essentials for Claude Code + Codex

Why I built it

Most AI code review setups are written for one stack. In a normal week I touch PHP, TypeScript, Swift, and Python. I wanted reviewers that figure out the codebase first and then review it the way that codebase expects.

How it works

/first-run profiles a repo once: language, framework, test runner, linters, and conventions. It writes a profile that every agent specializes to.

After that, /code-review looks at the current diff, picks the reviewers that matter for it, runs them in parallel, and merges the findings into one prioritized report. It only fixes code or leaves comments when you ask it to.

The reviewers:

• Security: injection, auth, CSRF, secrets, XSS
• Logic: correctness bugs that linters miss
• Performance: N+1 queries, unbounded reads, missing indexes

Migrations, routes, webhooks, dependencies, dead code, docs drift, copy, and UI each get their own agent too. A linter and a test writer round out the 13.

Multi-agent workflows chain them together. pre-pr-review runs the tests and then every relevant reviewer. codebase-health gives a full-repo baseline when you adopt a new codebase.

The Codex version

The Codex plugin ports the same ideas to 29 Codex skills. Claude commands and reviewer agents become focused skills, and the workflows become skills that orchestrate parallel Codex subagents.

Try it

/plugin marketplace add jshepelsky/claude-essential-plugin
/plugin install essentials@claude-essentials

Both plugins are MIT licensed. I use them across my own projects, including this site.