enrules

enrules

Claude Code plugin that makes Claude write like a human and work like a careful engineer.

The core problem with AI-assisted coding is not capability. It's behavior: editing before reading, broad rewrites for narrow tasks, padded explanations, generated-sounding prose, and overconfident claims about untested code. enrules addresses this with Markdown guidance loaded directly into context. No Python hooks, no external tooling, no post-processing.

The philosophy is in CORE.md.

What it does

When active, Claude:

• reads the relevant code and context before touching anything
• makes the smallest change that solves the task
• matches the repo's existing names, patterns, and style
• skips AI vocabulary, filler transitions, opener/closer boilerplate, non-keyboard characters (dashes, curly quotes, arrows, bullets, emojis, hidden Unicode), and hollow prose
• writes like a specific engineer made a specific decision, not like a language model composing a response

Skills

Four skills, each loaded independently:

Skill	Use for
enrules	general repo work: coding, reading, explaining
enrules-review	code review, diff review, PR review
enrules-debug	bug fixes, test failures, runtime errors
enrules-docs	README, comments, commit text, PR descriptions

Install

From the parent directory:

claude --plugin-dir ./enrules

Verify inside Claude Code:

/plugin
/agents

The default agent is enrules. It loads automatically.

Usage

Invoke a skill explicitly when the task is specialized:

/enrules:enrules-review

Or just work with the default agent active. Suggested prompts:

Fix this bug with the smallest safe change.

Review this diff for correctness, regressions, and missing checks.

Update the docs in the repo's existing style.

Explain what changed without filler.

Structure

enrules/
├── .claude-plugin/
│   ├── plugin.json
│   └── marketplace.json
├── agents/
│   └── enrules.md
├── skills/
│   ├── enrules/SKILL.md
│   ├── enrules-review/SKILL.md
│   ├── enrules-debug/SKILL.md
│   └── enrules-docs/SKILL.md
├── CORE.md
├── README.md
└── settings.json

Why no hooks

Markdown instructions are guidance, not hard enforcement. That's intentional. The plugin stays small, readable, portable, and cheap to load. If a project needs blocking behavior for risky commands, add a hooks/hooks.json layer on top.