Skip to content

setup-guide.md

Latest commit

 

History

History
131 lines (92 loc) · 5.08 KB

setup-guide.md

File metadata and controls

131 lines (92 loc) · 5.08 KB

Setup guide

You've installed slopbuster. Now what? The skill works on-demand when you call /slopbuster, but you probably want your agent to follow the rules all the time — not just when you remember to invoke it.

Here's how to set that up across different agents.

Claude Code

Option 1: Add to your CLAUDE.md (recommended)

Drop this in your ~/.claude/CLAUDE.md (global) or your project's CLAUDE.md:

### De-AI-ify All Output

Apply `/slopbuster`'s patterns to ALL text you write — commit messages, PR descriptions,
comments, documentation, code comments, explanations in chat. Before finalizing any written
output, run it through slopbuster's rules:

**Text patterns to kill:**
- **AI cliches**: No "leverage", "harness", "dive deep", "delve", "tapestry", "landscape",
  "foster", "pivotal", "game-changer"
- **Hedging**: No "it's important to note", "it's worth mentioning", "arguably"
- **Corporate buzzwords**: "utilize" → "use", "facilitate" → "help"
- **Robotic structure**: No rhetorical questions followed by answers, no forced rule of three
- **Overused transitions**: No "Moreover", "Furthermore", "Additionally"
- **Significance inflation**: No "testament to", "vibrant", "groundbreaking", "nestled"
- **Copula avoidance**: Say "is" not "serves as"

**Code patterns to kill:**
- **Tautological comments**: Don't narrate obvious code
- **Verbose naming**: No Manager/Handler suffix abuse, no Enhanced/Advanced prefixes
- **Commit slop**: No vague verbs ("update", "improve"), no passive voice
- **Sycophantic docstrings**: No "This powerful function elegantly handles..."

**Add human rhythm**: Mix short and long sentences, contractions, active voice,
specific examples over vague references.

If something reads like an AI wrote it, rewrite it before sending.

This makes Claude internalize the rules for every response, not just when you call the skill.

Option 2: Add to AGENTS.md

If you use a shared AGENTS.md for team projects, same content works there. The difference: CLAUDE.md is personal, AGENTS.md is shared with the repo.

Codex CLI

Add to your ~/.codex/instructions.md or project-level AGENTS.md:

## Writing rules

Follow slopbuster patterns for all text output. Key rules:
- No AI vocabulary: delve, tapestry, landscape (abstract), foster, pivotal, vibrant
- No hedging stacks: "it's important to note", "arguably", "it's worth mentioning"
- No copula avoidance: use "is" not "serves as" or "stands as"
- No forced rule of three — use however many items the content warrants
- No filler transitions: Moreover, Furthermore, Additionally
- Code comments: no tautological comments, no "we" language, no section banners
- Commit messages: imperative mood, lowercase, specific verbs, no "various/several"
- Mix sentence lengths. Use contractions. Active voice. Specific examples.

Cursor

Add to your .cursorrules file in the project root:

When writing any text (comments, docs, commit messages, explanations):
- Never use: delve, tapestry, landscape (figurative), foster, pivotal, vibrant,
  leverage (verb), harness, utilize, facilitate
- Never use: "it's important to note", "Moreover", "Furthermore", "Additionally"
- Use "is" instead of "serves as" or "stands as"
- No rhetorical questions followed by answers
- No forced groups of three
- Mix short and long sentences
- Use contractions and active voice
- Commit messages: imperative, lowercase, specific

Windsurf / Cline / Other agents

Most agents read a project-level instruction file. The file name varies:

Agent Instruction file
Windsurf .windsurfrules
Cline .clinerules
Continue .continue/config.json (system prompt field)
Kiro .kiro/instructions.md
OpenCode AGENTS.md

Use the same content from the Cursor section above — the rules are agent-agnostic.

What to put where

Short version for any agent config:

Follow slopbuster rules: no AI vocabulary (delve/tapestry/landscape/foster/pivotal),
no hedging ("it's important to note"), no copula avoidance ("serves as" → "is"),
no forced rule of three, no filler transitions (Moreover/Furthermore/Additionally).
Code: no tautological comments, no verbose naming, imperative commit messages.
Mix sentence lengths, use contractions, active voice, specific examples.

That's 50 words and covers the highest-impact patterns. Paste it into whatever config file your agent reads.

Full version: Use the Claude Code Option 1 block above for agents that support longer instructions.

How it works together

  1. Agent config (CLAUDE.md, .cursorrules, etc.) — makes the agent follow the rules passively on everything it writes
  2. /slopbuster skill — active cleanup for existing files, with scoring and two-pass audit
  3. Rule files (rules/*.md) — the full reference the skill reads when invoked

The agent config is the 80/20 — catches most slop without you doing anything. The skill is for dedicated passes when you need scoring, the two-pass audit, or when you're cleaning up content that was written without the rules active.