skill
deslop-copy
Rewrite a project's copy to read human, specific, and short — a Claude Code skill that hunts the shape of AI writing and kills it.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/JaimeOrtegaxyz/deslop-copy itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 23, 2026deslop-copy
A Claude Code skill that rewrites your product's copy so it stops reading like ChatGPT.
Run it in-session while you're building a site or app. It reads the codebase to
learn what the product is and who it's for. Then it flags every line that reads
AI or fails to sell the idea, and hands you a section-by-section rewrite to
approve. It started as a teardown of the "5 viral copywriting skills" going
around. The parts worth keeping are in references/persuasion.md.
How it works
v2 was instructions only, and it shipped lines its own catalog bans. So v3 added two checks that run after the writing:
- Reads the codebase first. README, docs, brand files, and wherever the copy
actually lives:
src/content, section components, MDX, SEO meta. Builds a one-paragraph product model, and asks you 2–3 questions if the concept is still unclear. - Runs the linter on every line.
deslop-lint.pycatches the mechanical tells: negation-contrast, tricolons, setup→reveal, vague enders, the slop-vocabulary blocklist, each with a fix hint. The model then hunts what regex misses, mostly balanced mirrors and lines whose only job is sounding smart. - Gates its own rewrites. Every rewritten line has to pass
deslop-lint.py --review COPY-REVIEW.mdwith zero errors. There is no override. Disputed hits go in a list for you to rule on. - A second model judges it. A subagent with no conversation history reads the shape catalog and the finished lines, then tries to refute each one. Failures get rewritten and re-gated. Anything still failing after two rounds is reported to you as unresolved.
- Never invents proof. When a rewrite needs a number or a customer name you
haven't given it, it leaves a visible
‹NEEDS: …›flag and asks. - Writes
COPY-REVIEW.mdat the project root: diagnosis, before/after per section, proof gaps, and alint/judgestatus line. You tick the rewrites you want and it edits the real files.
The linter runs on its own too. Python 3, no dependencies:
python3 scripts/deslop-lint.py page-copy.txt # or --review COPY-REVIEW.md
python3 scripts/deslop-lint.py --self-test # 45 regression cases
Install
git clone https://github.com/JaimeOrtegaxyz/deslop-copy.git ~/.claude/skills/deslop-copy
Or project-scoped: clone into .claude/skills/deslop-copy/.
See SKILL.md for the full method and references/ for the slop blocklist, persuasion principles, and hook bank.
Credits
The lint-gate architecture comes from woosal1337's
"The cure for AI slop is a 1986 aircraft manual":
check the model's writing with a deterministic linter instead of trusting the
prompt. That kit's ste-lint.py and cross-model experiments showed a
machine-checkable writing system cuts slop where a banned-words list doesn't.
deslop-lint.py applies the same idea to marketing copy with its own rule set.