Skip to main content

The anti-slop writing skill

A single file that makes any agent's writing read like a person wrote it, on any channel, in any repo. It bans the tells of AI "slop" (em-dash overuse, delve/tapestry/robust filler, the "it's not just X, it's Y" construction, throat-clearing intros, restated conclusions), enforces natural prose, lets you tune it to your own voice, and makes the agent self-review before it hands anything back.

It's the portable, standalone version of what the Reactor engine does at full strength, where the same discipline is applied automatically, tuned to your brand voice, and checked against everything you've already published.

Install​

The skills package is the default path - it installs into whichever agents you use (Claude Code, Codex, Cursor, any Agent-Skills-standard harness) and keeps the copy updatable:

npx @reactor-tools/skills add anti-slop

Or run npx @reactor-tools/skills with no arguments to pick skills and agents interactively.

Prefer the raw file? Download SKILL.md and place it yourself - e.g. for Claude Code:

mkdir -p .claude/skills/anti-slop
curl -o .claude/skills/anti-slop/SKILL.md https://docs.reactor.tools/anti-slop/SKILL.md

It's self-contained, so it also works pasted into a system prompt or a CLAUDE.md / AGENTS.md rules file.

Using it​

The skill is model-invoked: once installed, the agent applies it whenever it writes or edits prose, no prompt needed. You can also fire it explicitly (/anti-slop in Claude Code, $anti-slop in Codex) or by asking: "Apply the anti-slop skill to this draft."

What it enforces​

The skill is built from current, sourced best practice on detectable AI patterns. In short, it:

  • Bans the punctuation tells: no em dashes, straight quotes only, no bold-term list padding, no emoji bullets.
  • Bans the constructions: "it's not just X, it's Y", "stands as a testament to", the forced rule of three, "in today's fast-paced world", "in conclusion", and prompt-restating.
  • Bans the vocabulary: a long, consolidated blacklist (delve, tapestry, testament, boasts, leverage, robust, seamless, pivotal, vibrant, groundbreaking, moreover…).
  • Adds positive technique: vary sentence length (burstiness), be specific with real numbers and names, plain verbs, contractions, active voice, cut the preamble and the summary.
  • Tunes to your voice: a VOICE.md pattern (samples, a preferred/banned lexicon, register, like-this/not-like-this pairs) so the output sounds like you, not like clean-but-generic AI.
  • Self-reviews: a mandatory six-step pass (grep your own tells, rewrite the three most AI-sounding sentences, read aloud, check burstiness, kill intro/outro, cut 10%) before returning anything.

Use it to audit, too​

Point it at existing content: "Audit this against the anti-slop skill." The agent will flag each violation with the rule it breaks, suggest a rewrite, and return a cleaned version, handy in CI or a pre-publish check.

Want it tuned and automatic?​

This skill is deliberately generic so it works anywhere. Reactor is the same idea taken all the way: the anti-slop rules become a weighted part of a brand engine that also holds your voice, positioning, audiences, and competitive landscape, applies them on every generation, checks each draft against your recent posts so nothing repeats, and closes the loop with analytics. Same principle, tuned to you and wired into publishing.