lean-anti-ai-slop

Writes things under the user's own name — Slack messages, thread replies, PR descriptions, commit messages, emails, social posts — without the default LLM tells that betray AI authorship.

Included assets

• skills/lean-anti-ai-slop-usage/SKILL.md — explains what this plugin does, how to configure it, and when to invoke it
• skills/anti-ai-slop/SKILL.md — the drafting workflow: sample load, fact gating, draft, anti-slop self-critique pass
• skills/anti-ai-slop/references/anti-slop.md — concrete blacklist (banned words, antithesis patterns, em-dash rules) used during the self-critique pass

Skills

Skill	What it does
/lean-anti-ai-slop-usage	Explains the plugin and points at this README for setup
/anti-ai-slop	Drafts a single message/post in the user's voice and self-checks against the anti-slop blacklist

Setup

This file is the canonical source for setup instructions. The skills link here instead of duplicating the steps.

The drafting skill reads writing samples from a directory you configure with the ANTI_AI_SLOP_SAMPLES_DIR env var, set in .claude/settings.json (project) or ~/.claude/settings.json (user-wide):

{
  "env": {
    "ANTI_AI_SLOP_SAMPLES_DIR": "/absolute/path/to/writing-samples"
  }
}

Drop 3–5 of your own writing samples into that directory, each in its own file. Any format works — exported Slack messages, PR descriptions copied from GitHub, sent emails. The skill loads the most-recently-modified files on every invocation and matches rhythm/register before emitting.

If the env var is unset when you invoke /anti-ai-slop, the skill stops and tells you how to configure it instead of drafting.

Example prompts

/anti-ai-slop draft a Slack message saying I'm pushing the demo to Wednesday because the staging migration failed

/anti-ai-slop write the PR description for this branch — it's the auth middleware rewrite

/anti-ai-slop turn this braindump into a thread reply: <paste>

Notes

• Never auto-posts. Output is always a fenced draft that you copy and send yourself.
• Samples beat persona priming. Without samples in ANTI_AI_SLOP_SAMPLES_DIR, the skill refuses to draft — generic "write casually" prompts revert to default Claude prose within ~2 sentences.
• One draft per invocation. Re-invoke with feedback if you want a revision; the skill does not produce variants.