Roast My Agents Md — Testing skill for Claude Code

roast-my-agents-md

Brutally honest AGENTS.md/CLAUDE.md review — backed by real A/B test evidence, not just opinions. Built on [Anthropic's skill-creator](https://github.com/anthropics/skills/tree/main/skills/skill-creator) eval framework.

Most AGENTS.md roasts are vibes. This one brings receipts.

How It Works

1. Static Roast — Quick audit with comedy. Finds sins, calculates token tax, delivers zingers.
2. Evidence Round — Actually runs your rules through A/B tests. Proves which ones the model already follows without being told.
3. The Verdict — Scoreboard with real data. Guilty (dead weight) vs. Acquitted (actually useful).

The Key Finding

In controlled testing, **25 of 26 assertions passed identically** with and without a 755-line instruction file. Most rules are redundant because the model discovers patterns from existing code. This skill proves exactly which of YOUR rules are dead weight.

Install

npx skills add vltansky/roast-my-agents-md

Or manually:

git clone https://github.com/vltansky/roast-my-agents-md.git ~/.claude/skills/roast-my-agents-md

Usage

• "roast my agents.md"
• "roast my CLAUDE.md"
• "prove my rules are useless"
• "audit my instructions with evidence"

License

MIT