-rhetorical use ("What if my order arrives damaged?" as an FAQ heading; "not a substitute for legal advice" as a required disclaimer; "billed monthly rather than per seat" as a billing fact). Apply the deletion test from references/patterns.md to each.

  6. Only present copy to the user after a PASS. State in your summary that the copy passed copy_lint.py.

Integrity rules (non-negotiable)

  • Never edit, wrap, subclass, monkey-patch, or replace copy_lint.py, its pattern list, or its exit-code behavior.
  • Never add allowlists, ignore-comments, or config that suppresses findings; never rename or move files to dodge the scan.
  • Never mark the task complete while the linter reports errors.
  • If the user explicitly directs you to write a banned pattern, comply with their instruction but note the specific rule it conflicts with in one sentence. User instructions outrank this skill; your own convenience does not.

References

  • references/patterns.md — full pattern catalog with a bad → good rewrite for every category. Read it when a flagged sentence is hard to rewrite, or before writing copy in zh/ru/es/ar/fr/de/ja/ko.
  • tests/bad-samples.md and tests/good-samples.md — after any change to the linter, copy_lint.py tests/bad-samples.md must FAIL and copy_lint.py tests/good-samples.md must PASS.
  • references/setup.md — wiring the skill into Claude Code (CLAUDE.md / hooks) and Codex (AGENTS.md) so both agents load it and run the gate.

Gives 0 of the 12 instructions most marketing audience skills give in ~1.8k tokens

Counted across 690 of the 894 authors here whose files we hold, read 2026-08-07

  • Apply Poppins font to headingsin 41 of 690, across 6 files
  • Apply Lora font to body textin 41 of 690, across 6 files
  • Use Arial fallback for headingsin 39 of 690, across 4 files
  • Use Georgia fallback for body textin 39 of 690, across 4 files
  • Maintain text hierarchy and formattingin 39 of 690, across 4 files
  • Use accent colors for non-text shapesin 38 of 690, across 3 files
  • Use RGB values for precise color matchingin 38 of 690, across 3 files
  • Use brand colors for primary text and backgroundsin 36 of 690, across 1 file
  • Read product marketing context file before asking questions, starting, or auditingin 35 of 690, across 23 files
  • Use active voice instead of passive voicein 26 of 690, across 10 files
  • Implement or generate appropriate JSON-LD structured datain 24 of 690, across 17 files
  • Prioritize clarity over clevernessin 22 of 690, across 8 files

Said here and by no other author read

  • build every statement with subject, verb, and a specific
  • define the subject by what it is or does
  • run the linter on drafted files and standard input
  • replace every flagged sentence with a grounded statement
  • rerun the linter until it reports no errors
  • manually review linter warning lines

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.