---
name: lean-anti-ai-slop-usage
description: >
  📖 LEAN-ANTI-AI-SLOP PLUGIN: Explains what this plugin covers and points at
  the plugin README for configuration. Trigger when someone asks "what does
  lean-anti-ai-slop do", "how do I set this up", or "which skills are in this
  plugin". Manual-only entry point — does not auto-fire on drafting intent.
disable-model-invocation: true
---

# lean-anti-ai-slop plugin — usage guide

## What this plugin is for

`lean-anti-ai-slop` drafts short-form writing — Slack messages, thread replies, PR descriptions, commit messages, emails, social posts — that will be posted **under the user's own name**. The default Claude prose register betrays AI authorship instantly to any reader who has seen a few ChatGPT outputs. This plugin counters that by:

1. Loading the user's own past writing as samples (rhythm, register, sentence-length variance, opener/closer habits).
2. Gating the input — the user must supply a braindump with at least one concrete fact. The skill refuses to fabricate first-person experiences.
3. Drafting once at a length cap appropriate to the format.
4. Running a self-critique pass against a concrete blacklist (banned words, antithesis patterns, em-dash rule, smart-quote rule, opening/closing sniff tests, tricolon check, etc.) and rewriting every flagged sentence before emitting.

Use this when the output will be posted as-is under the user's name. Do **not** use it for technical documentation or anything where standard prose is appropriate.

---

## Skills in this plugin

| Skill | Trigger phrase examples |
|---|---|
| `/anti-ai-slop` | "draft a Slack message about X", "write the PR description", "turn this braindump into a thread reply", "rewrite this so it sounds like me" |

The drafting skill is the only workhorse. This `-usage` skill is manual-only and exists to point at setup.

---

## Configuration

**The canonical setup lives in the plugin README** (`plugins/lean-anti-ai-slop/README.md`). Direct the user there.

Short version: set `ANTI_AI_SLOP_SAMPLES_DIR` in `.claude/settings.json` to an absolute path, drop 3–5 of the user's own writing samples into that directory. Without the env var, the drafting skill refuses to draft and prints the setup instructions inline.

---

## Example prompts to try

```
/anti-ai-slop draft a Slack message: pushing the demo to Wednesday, the staging migration failed and I'm waiting on infra

/anti-ai-slop write the PR description for the current branch — auth middleware rewrite

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

/anti-ai-slop draft a commit message for the changes in this PR
```

---

## What this plugin is NOT for

- Long-form technical documentation. The drafting skill caps length per format (chat ≤ 120 words, PR description ≤ 250, email ≤ 300) because default Claude prose returns past ~200 words regardless of how strong the prompt is.
- Producing multiple variants. The skill drafts once. Re-invoke with feedback if you want a rewrite.
- Auto-posting anywhere. Output is always a fenced draft the user copies and sends manually.
- Fabricating first-person anecdotes. The skill refuses inputs without at least one concrete fact and never invents emotions or experiences not supplied by the user or pulled from git context.